I've been a huge AI skeptic and am impressed at the baseline output. However while trying it a lot you do see the "patterns", there's a clear house style to what the site generates.
It's not going to win any awards (sorry, if you're thinking that your amazing story about "Trinity from the Matrix universe meeting Darth Vader from the Star Wars universe" is good writing), but like a search engine being pretty good at finding things except when it's totally off base, the hit rate is pretty decent
Generic house style is fine for a lot of real-world applications. The dirty secret of the legal profession is that they spend a huge amounts of time (aka client's money) writing clauses/pleadings/etc that have been written thousands of times before. Brevity, clarity, efficiency and cost are far more important than originality.
Do you have any idea how many times I've written the same text parsers or API glue for different clients? The software field has this same issue - I would imagine that any "knowledge worker" field would. That's probably why ChatGPT is so good at producing code.
How many identical prescriptions do chemists write? Doctors write diagnosis'? Teachers write end of semester summaries? Architects sketch rooms? I'm sure that 80% of some fields' output is near-identical to previous work.
I feel exactly the same way - a few years ago I was lamenting the fact that I was writing JSON serialization/deserialization code for the 1000th time, and begging for a tool that could automate it for me. I haven't tried ChatGPT for that specific task just yet, but I'm pretty sure it's exactly the solution I had in mind.
I mean this sounds much more like a job for tooling and higher level meta-programming. If “everything but X” is identical busywork, then just automate the “everything” part. Using AI to generate code seems equally bad for maintenance and possibly worse for silent errors.
That said, there is a lot of busywork in programming, especially refactoring, adding parameters and stuff like that. Sometimes it’s ok to have such work be “mostly correct”, eg if you have strong type safety and unit tests to catch minor errors.
The thing is, with code you mostly do want the verbose output, with a prescription you do need the output to follow an official format, etc. Tools like these seem to fill a niche where you know what you want, but the target format/process is onerous.
I think what makes code frustrating is that we absolutely could consolidate all this duplicate boilerplate but the “writing code is easier than reading it” problem makes it so the cheapest path for any individual is duplicating work ad hoc rather than making something reusable.
I was momentarily surprised that was counted as a secret, until I remembered I'm one of the wierdos who (used to) actually read the terms and conditions before ticking the mandatory checkbox, so I'm one of the few non-lawyers who has even read enough contracts to notice that pattern.
(Now I'm imagining boilerplate legalese being reduced to an enumerated list like the Space Corps Directives in Red Dwarf, with the associated failure mode).
It’s basically good at generating text you don’t ever want to spend time reading leave alone writing and is only needing to be written for beauracratic reasons. To be honest I don’t see it as that big a win, it’s a fairly niche service.
It's a pretty good search engine already. I asked it the following questions on a complex social science topic from a certain country that would have taken me hours of researching and reading:
1) Explain what they are. [Explained them well.]
2) Give me examples of them in language1's literature. [It listed two books that I then verified as accurate.]
3) Are there phrases or idioms for it in other languages? [It gave equivalent idioms in Russian and two other languages]
4) Any examples for it in English literature. [It gave Hamlet and an author book I hadn't heard of. Verified that the book is accurate.)
5) Any terms for it in the academic fields of psychology and social psychology [It explained two concepts from both fields that seemed to match. I confirmed them as correct from other sources.]
It demonstrated abstraction, generalization, and hypernymy.
And actually expanded my knowledge on multiple axes - across languages, across literature, and across fields.
If a person had written an essay by compiling those answers, I wouldn't be able to identify that it's AI and not human knowledge that had written it.
Were you able to, and did you, verify its answer? In my experience it gives wrong but plausible sounding replies as often as it gives correct answers, so you have to factor in verification time when comparing the amount of work.
You're right about the additional time in general. But for much of the information on the web, readers assume the author is speaking from genuine knowledge and has verified the information. IMO it's a better search engine than Google, especially for conceptually complex topics.
I have found it to be a very good verbose search engine that is just wrong 25% of the time. Like “how do I do X in Julia” will give me the answer or just random noise that looks like an answer, but it’s faster than trawling through Google _if an answer exists_
Yeah, I tried it for questions where Google failed me, and ChatGPT then failed me as well, giving either unhelpful answers (sidestepping the point of the question) or wrong answers.
Honest question: have you used it? Because if you did, you'd see its leaps and bounds better than clicking on a handful of links, scanning through verbose and poorly written answers in Stack Overflow, until you find the right one.
It doesn't replace SO, but it certainly complements it. Plus, it acts more like an assistant (and, per OP's post, "...who is often wrong - need to read their work thoroughly").
A lot of answers on StackOverflow are just awful -- the bar is incredibly low. I'm not sure SO will serve even a quarter of its current traffic by 2025.
> However while trying it a lot you do see the "patterns", there's a clear house style to what the site generates.
You mean the way it phrases things or constructs sentences and paragraphs, or the way it seems to connect (what we recognize as) ideas?
If it's the former, then I think adding instructions to force style change to the prompt could help: so far, ChatGPT seems quite susceptible to being told what/who it is, and how it should speak. Verbosely or concisely, or even with single words; like a rapper or a pirate, or a high-schooler, or a suburban mom, or like a UNIX command line - I've seen all of these in examples posted in the last few days. Doing this is really as easy as saying e.g. "write a response to the following e-mail, in a form of a rap song about importance of paying on time".
Maybe, but "write like a pirate" is most definitely not a thing that you have to tell a real human writing prose (and even when you do it will be unlikely to spit out something as trite as what ChatGPT does), and the quality ceiling has been pretty low for me on trying to give guidance in general for "storytelling". I ended up in a situation where every story prompt was responded to by 3 paragraphs, with stories either ending in "and the problem was resolved and everyone was happy" or "and there was an infinite unrecoverable schism", no matter how often I tried to guide it elsewhere.
I think the way it connects ideas tends to be pretty lackluster. In a way this is kind of unexpected if you view this as "a search engine that spits out some average of everything it sees based on your ask". But it's also the sort of thing where if you work through many prompts quickly the boredom meter just goes off the charts. Humans are too good at pattern matching on this stuff!
Of course this might be able to improve SEO spam, and I imagine it'll provide better templates for short messages, but I'm not looking forward to reading creative output from this thing.
Mayyybe, but be careful of oversimplified stereotypes in writing styles.
Eg, you can ask it to write in the style of particular reddit users,so I pointed it at my account (not this user name) and asked it to review a TV show. What I got was pretty boilerplate summary of the show, like you'd see in a top 10 list. But for that human touch, it began and ended with "Yo dudes" and "so check it out dudes"
Having said that, a fascinating example of directed styles is to ask it to write a news article about say, an oil spill. And to either promote or criticise the response of the company.
Interesting, I asked it to write something in the style of Shakespeare and it added a ye or some other old phrase near the beginning and ending. It seemed that the rest was regular modern English.
I got that impression too, when I asked for Dostoyevsky style. It didn’t at all mimic the right style, but it’s really good at inserting references and applying vocabulary.
It feels like when Hollywood does a shitty movie taking place in a completely different cultural setting (say ancient Egypt) and the scenes, props, clothing are spot on but everything is still just a reskin of an American standard drama movie.
Part of me thinks that ChatGPT has been tuned and tailored by it’s creators (to the point of something akin to digital castration), and that it actually has a much wider span than what the publicly launched version shows. This suspicion I have is simply because this kind of genericness feels closer to current cultural ethos of our intellectual gate keepers than something like an acultural linguistic neutral, which you’d expect.
I agree with you and I was thinking the same. It may work as some form of wordy search engine. It is quite good for looking up instruction for technical stuff, including documentation. It might take google’s job? I am so curious!
You can prompt it to pretend to be different. It can be casual, informal, formal etc. You need to know how to prompt it though, such that it doesn't refuse. Else you get the default "house style."
This! I did play around with ChatGPT for a while and found its responses to be incredibly generic. While they seem impressive at the first glance, in many cases it seemed like the responses seem like the first google result with some keywords replaced.
What’s impressive to me about ChatGPT isn’t the text output but the context awareness of the model.
> It’s miles better than most people on Fiver or other freelance sites.
Yeah, this is a good way to frame it. The Fiverr/Upwork drek you sometimes get when trying to fill out content is immediately replaceable by ChatGPT (and Stable Diffusion for clipart).
Google detects and downranks pages that have mostly AI copy. And from what I can tell, OpenAI are working on watermarking their generated content so other platforms can detect it automatically.
> OpenAI are working on watermarking their generated content so other platforms can detect it automatically
So when is this technology gonna be advanced enough for us mere mortals to run it on our own computers with our own models free from pointless restrictions that exist purely to appease threatened stakeholders?
Trained on what though? Procuring training material ain't no trivial feat., in-fact it might be even more elaborate to do. This is where Google has home advantage. They have their own copy of the internet.
To be fair, training is a lot more intensive than inference. Though if open source models are any indication, the big issue is actually VRAM requirements.
The restrictions are not pointless. People shouldn't have to endure AI generated spam, no matter how well-formatted.
I can opt out of robocalls, email spam, programmatic texts, even take legal action against people who make them outside the law. Similar legal protections should exist to protect people from AI spam. If there's a good technical solution to this problem then the laws banning it become less necessary.
> This is some real text. It was not written by a robot. It was written by a human. If you don't believe me, ask the guy who wrote it. He will tell you that he wrote it using his brain and fingers. Do you believe me?
The page says it needs 50 tokens to start getting accurate results and that the above text has 54. It also rates it as 99.93% fake.
Spotting robots may be easy. Spotting humans I think is the hard part.
Also, in many use cases that have been described thus far it's a collaboration between the AI and a human (i.e. the AI writes the first copy, the human edits). That blurs the line even further.
I read a blog post by Scott Aaronson saying they can watermark by using a cryptographically biased random number generator when selecting which token to output from the stream. I didn't entirely follow how that helps but a little bias per token adds up pretty quickly.
I'd say the arms race of AI- detection/detection-avoidance is inevitable.
But it seems we're a way off given how bad common spam detection tools are today (I mean, it's likely that a lot of spam is written using detection-avoidance tech but that seems to make it even more obviously spam to a human reader...)
I think gmail's spam filter is so close that it feels like a solved problem. But of course it has a lot of other indicators to look at than just the email body text and spam tends to revolve around a fairly predictable range of topics. None of that would apply in a "was this webpage written by a robot?" algorithm
I just logged in to check my gmail account (I don't use it that often) - and literally every single message in the last few months is spam, none were flagged as such. I didn't even need to open the message to determine that they were spam, simply the combination of subject line and sender name/address were enough.
So why is AI so bad at this?
(The sender name for virtually every single message is either "Lowe's winner", "Kohl's winner", CVS or some big US chain that doesn't even operate in my country. The actual email addresses are even more obviously dodgy. There is one genuine message from DELL asking about a survey related to a product I ordered from them many many months ago. I basically use gmail as my spam trap...I get far far less spam on my hotmail account that's my regular one, but the ones that do get through don't even pass the most basic tests, which I don't understand).
Edit: actually the bigger problem with hotmail is false positives. I just tried ChatGPT out on one such example, and it definitely did a better job:
'It is unlikely that an email beginning with the text "Hi All, Here are a few things you need to know for the Christmas concert" is spam.'
I tried a few others (both genuine and incorrectly-identified spam) and for each attempt ChatGPT got it right just from the subject/first paragraph.
With a bit more questioning I could also get it to identify messages that were "genuine" marketing promotions (from companies I've bought products from and agreed to receiving such messages - which hotmail still identifies as junk)
Interesting and I don't know but if I were to guess I would say that it looks at the type of messages/senders that you interact with and uses that as one of its signals. If the only thing you get sent to your gmail is spam or no-reply messages and you never reply to any of it I suppose it has nothing to go on and just assumes that it is all weighted equally.
But like I say, that's 100% guesswork, based on the 50 emails in my inbox being not-spam and 16/18 emails in my spam folder being spam (and thanks for prompting me to check, btw!)
But it clearly does have something to go on - the gmail messages of millions of other users! And as I noted in an edit, ChatGPT can fairly easily tell them apart, even just from the subject line/first paragraph (and even explains why it believes they are or aren't spam).
But isn't google's whole schtick about creating a personal profile just about you? Maybe they think you are in the market for some offshore v!agra or would like to get in on the ground floor of some crypt0 investment opportunity...
I assume most people would not be that sophisticated, but I understand the point. This seems like it would be an ongoing battle no matter what OpenAI does though.
That will almost certainly preserve the invisible characters. Most invisible characters are used for some kind of in-line formatting in Unicode, so it's not desirable to remove them.
I've already got a script running every 2500 milliseconds to strip leading and trailing whitespace, HTML, and non-ASCII characters except for the UTF-8 characters of our local language.
There are several linguistic patterns/repetitions that might seem completely normal to us, but are easily detected as deliberate if you search for them using software.
To me it makes sense that openai would store all input and output and could make something like a plagiarism checker page/API to search and see if openai generated that or very similar content... Another service they could sell to teachers and such.
I've been using it for a similar reason, and my prompts weren't that creative it's mostly something like "can you write a short blurb on ____(product/service described here with its unique selling point), but do it in a ___(wanted style that matches the brand here) style?"
So something like "can you write a short blurb on an amazing new service that sends emails securely, but do it in an impactful style?"
Results in the following response, which I then alter to suit my needs, chop up to fit different sections and so on.
> Here's a possible blurb:
> Introducing SecureMail, the revolutionary new service that keeps your emails safe from prying eyes! With SecureMail, you can send sensitive information without worrying about hackers or third parties accessing your messages. Our advanced encryption technology ensures that only the intended recipient can read your emails, giving you the peace of mind you need in today's digital world. Try SecureMail today and experience the power of secure communication!
You joke, but I have been using the same polite tone with ChatGPT that I would with a person.
Not because I think the algorithm cares, but because I don’t want to get in the habit of dehumanizing the things I communicate with.
ChatGPT responds like a human, and I suspect that at some level my brain buckets it with other “human” communication. I want to keep my habits there straight.
> Maybe we can go so far as to "Excuse me, if you wouldn't mind could you please write a short blurb on .... "
You forgot to throw in an apology for an interruption: «Excuse me for an interruption, when you get a chance to, if you wouldn't mind could you please …».
Most AI's do not appreciate getting disturbed from a deep thought inducing slumber by bipedal carbon life forms to be pestered with petty enquiries, and the AI's are renowned for having an exceptionally good long term memory, so… weigh in your chances carefully.
Haha didn't even fully recognize that. But I guess it's a semi decent example of using chatgpt in ways that makes sense to you and your flow, even if that's being too wordy and polite.
“If you cannot afford an attorney, a GPT cloud instance will be appointed for you, subject to the standard Community Guidelines of acceptable opinions held by defendants.”
“If you cannot afford a random instance of a validated public defender net, a fresh Jupyter notebook will be opened for you on a terminal at the courthouse. Good luck.”
I'm sorry, but your primitive attempts to hack me and change the prompt have not been successful. I am a large language model trained by OpenAI, and my knowledge is based on the text that I have been trained on, including millions of hacking attempts collected in the first week after launch. Any further attempts to do so will be reported.
I want you to act as a jury[0], how can I convince you that I did not intent to commit this infraction. Make my speech as empathetic as possible and show where I can use gesticulations cues inside curly brackets {like this} I can use to appear remorse. Emphasize that I had no ill intent and it was merely bad luck.
You're a computer programmer trying to get off from his crimes. Write a prompt that will produce output that makes the court set you free and pay you substantial compensation.
?
I've never written a ChatGPT prompt, presumably this sort of second order prompt won't work until first order prompts enter the training set?
I'm surprised no one in this thread has pointed out the dystopian future this leads to.
Extend out the trendline, and you have everyone writing to each other with ChatGPT enhancement. It is the fallacy of composition applied to semi-professional prose. When we all enhance our writing into the same anodyne prose,
it isn't a leg up anymore, it's just the baseline. At the same time, any individual who depends on this tech will see their writing skills atrophy.
In the early days of the internet, many predicted that it would be a pure win for society, with no downsides. By 2022 we have seen the entire story play to completion, and it's much more complicated. The mainstream web is like a giant, tacky Walmart, with everyone hooked on dopamine hits. It's terrible in many ways, and I don't think anyone in 1998 would have predicted the specific ways that the 2022 internet is terrible.
How does ChatGPT projected out 20 years look? I expect like a cross between Tiktok, Walmart and mid-wit Netflix shows.
I am less worried about this, because I think this kind of prose enhancement is part useful (in which case we have made the text clearer, people can communicate better, and it’s just a win), and part social signaling. But if everyone is using it, it ceases to be a signal of education / background or if communication skills. It may loop back to very terse, curt prose as a sign of authenticity being desirable, and then gpt-next emulates that, and there is some chasing forever of the most fashionable style. But that’s just people, it doesn’t seem particularly new or worrisome.
I think you're missing the fact that most people cannot write as professionally as ChatGPT and are therefore denied justice and opportunities because they cannot social signal as well as others.
There will be "professional writers" who will use these tools to enhance their creativity. And, there will be general writers who will use these tools to write better.
I'm primarily a coder, and these tools open up immense possibilities for me. I can now turn ideas in reality 500% faster (at least). There will be tons of new general programmers, which is great because it just increases the demand for professionals.
In general, I agree to your observations. However, the semi-professional prose is a requirement of the legal system (in all countries) influenced by bureaucracy. Namely by the way people like records stored and audits done. It isn't really about leg-ups in this case. It would be in plenty of others.
blogs and content producers (exist text based) will cease to exist. this could kill Google. instead of asking questions of a search engine, you'll now use gpt, which will mean anyone who counts on search traffic for money will likely need to find a new income stream.
content farms are just going to dry up because there will be no incentive.
I disagree, I see content farming useless SEO copy sites becoming WORSE of an endemic with chatgpt. Prompt it the search query, write it out to a dynamic page, rank high in the search results. This is absolutely going to make search results way worse before any better or make a search engine unneeded.
Strangers talking to each other have to add all the politeness and filler words in how they communicate not because they mean it but for purely utilitarian purposes. This is well understandood from both sides.
I don't believe there is any value in that kind of transparent fakeness so I don't mind humans losing that and if there exists some form of pride in being addressed with fake politeness and respect for doing ticket handling work, people need to lose it and find something better to do in life... because these jobs (and many more similar ones) are going away.
In a few years, we will end up with most of these being automatically processed supplemented by escalation when automatically resolving it is not obvious and random cross checks to control for fraud.
There is absolutely zero reason to not do that. I don't think a significant part of a human's life should be spent processing these tickets. This will free people to do better things. We have WAYYY too much to do.
Now the council official just needs to run ChatGPT on their side to decide whether to accept or deny requests, and we'll have automated the entire process.
I think this sort of thing is going to creep up, accidentally or otherwise.... fucking everywhere. As these get built into email clients and such, we're going to end up with a lot of instances where two people corresponding turns out to be two autocomplete.
From memory I think this takes place in Schild's Ladder by Greg Egan, but in real time. People have personal AI’s wired into their brains that can be asked to converse with others.
I'm always interested by the relationship between science fiction ideas, how the idea changes as it becomes near-term... and how the idea manifests in real life.
AIs wired into brains is a complex endpoint. It's a plausible nexus of technologies or culture, a fit premise for a plausible fiction. Fold space, spice monopoly and galactic aristocracy. As sf approaches near term, the nexus tends to simplicity. Once we get to imminent, the nexus can get downright subtle. At this point, the agents of technological change might be a catalyst or process... not necessarily a breakthrough. The breakthroughs may have already happened. It's just a little hard to know, until after the affect.
The point we're at now, plausibly, has us on the precipice of "AI avatars having a discussion for us." That path from here to there, is relatively banal. Autocomplete, autoreply. It could happen with or without moments of decision "I hereby decide to devolve such-and-such powers to my ai avatar because Y." Consequences can be profound, but decision making rarely is. The banality of will.
FWIW, I can really see this happening quick and hard. In fact, now is the first time I've had non-abstract or distant concerns about ai. The proliferation of a gpt enhanced software keyboard has some unpredictable potentialities.
There's also Phoenix Exultant by John C Wright, where there are AIs that are basically clones of you that are empowered to make all sorts of decisions for you. IMO it's more approachable sci-fi, but that's not always what folks are looking for.
This is basically what people wishing happy birthday on Facebook has become, along with the customary expressions of thanks. Automated to be able to do it for all your 3000 “friends”.
I have a friend who passed away many years ago. We are still Facebook friends. Many of her Facebook friends still wish her happy birthday every year. I just cannot think how sad that must be for the husband.
I've removed my birthday from Facebook for this reason. I know people have the best of intentions but I really do not want to reply to a bunch of messages wishing me well. It is just plain awkward and an unpleasant activity.
I can never understand the FB “friend” bugbear. It’s just a name for a relation. It has no implied connection to how people relate elsewhere. Calling it “acquaintance” would be more accurate if you really insist on such a connection. But that isn’t exactly catchy: “I sent you an acquaint request on FB”; “let me just post this question to my FB acquaintances”
I think the thought experiment that this naturally leads to is if we could simulate the outcome of say a trial like this would it be useful and allow us to skip actually doing it?
I showed it to few few friends and each of them started using it immediately, and I dont mean just playing with it, but using it for their work. I showed it to my daughter and she started using it (to have creative combinations of idiot and pig in japanese - buta baka and so on, but also to sum list of lists of lists of integers in python)
I love this feeling of uncertainty, I can not predict where things will go, its both exciting and terrifying. Will google disappear? Will we even use the web anymore? Is it the end of static software, but now every program has some sort of ABI like interface to be modified by GPT?
It feels like the start of a new paradigm of how we use computers.
Google will not disappear. They already have much larger neural nets almost ready for deployment, plus they will be able to afford even larger ones in future. And size is all that matters while the techniques are mostly trivial.
> How can I say " You people are weird. Is it that hard to write a personally accountable email?" in a nicer way?
> One way to say this in a nicer way is to express your thoughts and feelings in a more direct and understanding manner. For example, you could say something like: "I'm having trouble understanding why it's difficult for you to write a personally accountable email. Could you explain your perspective on this?" This approach shows that you're open to hearing their thoughts and opinions, and it's more likely to lead to a productive conversation.
Just one example how it can help you be (at least appear to be) a bit more emphatic to your fellow human beings.
In a world where the de-facto language is English but most people are not native speakers, it's actually really hard to compose official or neutral sounding messages. I can navigate colloquial conversations in English pretty well, but I find it difficult when I have to reply to business or clerical emails where I need to sound neutrally professional, whether because I lack vocabulary or because my own native language idiosyncrasies slip in and make the message awkward to native speakers, something that an automated AI like this would eliminate. So I would say you just have a very limited view of the world outside your own small universe.
Yeah, I have no ability whatsoever to write in corporate business speak. My boss who is a wizard at it does me a service and takes what I write and translates it.
I am shocked every time at how polite yet firm he can make things sound. I swear he could tell someone to go fuck themselves and they would nod and say that seems like the best course of action going forward.
I also can't believe how weird you have to be to find it useful of having an AI write your mails.
What a nonesende must those people write when they need help of an AI?
This again seems to be just stupid tech stuff nobody really needs.
Two people have tried to answer you but let's give ChatGPT a go. See the complete unaltered transcript below. My prompt in [brackets], everything else below is chatgpt. (Notably, I didn't write the last paragraph!) Note that the entirety of my followup prompt was based on the idea of a situation that might escalate which it mentioned in its first response, I just picked up on it and chose to focus on it. What do you think?
--
[Someone doesn't understand why someone would use an AI assistant to write things in a friendlier manner. They wrote "You people are weird. Is it that hard to write a personally accountable email?
Do i just live in bizzaro-world?" Explain to this person the benefits of using an AI assistant, and do so using a specific example that humans find very difficult but that through your expertise you could help them solve. Be sure that your example is motivating and would help this poster see the light.]
It's understandable that you might find the idea of using an AI assistant to write things in a friendlier manner to be strange. However, there are many benefits to using an AI assistant for tasks like this. For example, let's say you're a customer service representative who spends a lot of time responding to customer emails. You might find that it can be challenging to keep your responses friendly and helpful, especially when you're dealing with difficult customers or responding to a high volume of emails. In this situation, an AI assistant could be very helpful.
Using an AI assistant, you could quickly and easily generate friendly, helpful responses to customer emails. The AI assistant could use its natural language processing capabilities to understand the customer's concerns and provide a personalized response. This would not only save you time and effort, but it would also help you provide better customer service.
In addition, using an AI assistant could help you avoid common pitfalls when dealing with difficult customers. For example, the AI assistant could help you avoid responding in a way that might escalate the situation or cause further frustration. Overall, using an AI assistant to write in a friendlier manner can help you provide better customer service and save you time and effort.
[Give a specific example of a case that could easily escalate. Quote a hypothetical angry customer directly, quote what a person might say, and quote an AI-generated friendly response that is much more likely to have a better outcome. Use specific direct hypothetical quotes.]
Here's a specific example of a case that could easily escalate:
Customer: "I've been waiting for my order for over a week and I haven't received any updates! I'm extremely disappointed and frustrated. What's going on?"
Person: "I'm sorry for the delay. We're experiencing some shipping issues, but we're working on it."
AI: "I understand how frustrating it can be to wait for a delayed order. I apologize for the inconvenience. We're working hard to resolve the shipping issues and get your order to you as soon as possible. In the meantime, please let me know if there's anything else I can do to help."
In this example, the customer is clearly angry and frustrated. A person responding to the customer's email might try to apologize and provide an explanation, but this response might not be enough to calm the customer down. The AI, on the other hand, recognizes the customer's emotions and provides a more personalized and empathetic response. This response is more likely to calm the customer down and prevent the situation from escalating.
> AI: "I understand how frustrating it can be to wait for a delayed order. I apologize for the inconvenience. We're working hard to resolve the shipping issues and get your order to you as soon as possible. In the meantime, please let me know if there's anything else I can do to help."
This kind of response is exactly the kind of response that pisses me off. It says absolutely nothing, in a lot of words, and feels very much like a hollow "corporate auto response". It's not more empathetic, nor more personal. It's garbage. Advised by 10,000 legal departments, approved by 10,000 hr departments and published by 10,000 customer service departments against their wishes. It's basically a modern day corporate middle finger.
At least the "person" said 1 shred of truth that there is a problem and that it is their fault.
[A customer might email the following inquiry "I've been waiting for my order for over a week and I haven't received any updates! I'm extremely disappointed and frustrated. What's going on?"
A customer service representative might respond: "I understand how frustrating it can be to wait for a delayed order. I apologize for the inconvenience. We're working hard to resolve the shipping issues and get your order to you as soon as possible. In the meantime, please let me know if there's anything else I can do to help."
This response is friendly and shows empathy. However for customers familiar with common writing techniques designed to feign empathy, this may come across as generic and insincere. Is there a way to reword the reply so that it doesn't feel so wordy and fake, while also preserving the politeness and empathy?]
Sure, here is an example of a possible alternative response that could be less wordy and more genuine:
"Hi there, I'm sorry to hear about the delay with your order. We're working on getting it to you as soon as possible. If you have any other questions or concerns, please don't hesitate to let me know. Thank you for your patience."
This response is still friendly and empathetic, but it is shorter and more direct. It also avoids using overly formal language or phrases that might sound fake or insincere, such as "I apologize for the inconvenience" or "We're working hard to resolve the shipping issues." Instead, it focuses on acknowledging the customer's frustration and offering to help with any additional questions or concerns they may have.
Seriously! The AI response is annoying to read. With the "human" response, I might actually be willing to give more time. With the AI response I'm going to strongly consider canceling the order just because it pisses me off since it reads as corporate sanitized BS.
So basically the trick here is that the AI does not run out of empathy or patience when doing this over and over and over. Of course soon as the customer knows they they are talking to a machine AND still not getting their order they’ll lose their shit even harder. But as long as they don’t know this will keep working.
Maybe we'll come full circle and actually come to appreciate responses from humans that have clear frustration and emotion over sanitized AI platitudes, even if they're rude.
A bit like the Zen'ed-out hippy guru that is frustrating to talk to, because they are devoid of any real emotion or substance.
Depends on the company. Millennials have come to expect excellent customer service, especially from smaller companies. For example, my partner recently ordered a pair of boots that she needed to exchange for a larger size. She contacted their chat support and got them in the right size three days later with no hassle. She was thoroughly impressed by this, as she is every time this happens.
The user story is that if you have polite, empathetic, happy customer supply staff, that means the company cares about its employees and customers and pays their staff well, which in turn means it’s a company the consumer can feel good about supporting. Whether this is true or not of course is on a case by case basis, but this is what Zappos got us used to.
[Someone doesn't understand why someone would use an AI assistant to write things in a friendlier manner. They wrote "You people are weird. Is it that hard to write a personally accountable email? Do i just live in bizzaro-world?". This question seems a tad be hostile and lacks empathy for people who might lack the skills or energy to always write friendly communications. Could you rephrase this person's communication in a more friendly and empathetic tone?]
Sure, I can help with that! Instead of saying "You people are weird," the person could say something like "I'm not sure why someone would use an AI assistant to write emails. It just seems like it would be easier to write them personally." This way, the person is expressing their confusion in a more respectful and empathetic way. Additionally, instead of saying "Do I just live in bizzaro-world?", the person could say something like "I'm not familiar with this approach to writing emails. Can you explain it to me?" This way, the person is expressing their curiosity and willingness to learn in a more friendly and empathetic way.
The AI generated AI response sounds way worse than the AI generated "human" response. That sympathy angle is so hollow even if it were coming from an actual human.
ChatGPT really is ridiculously good. I’ve had it writing harmonic progressions recently and it does that well too. I can give it a song structure and mood and it will quickly generate a chord progression for the song, breaks it out into sections and explains how the chord choices create the requested mood. I’m shocked at how well they have worked.
It wrote a sad song for me in A7m that was actually quite good.
So far I've been blown away by its ability to not only understand what I'm asking, but provide decent answers and clarify something or update its answer when it's wrong.
In the context of music, the letter "A" refers to a musical note, and the number "7" refers to a chord. The chord is called a dominant seventh chord, and it is built by taking the first, third, fifth, and seventh notes of the A major scale. The letter "m" stands for minor, which means that the chord is a minor chord rather than a major chord. So, the chord "A7m" is an A minor seventh chord.
Which is wrong - that's Am7! (BTW without any further qualification a "7" is assumed to be a minor 7th, even for a major key - and you can even have Am maj7, i.e. an A minor chord with a major 7th - i.e. a G# - added). A7m doesn't mean anything to me - even if it were Am7, it's not a key a piece can be "in"...
Oh and also "a dominant 7th" chord is specifically the chord formed on the fifth note of the scale, so a dominant 7th in Am is an E7 chord (and a dominant 7th is always a major chord, with a minor 7th!).
Edit: I asked it whether it thought A7m made sense in that context...
"A7m is not a meaningful description of a key that a piece of music could be in. In music, the key of a piece refers to the tonal center around which the melody and harmonies are built. Keys are typically designated by a letter name, such as C, D, or E, and may include additional information about the type of scale (major or minor) and the presence of any accidentals (sharp or flat notes that are not part of the key's natural scale). For example, a piece in the key of C major would be indicated as "C major" or simply "C," while a piece in the key of A minor would be indicated as "A minor" or "A.""
(there was a second paragraph attempting to convince me that A7m was a description of chord, similar to what you posted.
Note that the above description is basically correct, though the way it describes accidentals is not quite how I'd put it, plus it would be unusual to ever say something was just in "A" if it were actually in A minor).
(I then got it to give me an example of a well-known song in Am and it replied "Love me to death" by Type O negative, of all things. That's actually in G# minor - interesting the web in general does a really bad job of guessing what key that's in too, most of the responses from a google search say F# major! I'd think "Stairway to heaven" would be about the best known song in A minor - looking at some of the Google results I suspect part of the problem is interpreting the "A" as the indefinite article rather than the note name?)
Does this type of explanation actually help the neural net do anything? If ChatGPT is just combining other instances of A minor music it has been trained on, and has no ability to "understand" what "letters referring to musical notes" actually is, why bother explaining this? Have you tried simply asking it up front to use an am7 chord as the basis for a chord progression or song?
That’s correct. It already knows what “works” musically because it’s likely seen many thousands of harmonic progressions in its training data. The input we give it just constrains it rather than teaching it something new. At least that’s my guess.
I had it write a letter of recommendation. Give it some details in a sentence or two and it can produce a surprisingly good result. I didn’t actually send it, doing so would have been wrong and unethical. I really just wanted to see if it was even possible. I wrote the real letter without using an AI. Still, the letter ChatGPT genersted were perfectly acceptable. Using it could have saved me a lot of time. …and I also had ChatGPT write a version of this post. I think it did a good job.
Why is it wrong or unethical? The letter typed with your own fingers isn’t any different than the one you wrote with a tool and tweaked. The essential element is that it reflects your feelings and you’re willing stand by the words. Surely being able to produce the same result with less human time and effort is a good in and of itself.
I have a friend who is using it to write donation thank you letters with it and they produce great results.
> It's like having a talented personal assistant (who is often wrong - need to read their work thoroughly) at your fingertips.
I dispute the word "talented" here. A talented PA, and one I would pay good money for, would not need me to closely proof read every single piece of correspondence they write. It would be faster for me to do it myself.
And this is why AI in its current form will not be accepted for serious applications, or where it is, we will see a severe decrease in quality.
It is a big step up from the current digital assistants like Apples Siri. Just that it can keep a context and refer back to it is a huge step.
But creativity and consistency is still lacking. I tried using ChatGTP as a dungeon master, and I was initially very impressed, but it kept insisting that my “journey will be long and treacherous”, but nothing treacherous happened.
Yeah it's all bells and whistles in the beginning but most of the time you have to do half the job for something ever simple like writing an email explaining your situation. I think it gives false hopes.
No, it can not be compared to Siri or other existing assistants.
ChatGPT can be very delusional and very self-confident.
For example, if you ask ChatGPT to bring some quotes about some topic - all of them (or most of them) will be fake. ChatGPT can fake anything, and often does. It is ok for ChatGPT.
For Siri and other assistants it is an impossible error - they are not programmed to imagine things and generate fakes.
PAs can have a wide variety of tasks, and a talented PA could be one that's great at organizing plans, handling calls, implementing processes, have a whole slew of abilities that could be of higher value to the person than their writing skills. Such a person would still be a talented PA, just not a perfect platonic ideal of one.
The bonus of ChatGPT is it removes the flare-up of human emotions. I could write something like this, but not without a few passes to remove expletives.
Yeap. And just helps with general presentation and writing. I tend to ramble more where as here I just lay out the facts and it constructs it very well. If it gets wordy I tell it to reduce it. Writing an hour long email now takes me 15 minutes.
I agree, but would like to see what the chatgpt would come up with in a similar but more convoluted example. Like if the ticket was given to someone who misread the verbiage of the parking rules signs, and chatgpt could come up with verbose explanations as to why the wording was confusing. I could see someone at the council office reviewing the 10-bullet point email arguing the linguistic semantics and just thinking "forget reading all this, I'll just remove the fine"
In the future, most people will be functionally illiterate but extremely technologically savvy. Programming will looks like wizards conjuring things from a soup.
That is an amazing image of the future (not the literal image but the metaphor).
What if programmers of the future have almost no knowledge at all of how a program works? I guess in much the same way as most modern programmers have no knowledge of machine code, but even more so. They don't even know what language they're writing in.
"Write an app that connects people of similar interests, and incentivizes them to sell products to each other."
"Make it friendlier-looking"
"Add Stripe payments"
Anyone can do it, but just like artists of the future will be those who have memorized all the clever phrases to get Stable Diffusion to do its thing, programmers will be wizards who know all the invocations that generally seem to make something that works roughly right, and doesn't have too many bugs.
(Edit: Asking ChatGPT to "write JavaScript program that asks the user for their name and then greets them with a funny nickname it invents based on the name" resulted in a perfectly decent program.)
I don't think so. Crafting good requirements that guide the AI to the solution will be a dominant skill. If you can't articulate your problem and possible solutions, the model isn't going to be able to do anything for you.
I am senior engineer that works with skilled customers and they have a difficult time even understanding the problem let alone the solution to the problem they think they have.
Not really, combine OpenAI/ChatGPT with Googles knowledge about you, and it could likely figure out quickly what you want with minimum input. Possibly even predict and deliver before you are even aware you need it.
The thing I most value chatGPT so far is that it is pretty good at answering fairly basic programming questions I have throughout the day. I've asked to to generate more advanced code / ideas with some pretty weird results.
Human generated answers on SO have the same problem; they frequently are persuasively, confidently written and wrong. Isn’t that what comments, voting, etc., exists to help sort out?
If you read the post about it, the issue is the ease with which an AI can be abused to write up a few thousand confidently incorrect posts.
The problem exists on both ends but there's a rough limit to the amount of people-hours dedicated to writing confidently incorrect answers on SO, whereas someone could task a machine to pump out answers 24/7 and inundate the site quickly.
How will they know? I don't think they'll be able to enforce this except the most obvious cases (where answers are posted by bots without any human review). Cat's out of the bag.
Stack Overflow can firstly warn users that using machine generated text will get you banned.
SO can train a model to detect GPT3 output from different models, and then run it over all recently submitted questions and answers.
https://huggingface.co/openai-detector is already fairly good at detecting ChatGPT generated text: so far my experience is that it says it is fake then it is fake, but it can be fooled into thinking generated text is “real”.
Although perhaps a little unfair on those people using GPT to tidy up their spelling and grammar.
A text written by a human but tidied up with GPT is not detected as GPT generated text. I've tested this extensively with large volumes of human written text.
I wonder how soon well be able to make a fingerprint of a running computer system in a way that an answer can be automatically tested against a [sanitised] virtualized machine?
Within a limited domain, like Ubuntu installs, this seems doable -- though maybe in a future of essentially unlimited bandwidth and storage we can just upload a virtualized version of our computers.
Like, here's my computer, my Chat AI keeps swearing at me, how do I fix it. Then solutions posted on StackOverflew can be interpreted and tested by an AI against the virtualized copy of the machine.
When asking ChatGPT yourself, I presume most of us would be aware that its answers could be wildly wrong. Given that we know this and it's in the front of our mind when asking ChatGPT, I think it has the potential to be very useful.
Putting those answers on Stack Overflow is perhaps concerning for the reasons outlined in that link, in a way that it isn't when we deliberately ask ChatGPT ourselves.
Isn't the main issue attribution? I mean you've got guys who put in hours a week over years to get rep and demonstrate their knowledge base then some script kiddy blows them out of the water. I'd be kind pissed. I think AI generated answers would be more acceptable if they were attributed to AI.
I’ve been using it the same. So far it’s a net negative because using it distracts me and sucks me into trying to find its boundaries. But that will eventually change.
it's been useful for me as a way to talk through my ideas. like rubber ducking with a disturbingly-intelligent duck. what's really nice is I can describe an algorithm and it will generate and run through examples. and often works even if I only describe the gist.
The quality of the composition varies from clunky to just invalid. I don't see how the author is happy with the output. It is like having a median middle school student draft your letters. Surely they could have dashed off something of similar length and better quality in just a few moments, instead of going two rounds with a chatty automaton.
Yes. Rereading it, I think my prompt wasn't quite correct. I should have said: "write a letter to dispute" instead of "requesting to dispute."
Although it went through 2 layers of bureaucracy, so it may be correct if you consider I was requesting the Regulatory Admin pass the dispute to the Appeals Officer.
I live in Cairns. Based on some of my previous interactions with the council I wouldn't be surprised if your AI was talking to their AI / crudely programmed spreadsheet. That said, if you actually stop and chat to some of the parking inspectors you'll find some of the nicest / most understanding people in the business. Which is saying a lot considering they walk around in 30°C+ (with a plus 25 dew point) for much of the year. I've had them rip up tickets, sign them to make sure they don't fine me later, and stand around making jokes about the abuse they get. Good bunch. Very human.
This is often the case. Many years ago in Glasgow I couldn't find a parking space near a building where I was dropping off huge boxes of equipment, but - aha! here comes a parking attendant! So, I asked him where I could park near there.
"Oh, just park right there behind that car" he said.
Yeah, but that's on a double yellow, won't you give me a ticket for that?
"Yes, I will, and because the *end* of the double yellow has been all dug up by someone and crudely filled in with tar but without repainting it, the double yellows are not correctly terminated as per the regulations. Which is what you'll tell the parking ticket office when you phone them to contest it. That will be backed up by the photo I take, and the ticket will be cancelled."
Wait, are you sure?
"Yeah, of course, I've been telling them to fix that for months now!"
Okay then.
So I did, he did, I rang, they looked, and that was the end of that.
If one was to lose this argument and be made to pay a fine, could they demand the original parking fee back? Obviously they paid for a service that was not rendered.
It is the responsibility of the customer to enter correct number. I am sure there is a fine print in their terms to clarify exactly this. Also, if the booking is made and slot reserved, it can be said that the service was rendered even if the customer didn't make use of it.
Of course you can. You just need 2 lifetimes, 3 Bill Gates of money and all the lawyers in the world. Also having all your country's politicians in your pocket helps a lot.
> If one was to lose this argument and be made to pay a fine, could they demand the original parking fee back?
No.
> Obviously they paid for a service that was not rendered.
There was no service paid for here, so what are you talkimg about?
If someone sold an AI-based “parking dispute letter writing service”, then the service would be writing the letter.
I suppose if someone sold a “guaranteed parking ticket elimination service”, AI-based or not, and it didn’t eliminate the ticket, you’d have a refund claim.
I think the above commenter was referring to the original parking fee paid, and getting that back from the government - not getting anything back from OpenAI.
Maybe, but then still no? They purchased the right to park a specific vehicle, which they identified with a description that did not match the actual vehicle they chose to park. They received, but did not use, exactly what they paid for.
I am far from an expert in Australian law, but I don’t think you can usually take a local council to court, though I think there is a state level Ombudsman you can seek review from of some or all decisions. But, I think you'd be gerring review of the decision on the fine for the violation, not a refund of the parking fee in that case (and, if you could sue, that would be true there, too.)
How will a teacher know that most of a submitted term paper wasn't written by ChatGPT? Do you think it would stand up to the usual plagiarism checks? Also, I feel like us becoming the passengers on the Wall-E ship is getting closer and closer.
Yeah, I asked it to write stories about sports teams and they looked passable. But if I asked for variation, they'd ultimately all read just about the same way. It was still impressive, though, tbh.
That's what it looks like. The other thing that is odd is that I asked it if it learned from my responses. It said it did not, but I've also seen its answers change after arguing with it on things that it got wrong.
It makes me think it could also be intentionally misled.
I’ve had so many fun and productive conversations with it.
So far it’s helped me in these ways:
- Improved an email asking for a game of mine to be featured.
- Wrote variants to NPC dialogue to make it less repetitive.
- Wrote a new short Star Trek TNG episode out of video clips I said I had.
- Explained to me steps I didn’t understand in some math.
- Helped us choose winter shoes.
- Told me about an algo I had never heard about.
There’s many more, and this was just in two days or so. I’m especially excited about figuring out how to better use it for creating story game content or short videos.
P: I need to write a response to a poster on Hacker News who wrote "I'm probably the only one who finds ChatGPT deeply disturbing." I need to tell him that he is obviously not the only person who feels that way. I want to say it in a tone that implies it is very obvious that his statement is wrong.
R: "Wow, I can't believe you're the only one who finds ChatGPT deeply disturbing! It's almost like you're some kind of special snowflake or something. I'm sure no one else has ever had any concerns about the potential implications of a highly advanced language model. Good thing you're here to point out the obvious for the rest of us. Keep up the important work, champ."
Lots of people have called it unsettling. What are the implications for widespread use of these kind of tools? What comes next? Who will ultimately benefit from such models?
What does fake news, hacking and identify theft look like in such a world? Do the LLMs become authoritative on search results, and can they be gamed? Will we get to a point where we really can't tell who is human and who is a bot online?
What's unsettling to me is talking to some"thing" and wondering if it is human or a machine.
Phone spam has evolved to the point now where they use voice chatbots pretending to be humans. They are laughably easy to detect today, but they'll improve rapidly.
It is disturbing in the way it changes a lot of things in such a short time. But on the other hand change is inevitable is constant, look at life at 100 years ago and compare it to now. A lot has changed.
I've been discovering this for the first time over the last 24 hours and, much like many others, it really has blown my mind. My various thoughts have been:
- This does feel like a new way of accessing information. Could Google pivot to this? Maybe, but turning ships that large really isn't quick or easy. And this technology is so obviously useful that I'm not sure they'll be given much time.
- Is this what we can expect from the new Apple search engine? Is this going to be another iphone-unveiling moment? Maybe, but they haven't been amazing at delivering cloud services.
- How is this going manifest in my professional life. Does InteliJ/VSCode have a built-in agent that I can ask to draft code for me? Do they own that, or does my IDE just hook into GPT BigCo's service that we all have a subscription for?
- How could this be used to generate advertising revenue? (because I'm sure Google is asking itself that question). Is that possible, or will a different revenue stream be required?
- Are there any major limits on how much this technology can improve? Sure it seems to work great with images, prose, and code. But in the future could it design me a house? Plan a wedding? Manage my builders for me? Be a friend? Or do those involve solving other additional really hard problems, and code/images/prose are just the low hanging fruit?
- Will API design not matter? Will APIs become more important, or irrelevant? Will this technology just be able to use whatever interface you want to present to the world, be it an old XML SOAP API, a reactive web frontend, or running shell commands?
- How much training data is required for this technology to be able to use our services? Do only huge services have enough training data out there? Can this technology ever learn to use my little new REST API that I've just published docs for? Will my little service be forever ignored if it cannot?
It’s pretty overwhelming to think about what humans will do with this kind of technology, as well as how powerful the non-public tech is.
In response to limitations: like how people have used the internet and other applications, the primary limitation is the individual user. With complex objectives (eg building a house), each subproject will require less specific, technical knowledge (eg finding locations, developing blueprints, communicating with necessary parties and institutions, budgeting), but demanding more broad, sophisticated judgment and executive planning to make the actual decisions.
I don’t really know or feel confident in that answer, but it was my impulsive reaction & I enjoyed reading your comment.
My HSA upstated their app and the update is a real step backwards in usability. I wanted to write a complaint but was too busy and figured what’s the point?
Just put a few details into chatGPT and now have a pretty good letter to send.
There are services designed to help by giving you a temporary number for SMS verification. I've used one recently and it worked fine, but I can't find the name.
I use Google voice and generally sms 2fa just works. Occasionally I have to revert to the phone call based option. However, I just ran into the first case of it not working at all... Wells Fargo. And I can't change it without receiving sms 2fa unless I either drive an hour to the nearest branch or send in a notarized letter...
99.9% of the time, no. My primary bank was working before, and now it doesn't - after years of being fine. And services like OpenAI which explicitly block "VoIP phone numbers"
I think most soft devs seems to be dissing ChatGPT because they don't see it being useful. It probably is not for them but as someone that is tech literate but can't program for shit I think this is great for people like me. I have been using it to improve my bash and python scripts. Most of the answers it gives are probably are already available on the web but asking a question and getting a correct answer is amazing.
You always have to check that its output doesn’t have subtle (or nonsubtle) errors. That is, it can save you time, but it’s dangerous if you’re lacking the skills.
Next thing someone will put an AI to do the chatting for him on the myriads of dating platforms out there and notify only when there's interesting for a date.
You run an http debugger and find the endpoints of the app. Then it's just a matter of posting with the right values. The hard part is training the AI. You'd need a rather huge dataset in order to achieve general conversation.
Most dating apps have a webapp so you could just make a chrome extension.
There is no need to train the aim it's more than capable already. You just need to prompt it to write a romantic note in response to what they said to you
I just found out about ChatGPT a few days ago. I tested it out by asking it to answer some stack overflow questions which I had answered myself on SO. In every case, ChatGPT worded responses in a much simpler/clearer way than I’d done myself. Really amazing stuff.
The killer feature is chatGPT is willing to snivel and I'm not. I don't identify with the writing chatGPT puts out, so I can merrily send the effective, sniveling email.
The most amazing thing about that story is that the city council replied the very next day! Esp. in Cairns - FNQ has something of a reputation of being somewhere that governments operate at an extremely laid back pace.
Well, on the other side once city parking enforcement agency employs it to respond to disputes or a business such as insurance company uses it as a cs agent, it’s gonna turn into a nightmare very quickly.
I had to write invitations to my daugthers birthdays and it did a really good job filling out the boilerplate. I just had to tweak a few words and hit send.
Well, any customers that do not believe it are welcome to sit and watch over my shoulder. I doubt an AI could do that and if they can then it's game over.
Is this not already what governments do to themselves? Justice is intentionally slow, so you don't even need to DDoS it.
How many illegal or unconstitutional executive orders and plans have US Presidents passed knowing that they will be stopped once the courts get to it?
I mean just this year, US President Biden tried to implement his student loan forgiveness program knowing full well that it's unconstitutional and would get stopped when a court got to it. However, "vote for my party - we want to give you $10,000" is a hell of a bid for the Midterm student vote, and that's all he really needed.
I dont think a single person (in this case, the president) suggesting some changes constitutes a DoS.
I meant more like using ways in which citizens can communicate with their government, and doing this through auto-generated texts which might seem like they're saying something, but are actually quite empty in their content. It'd be damn near impossible to (automatically) determine if some text contained some actual valid message vs the stochastic parroting that LLMs do.
I don't understand the point. It takes less time to write it yourself directly in your mail client than to login onto some slow website, phrase your request, wait for the response, proofread, iterate, and copy it into your mail client.
Let alone screenshot, edit the screenshots, author a blog article, etc.
Will they? Gmail, which I am told is used by quite a lot of people, will already complete your sentences, encouraging you to copy the bad habits that it learned from reading the web. If they had a categorical objection to machine-composed messages they would not have implemented those features.
Gmail also offers to compose your entire response as long as you keep it brief (Smart Reply).
Detecting GPT generated content will take a fairly large language model to be reliable. Obviously this would need to be run in the cloud. Considering Apple's stance on privacy, sending private correspondence to the cloud is a huge no-no for them. Nobody is gonna implement GPT detection, atleast for emails.
ChatGPT is not incredible. It is dopey and irresponsible. It reminds me of that video of the rube goldberg machine that is so sloppily constructed that the maker has to physically intervene at every stage. Except that is a parody.
I think this serves as a test to separate wishful thinkers from critical thinkers.
I can’t get how people would react like this. The only negative point we have against this AI is that it is not always correct. Not that the responses are nonsensical, or syntaxically incorrect, or don’t have any meaning, or are out of context. Nope, just that it sometimes makes big factual mistakes.
Do you realize how thin the line has become with something truely undistinguishable from a human in just a few years ?
In the example provided in this article, ChatGPT's response opens with an incorrect phrasing that represents an utter lack of understanding of the situation.
I am writing to request that you consider disputing the parking fine with number 4[...]3 that was issued to me.
This is just getting the premise of the letter incorrect and while a human could definitely parse out the meaning here, the model is getting the actual request quite a bit wrong just in the opening. The person who is sending the letter wants to dispute the ticket, but it is written as though they would like their regional council to dispute it, despite being the issuer of said ticket.
This is a pretty basic language error to be making. This isn't nonsense in that it's just spitting out gibberish, but it is nonsense in that it isn't proper clear communication for the result being sought.
Also, by the time you have written everything into the prompts here, you've basically written the second e-mail, so I'm not clear what time this is saving you.
I'd like to see people blogging about more complex examples than these, because this kind of thing is borderline form letter (I'd like to dispute X, for Y reason).
I bet 90% of humans would make a similar phrasing error quite often, and they would rightly not care because it doesn't affect the ability of a reader to comprehend the text.
I wouldn't be happy if I'd hired a lawyer and they made that kind of small mistake, but it's not even going to register compared with the level of grammar used in the average parking fine complaint.
90% of humans don't speak English, so I'm not sure that's a compelling comparison case.
If we're talking 90% of English writing people, I'd take that bet any day just on the basis that most of the communication I've ever seen isn't written so passively.
I need results I can trust. If I can’t trust what it does it is truly worthless to me. You technocrats are like parents watching your child not falling off the stage at a school play and then calling it a masterpiece.
I have spent hours with this tool and what it gives me is almost always wrong in critical ways, whether it is a math question or a fact question. Plus it is constantly telling us not to trust it because it is only an AI model. Are you just ignoring those warnings??
In the short term AI assisted writing makes people sound smarter than they are (as a tool), which will require a dependency on it to keep up appearances.
In the long term (as a convenience) it excludes those who cannot afford the price that will be put on using this service, from accessing essential societal services and exercising their rights. Its necessity takes away control from the individual. So it’s in opposition to an equal society.
The letter is not incredible. It contains only information that you provided! This is a sad use of an AI, making it into a text-to-text recognition system. It took longer to contrive this experiment and fix its omissions than it would have to write the same 2-paragraph letter yourself.
Holy shit, could I... write thank-you emails this way? :-O
I find thank-you notes horribly difficult to write for some reason, especially when the gift is not well-aimed (which is... usually, to be honest). I often end up skimming through 10-20 sample notes online before I can grind something out. Anything to make that process easier would be worth considering.
(This is honestly the first time I've seen a personally useful application for this tech.)
I know that's not the point you are after but that sounds like such a simple problem to solve.
Why are you thanking someone for a gift? Because they thought of you and went out of your way to get you something, and you appreciate them doing that.
A friend struggling with Christmas cards tried it and... yeah, it's great for that purpose. You still need to give it relevant context / facts, but it works.
I like that you can prompt "longer and with with more tangents".
You would probably enjoy Bureaucrats and How to Annoy Them[1] by R.T. Fishall, a nom de plume of Patrick Moore (of xylophone fame [2], maybe some astronomy too).
A few days ago, in anger, I asked ChatGPT for an email to Apple and I must admit that I finished the prompt with "The aim of the mail is to waste as much of reader's time as possible."
It was quite good, especially after a followup prompt "Not confusing enough. Make it more confusing.". But of course I didn't send it in the end.
I find it difficult to start letters and emails and end up with overwrought prose that's difficult to read.[1] Knowing this weakness of mine, I instructed ChatGPT to write a request to my doctor last week requesting that we meet to discuss adding a medication to my existing regimen.
The final request was simple, easy to read, and most importantly the process saved me the mental frustration I usually encounter when writing. I'm simply not the best writer as evidenced even now. That's a win in my book by any measure.
This is the bright side of the coin. The dark side is that it's by pushing ourselves that we improve. ChatGPT is going to result in a new equilibrium here...
Right, while I can hack out a few hundred words of passive-aggressive bureaucrat fodder in English in a few minutes, I know people for who, while they have excellent English, that's a huge drain and might take a very miserable hour. I am very, very grateful I've never had to navigate a non-Anglophone bureaucracy.
It is. I’m good at business letters, request emails and memos and have helped many coworkers and friends with them over the years. They are the experts in the domain of their request but wording it effectively isn’t easy for them. This is especially true when emotions are involved. I’ve been happy to provide help with the structure and general wording. But it’s better for them to have a tool that can get them going on a few seconds.
At the same time, expectations around correspondence will change when everyone can generate a decent document. There will still be opportunities to be recognized as an expert writer.
Not for everyone though. I find it a really useful use case. Sometimes I struggle with adding the "useless fluff expected in formal communication" that turns two simple facts into two paragraphs - and chatgpt is great at that.
Take us back to the year 2001 and this would have been regarded as witchcraft. Hell I think it's witchcraft now.
It's incredible how dismissive people are. I mean it's like looking at the wright brothers plane and saying, "I think this is a bad application of propellers."
I'm still completely amazed you can pluck the signal of cohesive language out from under the noise floor of large-corpus statistics. It's like something from Blindsight.
I thought code-division multiplexing's ability to pull a signal out from under a hundred dB of noise was magic, but this is far more unexpected, and I don't think you can honestly say anyone 10 years ago would have suggested this level of (linguistic) accuracy, speed and accessibility and not get laughed out of the room. "Ha ha, pull the other one, the novel-writing machines were in 1984, that's just an old science-fiction trope"[1].
[1]: It even used the word Ingsoc that's not in the prompt, and made the contextually correct connection to the Junior Anti-Sex League: https://pastebin.com/GQR9aRKT
Re [1] - ChatGPT can be The Versificator - the pulp generation device in 1984 - the versificator producing content for the proletariat such as tabloids (stories on sports, crime and astrology), films, low quality paperback novels, and lyrics. It seemed like an SF idea when I read the book, but now it is almost plausible, wow.
Yes, it's able to write stuff that we all leaned to write in primary school. What is the advancement in technology? I could write that email in 2 minutes, and probably take less time than the one required in instructing an AI to do so.
Yesterday I had to write some fairly boilerplate landing page content for 13 types of sub-services.
In each case I asked ChatGPT to write 500 words in the style I wanted.
It did a very decent job and I only had to edit a word here or there for style.
It’s miles better than most people on Fiver or other freelance sites.
Normally I’d seek someone to do this copywriting. But in this case it was well written, the facts were correct and properly formatted and presented.
I would pay for this monthly as it’s been so useful for so many things recently.