ML/AI at https://xitter.cf/t.co/8o8gaO50rv, https://xitter.cf/t.co/LnWYE5LAZ8, https://xitter.cf/t.co/sYatSqOfYd
Seattle, WA
Joined August 2012
- Tweets1.7K
- Following396
- Followers432
- Likes1.6K
Sergey Feldman retweeted
My Claude found a critical zero day vulnerability in the Code of Hammurabi, unnoticed by developers for 3,779 years.
Sergey Feldman retweeted
My attention span is worse than ever, agents have made it much harder for me to focus on any given thing
Can't make this up.
Really? 2008: "Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes" arxiv.org/abs/0812.4360 Tweet: xitter.cf/SchmidhuberAI/status/1…
Sergey Feldman retweeted
Introducing AstaBrief 8B, an open model that turns complex research questions + literature excerpts into cited reports.
Run it locally on your own hardware, with open weights + training data you can inspect & build on. 🧵
🤗 Download: huggingface.co/allenai/AstaB…
Sergey Feldman retweeted
Whenever I see a model top this chart, it's normally a bit of a red flag. Would love to be proven wrong!
Sergey Feldman retweeted
... do we need to start publishing graphs of indicators of AI impact on the world that show no change. We could publish a lot of those. Here are some (very vibe coded) examples:
Sergey Feldman retweeted
We have evidence that
1) AI agents are very good at improving software used to train AI
2) LLMs can be superhuman in many domains
These do not combine to AI will be so powerful it will kill us in N years. We're forecasting inevitable saturation and unknown future breakthroughs.
Sergey Feldman retweeted
Here is the grumpy blog post that @SergeyFeldman and I wrote 13 years ago when @finkd invaded NeurIPS to buy up all the researchers:
explainmydata.blogspot.com/2…
Everything we were worried about happened almost immediately but the field also made prev unimaginable progress
Sergey Feldman retweeted
Includes a section on my flailing attempts to port MHCflurry from TensorFlow to PyTorch, which was mercifully finally dragged across the finish line by a combination of Claude Code and Codex in MHCflurry 2.2.0
(with @timodonnell & @SergeyFeldman)
Coding agents are helping scientists spend more time advancing research, taking on everything from routine maintenance and targeted optimization to complete redesigns and new systems.
While agents can reliably execute on ambitious projects, researchers must still define the scientific questions, verify results, and take a stance on long-term ownership.
Sergey Feldman retweeted
Late 2020s going to be full of breakthroughs communicated as tweets like “lol fine tuned SlopCoder-V7 and it proved P=NP” with a link to fully correct alien math
Sergey Feldman retweeted
Deep conversation about the specs of the next pretraining run on our new B300 cluster.
What do you want to see in a future Olmo release?
With @lambdaviking @RenZhongzheng @TaoTianhua @yashassamaga @jowenpetty
(I am hiring Sr. REs)
Sergey Feldman retweeted
On might feel that the AI coding era diminishes the value of a library like @scikit_learn, as coding is easy.
It's the reverse: one needs reliable building blocks, and fighting with Claude for eg reliable convergence of LogisicRegression takes energy, expertise, and tokens
Sergey Feldman retweeted
It is not every day that I can tell you about a new explanation for the origin of the universe -- and the origin of time itself -- that is not complete bogus. But recently physicists have come up with one that works because it removes the problem of quantum gravity.
