Your post isn't providing any argument, you're just providing statement after statement of a position that is pretty well known (no emergent behavior, just pattern matching) and pretty popular here on hackernews. Maybe the link provides all the good stuff, but I'd suggest you at least provide some flavor of what your arguments are.
Each argument could easily fill a book. Some valuable insights are discussed in the linked article, but most of the evidence comes from a series of research papers demonstrating that LLMs perform poorly when confronted with out-of-training distribution prompts. For example, even state-of-the-art LLMs can fail simple tasks like counting the number of "r" letters in the word "strawberry" or fall victim to the reversal curse (e.g., LLMs trained on "A is B" often struggle to deduce "B is A"). Another interesting read on this topic comes from MIT's research, which further explores these limitations: (https://news.mit.edu/2024/reasoning-skills-large-language-mo...)
The reversal curse is mostly about understanding specific logical relationships rather than identity statements. It's not about statements like 'Joe is human' implying 'human is Joe,' but rather about LLMs struggling to reverse relationships or roles, such as turning 'A is B' into 'B is A.' Check out the summary of the paper: "We expose a surprising failure of generalization in auto-regressive large language models (LLMs). If a model is trained on a sentence of the form "A is B", it will not automatically generalize to the reverse direction "B is A". This is the Reversal Curse. For instance, if a model is trained on "Valentina Tereshkova was the first woman to travel to space", it will not automatically be able to answer the question, "Who was the first woman to travel to space?". Moreover, the likelihood of the correct answer ("Valentina Tershkova") will not be higher than for a random name. Thus, models do not generalize a prevalent pattern in their training set: if "A is B" occurs, "B is A" is more likely to occur. It is worth noting, however, that if "A is B" appears in-context, models can deduce the reverse relationship. We provide evidence for the Reversal Curse by finetuning GPT-3 and Llama-1 on fictitious statements such as "Uriah Hawthorne is the composer of Abyssal Melodies" and showing that they fail to correctly answer "Who composed Abyssal Melodies?". The Reversal Curse is robust across model sizes and model families and is not alleviated by data augmentation. We also evaluate ChatGPT (GPT-3.5 and GPT-4) on questions about real-world celebrities, such as "Who is Tom Cruise's mother? [A: Mary Lee Pfeiffer]" and the reverse "Who is Mary Lee Pfeiffer's son?". GPT-4 correctly answers questions like the former 79% of the time, compared to 33% for the latter.
Code available at: this https URL."
Wolfram is serious about his software, but, to my mind, has been closer to a crackpot scientist than a real researcher the last few decades. Him backing something up diminishes my subjective probability that this is something "serious"
I keep seeing this "expansionist dictatorship" applied to China when the USA is discussed. The USA has invaded plenty of countries in the last few decades, has a history of colonialism (Cuba, Philippines, Puerto Rico...sure, less than some European countries, but still).
Which countries has China invaded in the past few decades?
Did you say "last few decades" to conveniently exclude their invasion of Vietnam? Not that it matters. Policy should not be made based on a naive extrapolation of historical track record. Culture, interests and leadership are all things that change over time.
Their publicly broadcasted intention is to change the status quo, forcefully if needed. That's a euphemism for invading Taiwan. They keep saying it, over and over. Beyond that, there's a militarism, nationalism and irredentism that permeates Xi's leadership and the culture he has created in his country, which did not exist to the same extent under Deng. The confluence of such factors have historically been a bad omen.
This does not mean that the US should start a war with China. It means the US should pivot its focus to Asia and continue the policy of containment, which is a maintenance of the peaceful status quo through a combination of sticks and carrots. It means the US should be aware that there is a rival there who may start a war on their own terms and on their own schedule when they believe they are capable of defeating the US.
Since WWII
Tibet (annexed)
Korea (at invitation of the North Korean government, but invaded South Korea)
India
USSR
Vietnam
China has been fairly quiet and well behaved since 1980, but it is current quite publicly talking forcefully reintegrating Taiwan and has had continuous naval disputes in the South China Sea.
If we are talking about colonialism, “China” is a land empire that has absolutely dominated its neighbors and often conquered them. Making an apples to apples comparison to European colonialism isn’t very useful and I don’t pretend to know details, but historical China has plenty of expansionist and domineering episodes.
I think it will eventually be good public policy to make it illegal to post massive amounts of texts produced by AI without disclosing it.
As with all illegal things on the internet, it's difficult to enforce, but at least it will make it more difficult/less likely
How about articles written by human charlatans? Claiming they are 'doctors' or 'scientists'. Or posters claiming something that didn't happen? Like a... pro bullshtter claiming he was denied apartment renting because of his skin color. He could make a lot of money if that was true. But poster is still taking ads place, payed by poor 'suffering' minority. Another example 'influencers' who pretending, or really being, experts advise you on forums about products. The tell mostly the truth, but avoid some negative details and competing products and solutions. Without disclosing their connections to businesses.
Shorter version: intentional bullshtting never ends, it's in human, and AI, nature. Like it or not. Having several sources used to help, but now with flood of generated content it may be not the case anymore. If used right this has real affect on business. That's how small sellers live and die on Amazon.
That's a good point. Latin letters map to a big number of IPA sounds.
I think this is specially true with consonants. Vowels, however, might be the same. For example, Spanish has five vowels (a, e, i, o, u in IPA). Japanese also has five IPA vowels and only the "u" is different in IPA
As a speaker of Scandinavian languages, literally none of the vowels in English map correctly to how we pronounce the same vowels. Several of the English vowels are diphthongs, which we spell out with double vowels. Like eng. "i" is our "ai", eng. "a" is our "ei" - "a bridge" is translated to "ei bro", where the articles are pronounced exactly the same. Or the Scottish word for home, "hame", is pronounced exactly like we say the same word, "heim".
When learning German it was a revelation that languages could pronounce the same letters in the same way in every word, and therefore you could accurately predict the pronunciation of a word from the spelling.
English does not do this; every word has its own pronunciation, only loosely related to how it is spelled. I think every native English speaker has had the experience of learning a word from reading it, and subsequently mis-pronouncing it because we had to guess at the pronunciation. E.g. I suffered acute embarrassment from mis-pronouncing "Hermione" when talking to a friend about the Potter books. I grew up in the UK but had never met the name before and my guess at pronunciation was entirely reasonable but entirely wrong.
Though I lived for a while in Ireland, and they have it worse. My friend Mebd laughed at me a lot.
Yes, this is also part of the reason why spelling bees are a uniquely American thing - in most other languages it is trivial to predict pronounciation from spelling, so a competition makes no sense.
To add to the confusion, in English people might pronounce the article "a" as "ei" (like when reciting "ABC") or "uh" (like the start of "under"). I think most Americans do the latter. I do, at least.
I'm a native Greek speaker and to my ears all European languages' vowels are weird with the exception of Spanish, which is completely normal. All those airy sounds: caaat dooog, haaaaouse, taaaaime, etc like a little fish trying to eat a much bigger fish. Let me not start with French, or Italian. Conversely, when I hear myself or another Greek speaker speak English it's like there's a little guillotine in the back of the neck that snaps shut just when a vowel is starting to form: c't, d'g, ta'm, etc.
That's the difference between [e] and [ɛ] in IPA (in IPA, with "narrow" phonetic transcription, you enclose the sounds in brackets). In American English, "ace" is [eɪs] and "mess" is [mɛs]. But I don't think that's right; I'm pretty sure both the Japanese e and Spanish e sound exactly the same.
And if you pronounced [e] and [ɛ] to native speakers of either Spanish or Japanese, they most likely wouldn't be able to differentiate the two sounds consistently without having had training. I know that in Spanish, realization of e can be either vowel depending on the speaker and context; they might pronounce "tierra" as [tjɛra] and "mesa" as [mesa].
> And if you pronounced [e] and [ɛ] to native speakers of either Spanish or Japanese, they most likely wouldn't be able to differentiate the two sounds consistently without having had training.
I am a Spanish guy living in Japan, so I can confirm it. I didn't have any idea that the two sounds are supposed to be different because to me they sound exactly the same.
Fun fact: Japanese people are surprised that we have the same five vowels (although the U is a bit different), and that we can get the correct pronunciation very easily.
>they most likely wouldn't be able to differentiate the two sounds consistently without having had training
I learn English since I was pretty young, I believe I'm quite fluent (I mostly use English for work, I lived for two years in an English speaking country, I read books in English, etc), and I still have problems distinguishing some English vowels. I think sound acquisition is one of the hardest things to learn for a non-native speaker.
Japanese e would only ever be [ɛ]. The constructs to produce an [e] sound would be transliterated as "ei" (the quoted are just letters, not meant to be IPA).
So, while I disagree that they sound "exactly the same", I do agree that in most cases the differences between [e] and [ɛ] wouldn't be enough to cause confusion between speakers of either language.
What's your source for this?
Wikipedia says "/e, o/ are mid [e̞, o̞]" [1]
>The mid front unrounded vowel is a type of vowel sound that is used in some spoken languages. There is no dedicated symbol in the International Phonetic Alphabet that represents the exact mid front unrounded vowel between close-mid [e] and open-mid [ɛ], but it is normally written ⟨e⟩.
>it breaks ability for non-speakers to infer meanings
Not sure what you mean by this. Do you mean that it's less convenient for people that don't speak / read Chinese? Why would that be a relevant metric?
You may be missing that character standards have changed over time and that different writing styles (草书，行书) are implicitly simplifications. You can think of latin or Russian cursive as a simplification of the printed letters.
In practice, the phonetic component has been mangled / evolved over time, so simplification doesn't make things more or less difficult for students (be it 5 year old native speakers or 50 year old non native speakers).
Do you think using capitals at the beginning of a sentence aids comprehension?
I view punctuation and spelling rules as a way to maximize comprehension (akin to having a linting standard). In non formal writing, I don't see any harm in avoiding capitalization (at least it doesn't seem to me to help understanding / reading speed, etc at all).
It's like people typing "K" instead of "OK". It's disrespectful to the reader, suggesting that the reader is not important enough to warrant typing an extra letter.
One would expect Altman to know how to use the SHIFT key when running a massive business, but, hey - once you achieve escape velocity from society, you don't have to live by its norms or grammar rules.
I can assure you that it would cost most people people here a promotion or a raise if they did this at work.
The problem is with the word "our".
If it's just private companies, the biases will represent a small minority of people that tend to be quite similar. Plus, they might be guided by profit motives or by self-censorship ("I don't mind, but I'm scared they'll boycott the product if I don't put this bias").
I have no idea how to make it happen, but the talk about biases, safeguards, etc should be made between many different people and not just within a private company.
