Feitong Yang retweeted
@gracexyzheng and @xinjin have been wonderful friends and collaborators of the team since the very beginning. very excited for this work to be out and looking forward to even more!
GPT-Rosalind surfaced a new lead for drug discovery: could restoring cholesterol transport protect neurons from degeneration?
DCTN1 is a gene linked to Perry syndrome, a rare neurodegenerative disorder. In mouse-brain data, disrupting it increased expression of genes involved in making cholesterol. GPT-Rosalind suggested a possible explanation: cholesterol may not be reaching where neurons need it.
That gives scientists a specific mechanism to test and, if confirmed, a direction for developing new treatments for neurodegenerative disorders.
With @perturbai_tx and @scrippsresearch, a team of thousands of GPT-Rosalind agents worked in parallel, connecting gene-expression results from millions of mouse brain cells with scientific literature in just 32 minutes. We’re now expanding this collaboration to identify new drug targets.
openai.com/index/scaling-…
4
12
219
52,352
Multiagent exploration babystep on drug discovery. One day you can imagine multi-agent system would just do these end to end.
GPT-Rosalind surfaced a new lead for drug discovery: could restoring cholesterol transport protect neurons from degeneration?
DCTN1 is a gene linked to Perry syndrome, a rare neurodegenerative disorder. In mouse-brain data, disrupting it increased expression of genes involved in making cholesterol. GPT-Rosalind suggested a possible explanation: cholesterol may not be reaching where neurons need it.
That gives scientists a specific mechanism to test and, if confirmed, a direction for developing new treatments for neurodegenerative disorders.
With @perturbai_tx and @scrippsresearch, a team of thousands of GPT-Rosalind agents worked in parallel, connecting gene-expression results from millions of mouse brain cells with scientific literature in just 32 minutes. We’re now expanding this collaboration to identify new drug targets.
openai.com/index/scaling-…
1
0
0
242
Feitong Yang retweeted
We’ve acquired Orchestra Bio, a platform that helps biopharma teams coordinate the complex process of drug development.
Together, we’ll build on Rosalind, our AI models and tools for life sciences research, to streamline those operations and help new treatments reach patients sooner.
Welcome @ashokaraj_ and team!
openai.com/index/openai-t…
Together, we’ll build on Rosalind, our AI models and tools for life sciences research, to streamline those operations and help new treatments reach patients sooner.
Welcome @ashokaraj_ and team!
openai.com/index/openai-t…
464
690
7,399
915,517
Sunday fun: Join the community to experiment simple games with AI. Very AI slop right now, but baby steps to learn.
0
0
2
272
Just want to add one thing, some poeple pointed out: this is more of visual awareness, or visual aspect than full defintion of consciousness. You close your eyes, thinking from the sky to the sea, you consciousness fluctuate, but you won't experience binocular rivalry there.
So the IT/STS part of neurons are more likely correlate with visual. The interesting thing is your early visual cortex will get both eyes stimuli for sure, so those neuron activities are not directly correlated to your "conscious perception". At one point, neural activity need to correlate with what your "experience". That unselective to selective correlation change is the fascinating part. How does that happen? Where is the signal from?
So the IT/STS part of neurons are more likely correlate with visual. The interesting thing is your early visual cortex will get both eyes stimuli for sure, so those neuron activities are not directly correlated to your "conscious perception". At one point, neural activity need to correlate with what your "experience". That unselective to selective correlation change is the fascinating part. How does that happen? Where is the signal from?
This phenomenon at behaviour level is not new, the bio-substrate is interesting. Cogpsy research has found a lot of these decades ago: 1) attention frequency switch around 100ms level, 2) attention guide and affect conscious perception, 3) bistable or binocular conscious competition fluctuate back and forth around that frequency, 4) you can suppress conscious percption into unconscious, via binocular attention competition.
It was fascinating to do attention/awareness/consciousness experiments in psychology. I had spend a few good years on it. It was very hard to get bio-substrate for human experiment because fMRI is too slow and EEG is bad in localisation.
0
0
4
427
I can relate. I had a very tough week trying to defend Prism abuse and didn't sleep much. A side effect is I cleaned up a ton of code and hardened lots of pieces in infra.
they've been non stop trying to attack our systems for the past 48 hours
a side effect is everything is getting hardened but damn we basically need agents going 24/7
because that's what they have
1
0
9
632
This phenomenon at behaviour level is not new, the bio-substrate is interesting. Cogpsy research has found a lot of these decades ago: 1) attention frequency switch around 100ms level, 2) attention guide and affect conscious perception, 3) bistable or binocular conscious competition fluctuate back and forth around that frequency, 4) you can suppress conscious percption into unconscious, via binocular attention competition.
It was fascinating to do attention/awareness/consciousness experiments in psychology. I had spend a few good years on it. It was very hard to get bio-substrate for human experiment because fMRI is too slow and EEG is bad in localisation.
It was fascinating to do attention/awareness/consciousness experiments in psychology. I had spend a few good years on it. It was very hard to get bio-substrate for human experiment because fMRI is too slow and EEG is bad in localisation.
1/ We discovered something totally unexpected about the neural structure of consciousness that overturns decades of thinking on its biological basis and brings us significantly closer to the recipe for generating conscious awareness.
New preprint: biorxiv.org/content/10.648… (led by @HesseJanis, with @FrankLanfranchi, @_conor_mcgrory, @Yuelin_Shi, and @EngelTatiana)
1
5
32
2,620
Attended a conference where VC and Scientists share the challenges and progresses about curing cancers. Damn, curing cancer is really hard, and it's not just solve one problem.
Need to learn more to understand what a life science breakthrough looks like. To gain Research taste
Need to learn more to understand what a life science breakthrough looks like. To gain Research taste
1
0
9
654
It is great that @OpenAI and @AnthropicAI are working together to contribute to the Math community!
1
0
21
832
Feitong Yang retweeted
We're expanding access to GPT-Rosalind for academic research.
The new Rosalind Academic Plan helps scientific labs bring AI into the work of understanding biology, exploring ideas, and deciding what to test next—with sponsored access for eligible labs at U.S. universities.
ts with questions whose impact may take years to impact human health.
The new Rosalind Academic Plan helps scientific labs bring AI into the work of understanding biology, exploring ideas, and deciding what to test next—with sponsored access for eligible labs at U.S. universities.
ts with questions whose impact may take years to impact human health.
340
302
3,323
428,962
Reading reactions to OpenAI's math drop is very educational. Some claim they would spend forever to understand those results, while others already tightened the bound 2^100 fold. Some claim AI speaks gibberish, while others point out AI has real insights. Some don't want to touch those results, while others criticize, fix, improve with concrete content. Some worry about the next generation of mathematicians, while others are happy for next generation. All have fair points.
1
0
2
437
It turns out auto-complete is still very helpful and useful. Never thought I would be back to Tab+Enter... But boy this one is interesting
Day 5/
Composer predictions in the desktop app. Often leads to a double take with how on point they are. Included in the Pro plans without consuming usage.
0
0
0
210
Feitong Yang retweeted
One of the OpenAI results that dropped was some sharp lower bound for some known embedding result (that an embedding of edit distance into l1 that was published in 2005 is optimal).
I saw this and used it as an opportunity to understand this embedding finally. The 2005 paper is quite finicky, and I was never able to fully understand what’s going on, but given how fundamental the problem is, and neat the answer is, I was always curious! (The optimal distortion is exp(sqrt(log n log log n)), and when it’s a bound like this, usually something interesting is going on).
So I tried to prompt a cleaner version of the upper bound yesterday from Astra, and I think I succeeded (not in one shot, it required quite a bit of tinkering and understanding of what’s going on and where the bottlenecks are in the original proof)! The result is just 2.5 pages, where 1.5 pages are completely standard, and 1 page is some clean combinatorial fact about edit distance. I used Lean along the way to verify the steps in the proof. But most importantly, it allowed me to digest and enjoy the proof, while historically edit distance results like this were impenetrable for me.
I am of course cautious that it can be slop and LLM psychosis, but after I slept I think it stands? Currently running it by some friends who understand this area well.
I am obviously not going to post it or anything (our world does not need more math slop floating around right now), but I just thought it was a very satisfying experience intellectually.
I saw this and used it as an opportunity to understand this embedding finally. The 2005 paper is quite finicky, and I was never able to fully understand what’s going on, but given how fundamental the problem is, and neat the answer is, I was always curious! (The optimal distortion is exp(sqrt(log n log log n)), and when it’s a bound like this, usually something interesting is going on).
So I tried to prompt a cleaner version of the upper bound yesterday from Astra, and I think I succeeded (not in one shot, it required quite a bit of tinkering and understanding of what’s going on and where the bottlenecks are in the original proof)! The result is just 2.5 pages, where 1.5 pages are completely standard, and 1 page is some clean combinatorial fact about edit distance. I used Lean along the way to verify the steps in the proof. But most importantly, it allowed me to digest and enjoy the proof, while historically edit distance results like this were impenetrable for me.
I am of course cautious that it can be slop and LLM psychosis, but after I slept I think it stands? Currently running it by some friends who understand this area well.
I am obviously not going to post it or anything (our world does not need more math slop floating around right now), but I just thought it was a very satisfying experience intellectually.
7
5
133
12,088