I am slightly confused.
So, Sam Altman and collegues discovered a very powerful thing called Q*. This will make OpenAI very powerful and will make the board a lot of money.
So, why did this cause the board to fire him?
From the article, it seemed like the board was too afraid of Q* and fired him to stop it being released without proper security features.
Could someone please help clarify this?
Thanks.
It’s complicated. The board at OpenAI is (or was) focused on AI safety and is not entirely comprised of investors. Their goal was not to maximize profits.
I forget the exact phrasing, but the board said they fired Altman for not being completely honest with them. Based on the wording of the board’s rationale for firing Altman, it seems likely that Altman was not forthright about the capabilities of this breakthrough, possibly because the board would then halt its development out of safety concerns.
I'm a little curious why this post has so many upvotes. I guess it shows that things really have changed a lot.
I know very little about ML (essentially nothing, I have a background in economics and, a bit, of statistics), but isn't AGI still miles away from these models?
Like, my understanding of LLMs is that they essentially "predict" the right word to respond to a prompt and then write a new word based on the previous one and so on. Actual human level intelligence seems to me to be a degree of complexity higher.
Does it? In what quantitative way?
I dunno. I am asking dude
Im not that clued up either. But given how the 'predict the next word' method trivially enables the wonders of AI photo, video, audio, code, etc. generation it's not inconceivable that it can also be extended to areas like logic, cognition.
The rhetorical question of other guy is actually quite a good one. Since predicting the next best word alone already seems to do such a great job of convincing us of intelligence, perhaps the onus is on us to describe how our intelligence is anything more than essentially a predictor of next words.
It's Q-asterisk, not Q-star. Somebody read the footnote please
You might rent a GPU from runpod or another cloud provider.
Memory requirements:
34B Model Memory Requirements (infer)
Seq Len vs Bit Precision
SL / BP | 4 | 6 | 8 | 16
-----------------------------------------------------------
512 | 15.9GB | 23.8GB | 31.8GB | 63.6GB
1024 | 16.0GB | 23.9GB | 31.9GB | 63.8GB
2048 | 16.1GB | 24.1GB | 32.2GB | 64.3GB
4096 | 16.3GB | 24.5GB | 32.7GB | 65.3GB
8192 | 16.8GB | 25.2GB | 33.7GB | 67.3GB
16384 | 17.8GB | 26.7GB | 35.7GB | 71.3GBI don't understand how a change in management is linked to these news. If anything, shouldn't it go the other way ? That sounds like good news ?