Kai
@kaicathyc
Followers
16K
Following
2K
Media
43
Statuses
443
Occasionally here. Currently research @OpenAI.
Joined February 2013
3 years ago we could showcase AI's frontier w. a unicorn drawing. Today we do so w. AI outputs touching the scientific frontier: https://t.co/ALJvCFsaie Use the doc to judge for yourself the status of AI-aided science acceleration, and hopefully be inspired by a couple examples!
74
206
1K
I'm specifically excited about having tasks that go beyond math and coding. AI researchers tend to optimize for what's familiar, which is reflected in recent successes in math and coding competitions. Broader benchmarks help us see how models can improve outside those domains.
0
0
40
šØ Actress Diane Keatonās Cause of Death Revealed Read more
0
299
3K
As academic benchmarks become saturated, it's increasingly important to have benchmarks that actually reflect real-world capability. GDPval tasks are created by experienced professionals and reflect authentic work outputs from 44 occupations. šļø
In our pairwise grading setup, we find model performance over the past year has increased roughly linearly on GDPval. The best models are already starting to approach parity with industry experts.
2
2
47
1/n Iām really excited to share that our @OpenAI reasoning system got a perfect score of 12/12 during the 2025 ICPC World Finals, the premier collegiate programming competition where top university teams from around the world solve complex algorithmic problems. This would have
141
449
3K
We trained a new bio AI model and used it to design new Yamanaka factor variants for iPSC generation, achieving over 50x higher efficiency! š§¬
At @OpenAI, we believe that AI can accelerate science and drug discovery. An exciting example is our work with @RetroBiosciences, where a custom model designed improved variants of the Nobel-prize winning Yamanaka proteins. Today we published a closer look at the breakthrough. ā¬ļø
16
11
175
Have you integrated APOL1 genetic testing into your practice? Discover the No-Cost APOL1 Genotyping Program for eligible patients sponsored by Vertex Pharmaceuticalsāhelping you deliver precision care without added cost. Learn more today!
19
25
220
a very good group of humans
6/n Iāve been lucky to work with many fantastic teammates here at @OpenAI, specifically with @alexwei_ @bminaiev @oleg_murk for prepping for IOI and building on top of the long term work on competitive programming by @_lorenzkuhn @MostafaRohani @clavera_i @andresnds @ahelkky
2
0
44
"there will no longer be an excuse for ugly internal applications"
model creates a finance dashboard in 5 minutes that devs estimate would have taken many hours:
1
2
41
@michpokrass is unbelievably goated at intersecting research and engineering and sheās grown a whole team around it. Honored to ship under her leadership!!
7
25
309
AI agents are moving from demos to teammates. Part I of our playbook covers the foundations: what an agent is, how itās structured (objective, instructions, tools, content grounding), and how to spot high-impact use casesāso you can ship secure, reliable agents at scale.
0
8
51
Details of the study are shared in @Eric_Wallace_ās thread and our paper here: https://t.co/d8clQI7Pnl We hope that this approach can serve as useful guidance for other groups releasing open-weight models.
openai.com
In this paper, we study the worst-case frontier risks of releasing gpt-oss. We introduce malicious fine-tuning (MFT), where we attempt to elicit maximum capabilities by fine-tuning gpt-oss to be as...
0
0
9
While MFT improves performance, gpt-oss stays below the High threshold in OpenAIās Preparedness Framework. These findings contributed towards the decision to release these models for the world to use.
1
0
11
Open-weights canāt go back in the box. Before permanently releasing gpt-oss models into the ecosystem, we estimated marginal frontier risk by deliberately eliciting bio/cyber capabilities via malicious fine-tuning (MFT).
Today we release gpt-oss-120b and gpt-oss-20bātwo open-weight LLMs that deliver strong performance and agentic tool use. Before release, we ran a first of its kind safety analysis where we fine-tuned the models to intentionally maximize their bio and cyber capabilities š§µ
3
6
117
It's Fall, Get Ready to Ride! Use the VSEAT to ride farther, longer, and healthier. Reduce pain and discomfort and order today!
3
12
121
Enjoy these lil models. A lot of people worked very hard on them :)
Our open models are here. Both of them. https://t.co/9tFxefOXcg
6
7
640
gradually, then suddenly
To summarize this week: - we released general purpose computer using agent - got beaten by a single human in atcoder heuristics competition - solved 5/6 new IMO problems with natural language proofs All of those are based on the same single reinforcement learning system
2
1
47
want to add that OpenAI has been incredibly supportive during this kerfuffle. initial application was filed almost 3! years! ago before my time here - really sucks to get denied after waiting for so long and unable to return home, but all in all feel very lucky to be where i am
Update: It sounds like there might have been paperwork issues with the initial green card filing (done over 2 years ago). It's still a shame that this means @kaicathyc has to leave the US for a while but there's reason for optimism that this will all be resolved.
25
14
724
Hello tweeter. Iāll be in Vancouver for an indeterminate amount of time! I have no friends there so would be excited about meeting new people :) Hopefully will return home sometime this year but if not shall make the best of it.
It's deeply concerning that one of the best AI researchers I've worked with, @kaicathyc, was denied a U.S. green card today. A Canadian who's lived and contributed here for 12 years now has to leave. Weāre risking Americaās AI leadership when we turn away talent like this.
100
46
1K
