I’m a software engineer building high-performance kernels and compilers at Anthropic! Previously at Facebook/Meta (PyTorch, HHVM, ReDex)
Joined December 2022
- Tweets432
- Following402
- Followers2.7K
- Likes1.2K
Bert Maher retweeted
I am a proud American, and I’m proud to work at Anthropic. Labeling Claude a supply chain risk will only harm America’s lead in AI. It’s not un-American to support AI guardrails that protect our civil liberties.
A statement on the comments from Secretary of War Pete Hegseth.
anthropic.com/news/statement…
Bert Maher retweeted
One of the most common flaws of math textbooks is that they present only the logic, without the intuition. They give you the later, cleaned up version of the idea, which hides the way it was discovered.
Bert Maher retweeted
Introducing Claude Sonnet 4.5—the best coding model in the world.
It's the strongest model for building complex agents. It's the best model at using computers. And it shows substantial gains on tests of reasoning and math.
ALT A benchmark table for Claude Sonnet 4.5 showing leading performance across many domains, including agentic coding, computer use, mathematics, graduate-level reasoning, and financial analysis.
Bert Maher retweeted
The most important skill for a researcher is not technical ability. It's taste. The ability to identify interesting and tractable problems, and recognize important ideas when they show up.
This can't be taught directly. It's cultivated through curiosity and broad reading.
Bert Maher retweeted
I was talking to @tensorbert about why so many Chinese talent on AI especially on the engineering side. It’s because of a deep engineering / implementation/ optimization culture in the STEM education system that favors incrementally a lot more than innovation. DeepSeek was an outlier (its founder was heavily influenced by Renaissance)
Bert Maher retweeted
getting a phd is mostly about developing a system for maximizing the time you spend in flow state, over several years
Bert Maher retweeted
This is hilarious, the random activation function I made up for this talk because it demonstrated fusion well is actually useful 😂
Bert Maher retweeted
Prompted by investigations by Boyue Li and @tensorbert, I reduced the kernel launch latency of the open-source compiler github.com/openai/triton by a factor of 7 yesterday, from 105μs to 15μs.
Admiral Grace Hopper explains the importance of microseconds in youtu.be/9eyFDBPk4Yw
Bert Maher retweeted
Happy New Years Everybody! 🥳
One year ago today, I had:
- Never trained any model
- Did not know the first thing about AI
- Never worked in Tech
- Had 8 followers on twitter? (probably)
One year later and here I am!
Met many of my heroes and legends in tech and worked with hundreds of amazingly talented and influential people, went on some amazing trips to SF and NeurIPS and met so many of you, worked at three tech co's, built Nous Research with @theemozilla @karan4d and Shivani, made several of the worlds SOTA OS models, became a pro data engineer, got cited in at least 14 papers, and so much more!
Imagine where we'll be in a year from now, I'm optimistic I'll be able to continue contributing to the Open Source future!
Bert Maher retweeted
Ok I’m at NeurIPs to talk about our in person competition workshop on LLM efficiency on Friday Dec 15 between 1:30 - 4:30 pm CST
Competitors had to fine tune 1 LLM in 1 day on 1 GPU and the reception was incredible. This was one of the most popular ML competitions of the year.
If you’re interested in the foundations of eval and fine-tuning come learn from the best @jeremyphoward @rasbt and most importantly which techniques actually worked from the winners themselves!
Stacked lineup ->
neurips.cc/virtual/2023/comp…
Bert Maher retweeted
@MistralAI model is hot: with mixture-of-experts, like GPT-4!
It promises faster speed and lower cost than model of the same size and quality
Fascinatingly, the speed-up is unevenly distributed: running on a laptop or the biggest GPU server benefit the most. Here’s why 🧵 1/7
Bert Maher retweeted
Language is an imperfect, incomplete, and low-bandwidth serialization protocol for the internal data structures we call thoughts.
Bert Maher retweeted
I got Triton working on (local) Godbolt instances, including Python <-> SASS line correspondence. Pretty nifty.
Still got some infra issues to figure out, meanwhile PR is here if anyone wants to try it: github.com/compiler-explorer…
Bert Maher retweeted
In past companies, contributing to LLVM, there was a huge quality-gap between internal LLVM-code and LLVM public one.
As an engineer, OSS code reviews were protecting me against skewed short-term internal incentives and manager pushing for "just land it as-is and improve later".
Bert Maher retweeted
There's a huge opportunity to use ML to replace hand-written heuristics in compilers.
Want to try? Check out this @kaggle
contest organized by Ashley Chow, Bryan Perozzi, HCL-Jevster, inversion, Mangpo Phothilimthana, and Sami Abu-El-Haija!
kaggle.com/competitions/pred…
