Sara Hooker
@sarahookr
Building intelligence that evolves @adaption_ai. Built @Cohere_Labs, @GoogleBrain, @GoogleDeepmind. ML Efficiency, Multimodal\lingual.
11.7K Following 65.3K Followers
A model that knows your business and gets sharper with every task is the catalyst for faster innovation.
Co-founder @sarahookr will discuss turning AI into a competitive advantage at the @ftlive Future of AI conference with @Mastercard, @Cognigy, @BentleySystems.
Show more
my favorite bagel place in the world is now high fashion
Loved our conversation on generative and malleable interfaces with @HaijunXia!
So many interesting ideas around how people can explore, interact with, and shape intelligence.
Had a blast discussing Dynamic Task UIs with Simra @its_sshahid from @adaption_ai .
If you missed it, this article I published today
summarizes my view of the emerging interface paradigm with AI agents.
Show more
Millions of developers build in @cursor_ai. Now you can build custom models without leaving it.
Adaption is live on the Cursor Marketplace. Go from dataset to trained model, right from the Cursor agent.
Show more
had way too much fun editing this one. the reactions are everything 😭
i’ve watched this 40 times and the ending still gets me
At #PyTorchCon# 2026, Sara Hooker @sarahookr (@adaption_ai) examines the limits of scaling larger models and the shift toward adaptive intelligence.
Sara discusses post-training, test-time scaling, adaptive compute, and hardware co-design as alternatives to simply scaling pretraining. She highlights AutoScientist examples spanning agent harness optimization, agent memory, and dataset creation.
Her keynote explores how making AI easier to adapt and customize could enable more people to build and own frontier AI.
2026 prediction? "You will build and own frontier AI"
Show more
we all took the bait here. but I took the bait, the hook, and a front-row seat to my own farewell party 😂
Something’s cooking in Chicago 🎥
Can the Adaption team spot a made-up research paper?
Scientific Lead @hugo_larochelle and the team played Real or Fake Paper.
Will be at @PyTorch conference tomorrow.
looking forward to a great line-up of speakers. 🔥
The next generation of intelligent systems won't be defined by scale.
It will be defined by intelligence that learns from your work. Co-founder @sarahookr on what that takes, live at @PyTorch Conference on October 20.
Show more
This is actually more a side-effect of the death of pretraining.
As transformers saturate, most pretraining results are tweaking data recipes, not massive stepwise changes in performance.
Turns focus to innovation in postraining and test time compute. This is good for competition, since pretraining concentrated resources in a few labs.
Show more
GLM 5.3’s biggest leap came from post-training alone.
“GLM 5.3 was built entirely by post-training the previous model. For me, that was a wake-up call.”
“Post-training is far more GPU-efficient than pre-training from scratch. Plenty of companies can’t afford to train a model from scratch, but they have proprietary data. They can post-train themselves and beat some of the closed models.”
“There’s still so much juice left to squeeze from post-training. We compared GLM 5.3’s cyber capabilities with 5.2, and it’s a night and day difference. That’s all just post-training.”
Show more
Back in SF
Best part of jet lag. Early morning run.
Which review complaints get ICLR papers rejected?
I used @adaption_ai's Visualize to cluster the harshest review of 5,000 ICLR 2026 papers.
Complaints about weak baselines or novelty drop acceptance to ~23%, while those about compute cost or real-world validity leave it at 36%.
Show more
“Do you think that getting married young has made you less ambitious?”
Welcome to the first episode of SOTA Wives with Sarah Chieng (Head of DevX at Cerebras, @MilksandMatcha) x Vivian Midha Shen (General Partner at Y Combinator, @vivianmshen).
In episode 1: finding your person before getting rich, turning down startup funding to finish MIT, what a YC partner sees in founders, and how a birthday post became “psychological violence.”
4:08 Finding your husband before the exit
8:48 Did we marry away our ambition?
18:04 The internet called Sarah's birthday “psychological violence”
20:47 Does your YC partner secretly think you’re doomed?
23:47 Building a startup with your ex-boyfriend
28:54 SF cat parties
Show more
Intelligent, adaptive interfaces are becoming ever more important as AI becomes central to our lives.
Visualize is our first release in this area, enabling you to interact with your data, get actionable insights and improve it.
Awesome work from @its_sshahid and team!
Show more
Great take and quite true!
Basically all labs have an RL-shaped hammer and are paying data vendors millions for env-shaped nails.
The closer the env to digital worlds of real people, higher the price.
Show more
Test-time data is coming.
After continual learning real-time adaptive data will be the breakthrough.
@adaption_ai is so underrated.
My taxi driver in Singapore tonight.
Has a Lego collection. Representing only horror movies he likes. Quite fun.
I can’t tell what all of them are. I see Freddie, Adam’s family, the ring. Pretty detailed Lego landscape.
Show more
It’s too bad we already decided to sponsor Lucha libre after our Mexico City offsite.
Jokes. Jokes. Although somewhat serious about lucha libre.
Truly super special atmosphere at @F1. My voice and hearing is completely shot.
Show more
