research: {post-training, rl} @GeorgiaTech
Joined March 2019
- Tweets18.1K
- Following474
- Followers780
- Likes17.1K
Pinned Tweet
We can now do effective continual learning for #LLMs
I am happy to present Info-SDFT: Information-Proximal Self distillation. The method nicely balances plasticity versus stability and works very well.
We did over 500 experiments on this! Thanks to @KickItLikeShika!
We open source code and checkpoints for you to use:
Code: github.com/KickItLikeShika/i…
Checkpoints: huggingface.co/collections/K…
Paper: arxiv.org/pdf/2609.24646
holy fucking shit man
ABSOLUTE BLOODBATH IN THE MARKET.
Over $2 trillion has been wiped out from US stocks in just 50 minutes after the open.
This is the fastest wipeouts in US stock market history.
The S&P 500 is down 2.7% and has now erased all of its gains since the midterm results.
Most of the selling is in tech stocks:
• Nvidia: -7%, -$300 billion
• Tesla: -9%, -$140 billion
• AMD: -12.5%, -$100 billion
• Micron: -10%, -$95 billion
• Broadcom: -5.5%, -$90 billion
Health care, consumer staples and energy stocks are up today.
Did Democrats just kill the AI trade?
check out the Work Models by @MotoniqLabs!
I did always say, if you do robotics right, you can get some amazing stuff happening. Some amazing findings here:
1. You can improve VLAs if you do things right and have a model that thinks about the VLA's thinking. Really cool idea!
2. You can save tons of money if you do things in the right way.
3. Claude and GPT are not as great as advertised when it comes to EAI!
This is some cool next level results! You can even see improvements to LeWM!
Original Post: motoniq.ai/blog/making-robot… #AI #MachineLearning
Ahmed retweeted
I did always say, if you do robotics right, you can get some amazing stuff happening. Some amazing findings here:
1. You can improve VLAs if you do things right and have a model that thinks about the VLA's thinking. Really cool idea!
2. You can save tons of money if you do things in the right way.
3. Claude and GPT are not as great as advertised when it comes to EAI!
This is some cool next level results! You can even see improvements to LeWM!
Original Post: motoniq.ai/blog/making-robot… #AI #MachineLearning
flawless
immaculate
phenomenal
stunning
incredible
Ride onboard with Max Verstappen as he propels his Red Bull car to #F1Sprint pole 🤩
#F1 #SingaporeGP
Ahmed retweeted
Software engineering is dead... In other news, a software engineer with 18 years of experience (me) just spent 15 hours refining a big new feature, asking questions that the frontier models hadn't even considered. Software engineering isn't going anywhere.
Ahmed retweeted
Ten years of Sad Affleck memes and it turns out he'd just been trying to install CUDA drivers the whole time.
collateral damage of using llms, work isn’t being properly reviewed, and a whole lot of it is complete bull crap
Surprise: the verified Navier-Stokes proof doesn’t match the one in the paper.
arxiv.org/abs/2610.08144
one funny thing about this is how llms are getting more cracked in maths, and they still suck in the thing that they are optimized for the most, which is coding
let an llm touch a proper huge codebase, and you better get ready to revert these changes
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
