For the record, I don't think it's true. I think it was a power play, and a failed coup at that. But it's about as substantiated as the "serious" hypotheses being mooted in the media. And it's more fun.
Your comment sounds like a rhetoric way to say that GPT is in the same class as autocomplete and that what autocomplete does sets some kind of ceiling to what IO functions that work a couple of bytes at a time can do.
It is not evident to me that that is true.
As they learn to construct better and more coherent conceptual chains, something interesting must be happening internally.
I think trying to model the world based on a single projection won't get you very far.
I smell a fallacy. Parent has moved from something you can parse as "LLMs predict a representation of concepts" to "LLMs construct concepts". Yuh, if LLMs "construct concepts", then we have conceptual thought in a machine, which certainly looks interesting. But it doesn't follow from the initial statement.
Could you at least elaborate what they are “not”? Surelly you are not having a problem with “LLMs predict language”?
There is nothing special about human intelligence threshold.
It can be surpassed by many different models.
Are there any papers testing how good humans are at predicting the next word?
I presume us humans fail badly:
1. as the variance in input gets higher?
2. Poor at regurgitating common texts (e.g. I couldn't complete a known poem).
3. When context starts to get more specific (majority of people couldn't complete JSON)?
https://nonint.com/2023/06/10/the-it-in-ai-models-is-the-dat... The ultimate model, in his (author's) sense, would suss out all patterns and then patterns among those patterns and so on, so that it delivers on compute and compression efficiency.
To achieve compute and compression efficiency, it means LLM models have to cluster all similar patterns together and deduplicate them. This also means successively levels of pattern recognition to be done i.e. patterns among patterns among patterns and so on , so as to do the deduplication across all hierarchy it is constructed. Full trees or hierarchies won't get deduplicated but relevant regions / portions of those trees will, which implies fusing together in ideas space. This means root levels will be the most abstract patterns. This representation also means appropriate cross-pollination among different fields of studies further increasing effectiveness.
This reminds me of a point which my electronics professor made on why making transistors smaller has all the benefits and only few disadvantages. Think of these patterns as transistors. The more deduplicated and closely packed they are, the more beneficial they will be. Of course, this "packing together" is happening in mathematical space.
Another thing which patterns among patterns among patterns reminds me of homotopies. This brilliant video by PBS Infinite Series is amazing. As I can see, compressing homotopies is what LLMs do, replace homotopies with patterns. https://www.youtube.com/watch?v=N7wNWQ4aTLQ
From this, we get comedy. A funny statement is one that ends in an unpredictable manner and surprises the listener brain because it doesn't have the meaning of that one already calculated, and hence why it can take a while to "get the joke"
I personally think a far more fundamental change is necessary to reach AGI.
https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chat...
I get that this is a "just for fun" hypothesis, which is why I have just for fun questions like what incentive does anyone have to keep clearly observed ai risk a secret during such a public situation?
But if there's one thing that seems very easy to discern about Ilya, it's that he fully believes that when it comes to AI safety and alignment, the buck must stop with him. Giving that control over to government bureaucracy/gerontocracy would be unacceptable. And who knows, maybe he's right.
* Current-gen AI is really good at tricking laypeople into believing it could be sentient
* "Next-gen" AI (which, theoretically, Ilya et al may have previewed if they've begun training GPT-5, etc) will be really good at tricking experts into believing it could be sentient
* Next-next-gen AI may as well be sentient for all intents and purposes (if it quacks like a duck)
(NB, to "trick" here ascribes a mechanical result from people using technology, not an intent from said technology)
Yes, actually. This is overwhelmingly true for most people. At the end of the day, we all fear being alone. I imagine that fear is, at least in part, what drives these kinds of long-term "existential worries," the fear of a universe without other people in it, but now Ilya is facing the much more immediate threat of social ostracism with significantly higher certainty and decidedly within his own lifetime. Emotionally, that must take precedence.
His existential worries are less important than OpenAI existing, and him having something to work on and worry about.
In fact, Ilya may have worried more about the continued existence of OpenAI than Sam after he was fired, which looked instantly like a: "I am taking my ball and going home to Microsoft.". If Sam cared so much about OpenAI, he could have quietly accepted his resignation and help find a replacement.
Also, Anna Brockman had a meeting with Ilya where she cried and pleaded. Even though he stands by his decision, he may ultimately still regret it, and the hurt and damage it caused.
