I just tried it first time.
I was happy when I saw it integrates with GitHub Copilot subscription and then I went ahead with Claude Sonnet 3.7 (from the GitHub Copilot provided models) and made some cool changes to our app.
Spending drinking water for toilet flushes is indeed a problem. Perhaps not CO2 measurements directly, but informing people in general of how much high quality water is wasted on flushes alone will hopefully bring more momentum into more efficient flushing mechanism and introducing grey water systems to new and old buildings alike. Good idea!
people downvote your sarcasm, but if you do the calculations you're kinda right.
1Kg of Beef costs:
- The energy equivalent of 60.000 ChatGPT queries.
- The water equivalent of 50.000.000 ChatGPT queries.
Applied to their metric Mistral Large 2 used:
- The water equivalent of 18.8 Tons of Beef.
- The CO2 equivalent of 204 Tons of Beef.
France produces 3836 Tons of Beef per day,
and one large LLM per 6 months.
So yeah, maybe use ChatGPT to ask for vegan recipes.
People will try to blame everything else they can get a hold on before changing the stuff that really has an impact, if it means touching their lifestyle.
Wow, thanks. I’m even coming up with 500K chatGPT queries for the amount of energy consumed as a KG of beef, though I might have moved a decimal place somewhere - feel free to check my math :)
“average query uses about 0.34 watt-hours of energy” - or 0.00034MWH
Using this calculator: https://www.epa.gov/energy/greenhouse-gas-equivalencies-calc... - in my zip, 0.0002KG of CO2 per MWH. (Though, I suppose it depends more on the zip where they’re doing inference, however this translation didn’t seem to vary much when I tried other zips)
Then, 99.48KG/0.0002KG= 497,400 chatGPT queries worth of CO2 per KG of beef?
This is spot on because there can’t be two issues that exist simultaneously. There can only be one thing that wastes enormous amounts of energy and that thing is beef
You can try to misconstrue and ridicule the argument,
but that won't change the math that if you have one thing that causes 1 unit of damage, and another thing that causes 100.000 units of damage, then for all intents and purposes the thing that produces 1 unit of damage is irrelevant.
And any discussion that tries to frame them as somewhat equally important issues is dishonest and either malicious or delusional.
My guess, as I've expressed earlier in the comment chain, is that it's emotionally easier for people to bike-shed about the 0.01% of their environmental impact, than to actually tackle things that make up 20%.
And no it's not only beef (which is a stand-in for meat and diary), another low hanging fruit is also transport, like switching your car for a bike.
But switching from meat and diary to a vegan diet would reduce up to 20% of your personal environmental impact, in terms of CO2.
And about 80-90% of rainforest deforestation is driven directly or indirectly by livestock production.
So it's simply the easiest most impactful thing everyone can do. (Switching your car for a bike isn't possible for people in rural areas for example.)
>1 unit of damage, and another thing that causes 100.000 units of damage, then for all intents and purposes the thing that produces 1 unit of damage is irrelevant
You make a good point. A problem is only a real problem if you can’t find a bigger thing that makes it look small by comparison. For example, the worldwide concrete industry creates more co2 than beef does so there is no reason to stop eating beef if you enjoy it.
Now I know that some might say that “all of this is cumulative” or “the material problems that stem from entrenched industries is actually a reason not to invent completely novel wasteful things rather than a justification for them” but in reality only two things are true: only the biggest problem is real, and the only problem is definitely some other guy’s doing. If I waste x energy and my neighbor wastes y amount, a goal of reducing (x+y) is oppressive whereas a goal where I just need to try to keep x lower than y feels a lot nicer.
I agree. Humans have been eating meat and doing construction for the entire history of civilization, they are not the sort of things that could be affected by posting online. LLMs on the other hand are new, largely in the hands of a small handful of companies, and a couple of those companies are bleeding cash in such a way that they might actually respond to consumer pressure. It is cynical to compare them to things that we know will not change as a justification for a blanket excuse for them.
Seeing as these models being wasteful is integral to the revenue of companies like OpenAI and Anthropic, the more people that tell them that the right business strategy is to start perpetually building data centers and power plants, the less incentive they have to build models that run efficiently on consumer hardware.
They just suggested a different bike shed — one for the purpose of their argument won’t ever get fixed. J-pb’s point is that running a bunch of generators 24/7 in Memphis is fine because people eat meat. Inefficient LLMs in the real world are okay because people could theoretically become vegan but have not. It’s just a thought experiment
If something costs too much, and you find a way to completely pay for it, that's not bikeshedding.
And it's not a thought experiment. It's a very real suggestion. If you're worried about the resource cost from your personal use, doing something to 100% offset it lets you stop worrying.
> become vegan
For one day per year. Replacing a day you would have otherwise eaten meat. That is an extremely attainable action for anyone that cares enough about LLM resource use enough to strongly consider avoiding them. It's not something that "will not change".
By the way, your goal of running efficiently on consumer hardware isn't as great as it sounds. One of the best ways to improve efficiency is batching multiple requests, and datacenter hardware generally uses more efficient nodes and runs at more efficient clock speeds. There's an efficiency sweet spot where models are moderately too big to run at home.
And it really undermines your argument when you throw in this stupid strawman about elon's toxic generators. You know j-pb was talking about typical datacenter resource use and not that. Get that insulting claim out of here.
It is only a “very real suggestion” if you believe that your argument might be effective.
Do you believe that “skip meat for a day use LLMs for a year” will have a climate impact?
Because if not then you agree with me that in this case theoretical vegans are just being used to justify more real consumption, not less
>stupid strawman about elon's toxic generators
They exist in the real world, right now. It is a real phenomenon and no matter how many vegans I imagine it’s still there. I’m not really clear on why the real thing that’s really happening is a strawman unless you think that the existence of that system is so bad that it undermines your position. Even then it wouldn’t be a strawman though, just a thing that doesn’t support your position that using LLMs is categorically fine because you can picture a vegan in your head
> Do you believe that “skip meat for a day use LLMs for a year” will have a climate impact?
If "use LLMs for a year" is enough to count as having a climate impact (negatively), then yes I believe "skip meat for a day use LLMs for a year" is enough to count (positively).
I'd be tempted to write off both of those, but the whole point of your argument is to consider LLM resource use important, so I'm completely accepting that for the sake of the above argument.
There are no theoretical vegans involved.
And the suggestion doesn't even involve vegans, unless there's a massive contingent of americans that only eat meat one day per year that I wasn't aware of.
And to get at what I think is your core objection: The fact that people can do this isn't being used to let companies off the hook. If only 2% of LLM users set up a meat skipping day, then LLM companies are only 2% let off the hook.
But at the same time let's keep a proportional sense of how big the hook is.
> They exist in the real world, right now. It is a real phenomenon
The strawman is you accusing people of supporting those generators.
> your position that using LLMs is categorically fine
>If "use LLMs for a year" is enough to count as having a climate impact (negatively), then yes I believe "skip meat for a day use LLMs for a year" is enough to count (positively).
Sorry, I should have clarified. In this case I meant “argument” as a thing that leads real people to either understand or agree with your position, not the construction of an idea in your mind.
With that in mind, do you think that “skip meat for a day use LLMs for a year” will convince enough real people, in real life, to not eat meat, that it offsets the emissions from LLM use?
Like imagine the future.
Since LLM use is a new category of energy use, you would have to convince people that haven’t already been convinced to skip meat by animal cruelty, health, philosophy, or existing climate concerns. People that were vegan before LLMs became popular obviously don’t count. The group of people that resisted decades of all that messaging will now make a meaningful adjustment to their consumption to cancel that out — and there will be enough of these new part time/full time vegans that it offsets the entire chat bot industry’s energy usage.
Do you imagine that being what happens?
If not it’s just somebody advocating for increased consumption in real life by invoking imaginary vegans.
As somebody that’s spent years as a vegan I am incredibly wary of “vegans can recruit” as a pitch. I’ve only ever heard that from people that have never tried to recruit in earnest or charlatans. Like I’ve mostly heard that from people that are not, never have been, and have no interest in being vegan.
Edit:
>The strawman is you accusing people of supporting those generators.
That’s not what a strawman is and it’s not an accusation, it’s an observation. If you say “I want subscription based online batched mega-high-compute language models” you are advocating for that industry, and those generators are part of it. Saying you feel that they’re somehow special and different because they’re icky does not make them any different from the thing that you say is necessarily the future. That you want!
I think anyone that does get convinced and skip meat should be able to use LLMs without shame or guilt, while we continue to pressure everyone else to save resources and we continue to pressure LLM companies to save resources.
LLM companies only get let off the hook if a very large fraction of their users do the meat skip thing, which is not very likely but could theoretically happen.
LLMs being a new category of energy use should get them some extra scrutiny, but only some. Maybe 3x scrutiny per wasted kilowatt hour compared to entrenched uses? If our real motivation is resource use, and not overreacting to change, LLMs should get some pressure but most of the pressure should go toward preexisting wasteful uses.
Nobody is advocating to ignore LLMs. But we shouldn't overstate them too much either.
And the giving up meat defense is not a defense for the companies, it's a defense for individual users that actually do it.
Like not an if or maybe thing, what do you see when you picture the future?
Do you think “Skip meat for a day use LLMs for a year” will produce enough new vegans to offset the energy usage and co2 produced by the LLM architecture of your choice?
Not asking if you want it to happen or if it’s something you can imagine could happen, I’m asking if you think it will
[_] yes
[_] no
Because if no, then the idea is just advocating for increased real consumption by invoking imaginary vegans!
Edit:
>LLM companies only get let off the hook if a very large fraction of their users do the meat skip thing, which is not very likely but could theoretically happen.
The person I was initially talking to took the position that LLM companies have negligible impact because people can be vegan. J-bp was saying that LLM companies shouldn’t be on anybody’s radars because uh, meat is 100,000 times worse.
The person you hopped in to defend was saying that LLM companies do not and should not have a “hook” because meat eaters exist
> It was a yes or no question [...] I’m asking if you think it will
[x] no
> Because if no, then the idea is just advocating for increased real consumption by invoking imaginary vegans!
Wrong.
> The person I was initially talking to took the position that LLM companies have negligible impact because people can be vegan.
He said "LLMs are not the problem here", which is true.
And he was arguing for individual use being offset when he said "maybe use ChatGPT to ask for vegan recipes".
The top level comment was also about individual use. "I would really like it if an LLM tool would show me the power consumption and environmental impact of each request I’ve submitted."
The comments right before you replied were also about individual use. "lifestyle choice".
> J-bp was saying that LLM companies shouldn’t be on anybody’s radars because uh, meat is 100,000 times worse.
The 100,000 number was a throwaway hypothetical to make a point. Not a number he was applying to LLMs in particular. Two lines later he threw in a 2,000x too.
And what he said is that LLM companies are not "somewhat equally important". Which is true. He didn't say you should ignore them entirely, just to have a sense of proportion.
-
Edit: Here is an important distinction that I think isn't getting through. There are multiple separate points being made by j-bp:
Point A, about not eating meat for a day, is only excusing anyone that actually does it. It's not a hypothetical that excuses the entire company.
Point B, about the size of the impact, suggests caring less about LLMs based on raw resource use. Point B does not care about the relatively small group of people that take up the offer in Point A. Point B is just looking at the big picture.
Then it is not a “very serious suggestion”. It is a thought experiment which should be taken with commensurate weight.
>Wrong
Explain what “skip a day of meat do a year of LLMs” is then. If it’s not just an ad for feeling good about using LLMs, what is it?
>The 100,000 number was a throwaway hypothetical to make a point
>Two lines later he threw in a 2,000x too.
Alright he said that meat is 2,000 times worse than language models as well as 100,000 times worse than language models. He might have meant 100k but could also mean 2k.
Do you have a real problem in real life where if somebody called you and said “it’s gotten two thousand times worse” versus “it’s gotten a hundred thousand times worse?” the former would be fine and the latter alarming?
If yes, what is the problem? Why was it a problem at 1x? 2000x? 100,000x? Why was it a problem at at 1x and 100,000x but not 2000x?
> Explain what “skip a day of meat do a year of LLMs” is then. If it’s not just an ad for feeling good about using LLMs, what is it?
You can stop being part of the problem if you do it. The problem still exists, but you are no longer part of it. You reduced it by more than your fair share. While the problem would stop existing if everyone made the same choice, there's no pretense that that's actually going to happen. LLM companies are not being excused by such an unlikely hypothetical.
j-lb also made an argument to not care much about LLMs at all, but it was separate from the "skip a day of meat" argument. That's where the big multiplier comes in. But again, separate argument.
I don't want to argue about the example ratio he used. The real ratio is very big if the numbers cited earlier are correct. So if you're going to sit here and say 2000x might as well be arbitrarily large then I think you just joined the "LLM resource use doesn't matter" team, because going by the above citation 2000x is in the ballpark of the correct number, so LLM use is 1 divided by arbitrarily large, making it negligible. Congrats.
Just wanted to chime in and say you represented my case perfectly and got all my points (and their separation) 100%!
You're right, I never said we should not care about LLMs because we also "rightfully don't care about meat".
To me the whole AI resource discussion is just a distraction for people who want to rally against a new scary thing, but not look at the real scary thing that they just gotten used to over the years.
In a sense it's the `banality of evil`, or maybe `banality of self destruction`:
The “banality of evil” is the idea that evil does not have the Satan-like, villainous appearance we might typically associate it with. Rather, evil is perpetuated when immoral principles become normalized over time by people who do not think about things from the standpoint of others.
We've gotten so used to using huge amounts of resources in our day to day lives, that we are completely unwilling to stop and reflect about what we could readily change. Instead we fight against the new and shiny, because it tells a better story, distracting us from what really matters.
In a sense we are procrastinating on changing.
It's not the Skynet like AI that is going to be the doom of humankind, but the hot-dogs, taking your car for the commute, and shitty insulation.
> Whatever you need to tell yourself to keep eating meat buddy.
I’m not the one that brought up moralizing or food. I can’t really comment on your relationship with your diet but it kind of seems like you saw somebody mention power usage and unprompted shared “well I don’t eat meat or cheese or yogurt” so I guess keep that up while you use enough energy to power your home to write some code slower than you would without it?
Where does a Youtube LetsPlay video fall into that calculation? My understanding is that a single watch of a video is orders of magnitude more than a day's active use of ChatGPT.
ETA: The above link is at the bottom of the original submission's README. (https://github.com/sst/opencode) I posted it without context, and I have no opinion on the matter. Please read theli0nheart's comment below for an X rebuttal.
I’m the founder and CEO of Charm. There are claims circulating about OpenCode which are untrue, and I want to clarify what actually happened.
In April, Kujtim Hoxha built a project called TermAI—an agentic coding tool built on top of Charm’s open-source stack: Bubble Tea, Lip Gloss, Bubbles, and Glamour.
Two developers approached him offering UX help and promotion, and suggested renaming the project to OpenCode. One of them bought a domain and pointed it at the repo.
At the time, they explicitly assured Kujtim that the project and repo belonged entirely to him, and that he was free to walk away at any point.
We loved what Kujtim built and offered him a full-time role at Charm so he could continue developing the project with funding, infrastructure, and support. The others were informed and declined to match the offer.
I also mentioned that if the project moved to Charm, a rename might follow. No agreement was made.
Shortly after, they forked the repo, moved it into their company’s GitHub org, retained the OpenCode name, took over the AUR package, and redirected the domain they owned.
To clarify specific claims being circulated:
- No commit history was altered
- We re-registered AUR packages for continuity
- Comments were only removed if misleading or promotional
- The project is maintained transparently by its original creator
The original project, created by Kujtim, remains open source and active—with the full support of the team at Charm.
It's pretty funny to refer to your libraries for building a TUI as an "open-source stack". From the commonly accepted vision of a "stack" it's a pretty thin slice. It's like saying "my over-engineered component library is a stack because it involves 15 layers of abstraction!".
Neither of these companies are focused on LLMs or AI, they're both just using this as AI dust to sprinkle on top of their products.
You’re implying the door has now closed for people to get into coding agents. It’s a bit early for that don’t you think? These guys might one day be considered part of the founders of coding agents for all we know.
So which project is which here? Is Kujtim sst on github and is sst/opencode his project? Is opencode-ai/opencode the one that the two developers that went rogue made (if I understood the tweet correctly)? Or did I get it backwards?
I'm so confused by this. I saw this post on HN, and then ended up installing the opencode-ai/opencode one via homebrew somehow (I guess I did a google search and ended up on the wrong github). But then sst/opencode is the one that links to the website opencode.ai and I was reading the docs on that website. Which one is better?
Both are go based using charmbracelet's gui libraries. There's actually a note about the project you posted being developed under the charm repo now but it doesn't seem to be public. Maybe they are the same project?
Kujtim started opencode few years back, they were developing this it even before any other CLI tools were in the market. Few months back thdxr(dax)(SST) and Adam started contributing to opencode. And quickly became the biggest contributors to the project. I think they also wanted to make it more presentable and Dax bought a domain and stuff while working on it. At some point charm approached Kujtim for some deal to move opencode to charm and keep working on it under them. Dax and Adam wanted to keep it open source as is. (Dax's commits were somehow squashed and removed at this point too)
So they ended up rewriting opencode with the same name in TypeScript TUI away from Kujtim's vision. And thats where we are, since then opencode doesn't seem to have much progress done but Dax's opencode is being worked on non-stop.
This is a third party retelling of this story from some post I read, as I came to know about it only after Dax started working on TS TUI for opencode under SST.
In EU in some countries because of extra taxes it's really expensive to buy a bigger car so 3+ kids can fit in. But those rich people who sit alone in huge luxury SUVs they still sit alone. They have the money while families usually don't.
I just hope this doesn't go in this wrong direction.
Many people have legitimate needs and uses for pickup trucks and SUVs, and the definition of "dangerously oversized" and unsafe is rather subjective. Should semi trucks be banned because they are "dangerously oversized"? What should people use to move large and heavy items? Should farmers be prevented from having the vehicles they need to do their job?
While I agree that large pickup trucks and SUVs pose an increased risk to pedestrians, particularly in urban areas, and should be discouraged as single person commuter vehicles, particularly in urban areas, there are a lot of other use cases. What should someone who needs to move bulky/heavy/rough things with regularity and can only afford to have one vehicle use?
Discouraging negative externalities through taxation makes sense, but setting taxes to be so punitive that they make it difficult for people to afford the vehicles they need to do what they need to do is also harmful.
As an example, I have a pickup truck and a regular car. Driving around town I use my regular car, but when I need to move large or heavy things I use my truck. It's much more convenient to be able to use my truck when I need it rather than having to rent one every time or hire a company to move things for me.
The pickup truck is large (RAM 1500), and some would argue it's dangerously oversized, but its size is needed when I need to do truck things with it. The truck being affordable means I can afford to have a regular sized vehicle for tasks that don't involve moving big/heavy/dirty things.
It's not subjective. RAM 1500 and trucks like it kill more pedestrians and cyclists in collisions because they have a larger frontal cross-section that limits the driver's view in front of the cab and is more likely to cause fatalities when a collision does occur.
I agree that trucks like the RAM 1500 are useful in many applications. They should be taxed appropriately (in a way that offsets or negates what currently amounts to subsidies in the US market), and manufacturers should be required to enable the driver to see a certain minimum distance in front of the vehicle and obstruct a certain maximum angle around A pillars. Trucks and SUVs over a certain size should also have speed limiter governors that activate on city streets. It is not acceptable to have drivers of these vehicles - which were originally developed for specialized industry applications - speed in areas where it directly endangers pedestrians and cyclists.
What's subjective is the threshold of what is considered "dangerous". I do agree improving visibility would be a good thing, and incentivizing safer designs through taxation could be good. Insurance prices already do that to an extent. I do agree that SUVs and pickup trucks have become taller than necessary in recent years; pickups from the 1990s were appreciably lower and had better visibility without being any less useful for moving things around.
The CAFE exemptions for SUVs and trucks compared to cars don't make sense to me and encourage maximizing vehicle footprint, so changing those rules would make sense. What else do you consider "subsidies" for this size of pickup trucks? Not being subject to the "chicken tax"? (not directly relevant to me as I'm in Canada)
In the US it's CAFE gaming, 6000LB GVW tax incentives and arguably fuel, registration and tolls aren't (sufficiently) scaled with vehicle size. The chicken tax thing definitely contributes even to Canada's market since it's relatively rare for automakers to consider it separately from the US one.
Would you support more stringent driving licence for such vehicles? So that anyone who needs a truck puts up with the faff of getting one, but regular Joes who just want moar car might not bother?
> Should semi trucks be banned because they are "dangerously oversized"?
Poor example, they already are. You require a special license. This is a tax, and a rather severe one. Never mind that those drivers are also directly responsible if their vehicle malfunctions. That's how some big rig drivers are able to get 150 years in jail because their brakes went out.
Your computer is not unsafely designed in a way that kills tens of thousands of people a year in accidents.
Also, your computer's power consumption is in fact regulated on a semi-voluntary basis via a program called Energy Star. People are fine with it because Energy Star certification saves them money. By contrast, the problem with trucks is that their unsafe design is subsidized by the government, so smaller more efficient vehicles that would normally save people money end up being penalized.
Last I checked my desktop doesn't slam into pedestrians/bicyclists trying to use what public infrastructure we have in the US. If unbounded vehicle bloat is systematically contributing to the deaths of citizens then yes, we have a collective responsibility to regulate it.
I can’t think of many consumer products more dangerous and widespread than cars. Going from a suggestion that we make them safer to “What’s next, regulate everything?” is quite the jump.
Dude we are literally doing calculations of the power usage of our employees at their home office for a ISO 14001 certification. It’s beyond ridiculous
Why exactly is it insane ? To reliably differentiate (let's assume it's possible for the sake of argument) between "you made this" and "you didn't make this" or at least "a human made this" seems to carry mostly (if not only) benefits.
the problem is your parenthetical - it's not possible, so attempting to do so isn't actually really possible. what's worse than a watermark? one that doesn't actually work.
Open AI literally said they have a semi-resilient method with 99.9% accuracy. It will become full-resilient for practical purposes if all LLMs implement something similar.
> Open AI literally said they have a semi-resilient method with 99.9% accuracy.
They also said many other things that never happened. And they never showed it. I bet $100 they do not have a semi-resilient method with 99.9% accuracy, especially with all the evolving issues around "human vs computer" made content.
I bet you also the `semi-` in the beginning leaves a lot of room for interpretation and they are not releasing this for more reasons than "our model is too good".
I really don't see what's in it for them to brag about a non-existent feature that's not in their commercial interest when its non-implementation can be turned into a stick to beat them with, so I believe they have something, yes. I don't necessarily believe the 99.9%, but with that proviso I'll take your bet.
The Verge doesn't report this, but other reports have said that the watermark is easily beatable by doing things like a Google Translate roundtrip, or asking the model to add emoji and then deleting them.
> the problem is your parenthetical - it's not possible, so attempting to do so isn't actually really possible. what's worse than a watermark? one that doesn't actually work.
If it's not possible to watermark, then just ban LLMs.
Tech people have this weird self-serving assumption that the tech must be developed and must used, and if it causes harms that can't be mitigated then we must accept the harm and live with it. It's really an anti-humanist, tech-first POV.
The comment was referring to models close to the recent releases from Meta and Mistral, reaching up to 405B with performance competitive with large commercial vendors. These models absolutely can't be trained without significant investment, and their inference without a cloud provider isn't cheap either. As I had mentioned, nothing short of not having released the weights could have stopped the abuse, but still, a fraction of it could be deterred, hopefully adding up to a few billion less spam pages for search engines to serve back to you.
As for the rationality of watermarking itself, firstly I'd like to reiterate, no spam wave of this magnitude and undetectability has ever happened in the history of the web. A word processor cannot write a petabyte of propaganda on its own. A Markov chain can't generate anything convincing enough to fool a human. Transformer-based LLMs are the first of their kind and should be treated as such. There is no quick analogy or a rule of thumb to point to.
If statistical watermarking is proven to have sufficient recall and error, there'll be nothing to lose in implementing it. A demand already exists for detecting AI slop; half-working BERT classifiers and prejudiced human sniff tests already provide for it, with little incentive to reduce false positives. With watermarks, there'll be a less painful, more certain way to catch the worst offenders. Do you really think the same operations that produce papers with titles like "Sorry, as an AI model..." or papers with pieces of ChatGPT UI text will care to roundtrip translate or rewrite entire paragraphs?
We already had this exact dilemma back when email spammers tried Bayesian poisoning [0]. Turns out, it actually creates an identifiable pattern, if not for the system, then for the user on the other side. People will train themselves to look for oddly phrased sentences or the outright nonsense roundtripping produces, abrupt shifts in writing style, and other heuristics, and once the large enough corpus is there, we can talk about training a new classifier, this time on a much more stable pattern with less type-I errors.
That's how you can weaken a country, create a movement with the media that calls for the destruction of their own energy production. Try that in Russia...
Octoscreen is helping clients to monitor locations and important information.
We ingest a lot of data from various sources and display the most important information for our clients.
We are looking for an Elixir developer who is also a generalist and can work with a diverse set of technologies.
The ideal candidate is someone who is following tech trends in the industry, keeps his/her knowledge up to date and is proactive in improving the product.
We don't do agile or any strict process. We focus on the product and the customer.
Small dev team (<5 people).
Useless is probably not the right word but it's a good way of summing up a lot of the current problems. If the model can clearly identify when something is an exact quote and also know the source then its output could be trusted for the most part and much more easily verified. It would certainly elevate the output of the model from "random blog post or forum chat" to "academic paper or official report" levels of trustworthiness. Citing sources is hugely important for validation, cited text allows an immediate lookup and simple equality check for verification after which you can use it as context to validate the rest of the claims. Like I said, it's a standard we apply to humans who have an equal propensity for hallucination, mistakes, and deception because it's a tried and true method for the reader to check the claims being made.
I, for one, agree with the original statement. I think the hallmark of enlightenment (for example, in the scientific method) is that we are able to externalize the expert knowledge, that is, experts are usually required to provide reasoning behind their claims, and not just judgements. This is because we learned that experts cannot be 100% trusted, only if we can verify what they say we can somewhat reach what is truth (although expertise still provides a convenient shortcut).
So not demanding this (and more) from an AI (an artificial expert) is a regression. AI should be capable of wholly explaining its reasoning, if we are to consider its statements to be taken seriously. It is understandable that humans have only limited capability to do that, since we didn't construct human brain. But we have control over what AI brains do, so we should be able to provide such an explanation.
It is somewhat ironic that you yourself do not provide any argument in favor of your disagreement.
This isn’t meant to totally disagree with your point (there’s some stuff I agree with in here) but I’m having trouble seeing the point about regressions.
To use another example, a new NoSQL DB not having joins is a regression. Does that mean no one is justified in releasing a new NoSQL DB?
As long as from "NoSQL" it is clear that you mean "I can't do joins", then it is OK. I think LLMs (and similar models like Stable Diffusion) are really cool for things like fiction, but to rely on them to tell you the truth is dangerous. So I am not really sure why the models have to be trained on NYT articles in the first place.
Providing reasoning and providing citations are not the same thing. Reasons can be provided without citations; citations can be provided without reasons.
LLMs have astounding utility citations notwithstanding.
They are different, but perhaps you misunderstood my argument.
Issue of plagiarism aside, we reason from facts, and it's the facts (or some other analysis, which is itself a fact) that should be sourced. That's why I agree with the original statement, and I argue not from a (moral) POV of preventing misattribution or plagiarism, but from a (practical) POV of veracity.
I think you just identified another problem with LLMs - we don't know their values, either.
Of course when you listen to human experts you're using the shortcut (and you do it based on trust), as I already argued. You have an option (in most cases, in free societies) to dig beyond just experts judgement, you can study their reasoning, and understand their sources of both values and facts.
Anyway, I disagree with citations not being required. If Wikipedia had no citations it would be less useful (and more prone to contain misinformation). Same goes for Google. So the next best things we have to "artificial brain that contains all the human knowledge" have citations, and for a good reason.
What are the citations actually required for though?
Another way to ask this is: what value remains without them?
I'll add this as well: humans produced valuable knowledge for thousands of years without the use or standard of citations.
To be clear, I think citations are highly valuable and desirable and I very much want LLMs to cite when appropriate. However, I think the necessity of this is overstated.
Edit: what you said of experts can be said of LLMs as well.
> What are the citations actually required for though?
For me, yes! I do often reference (cite) myself when thinking, by making and reading notes, or materials that other people wrote. If I relied only on my own memory, I would be unable to think more deeply.
And this is, interestingly, where the current LLMs seem to break down - they can reason short proofs but cannot scale their reasoning to longer chains of thoughts correctly (were they capable of doing that, they would have no issue producing citations to back up their claims). They operate intuitively (Kahneman's system 1) and not rationally (Kahneman's system 2).
And thus lot of that "valuable knowledge" that humans produced over the years have been hopelessly wrong, precisely until somebody actually sat down and wrote things up (or communicated things out, people working together can sort of expand the working memory as well).
And also patently false. Knowledge is knowledge, it's useful without source citations.
Is the knowledge of how to do CPR somehow ineffective because I can't cite whether I studied the knowledge from website A or book B? Is reality a video game where skills only activate if you speak the magic words beforehand?
The statement is equally hyperbolic both as quoted and in the original context. LLMs often can't quote sources, and those models are nevertheless useful to lots of people. Makes it hard for me to take the rest of the comment seriously.
reply