CTO at https://xitter.cf/t.co/hvD3ZW7gNY | Passionate about Software Engineering. Search/AI practitioner #search #relevancy #ml Always Learning. Previously @Vimeo
London (NW) / New York
Joined September 2008
- Tweets2.5K
- Following1.9K
- Followers986
- Likes1.3K
Chris Simpson retweeted
We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them.
Chris Simpson retweeted
Today, we’re sharing that a general-purpose internal @openai model achieved a breakthrough on one of the best-known combinatorial geometry problems. Less than 1 year ago frontier AI models were at IMO gold-level performance. I expect this pace of progress to continue.
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
I read this piece by @seldo "The end of fine-tuning" and it resonated with me.
As many move away from fine-tuning to the iterating the harness I think retrieval expertise is more important than ever: chrissimpson.co.uk/the-harne…
I've been writing out a bunch of my thoughts recently. Today I checked out the new xAI algorithm on github, and posted my thoughts on this from viewpoint of a career long search guy: chrissimpson.co.uk/xai-algor…
Buried in the Elastic BBQ vs TurboQuant post: on isotropic data, Hadamard's dot-product gain is pure padding. 1.15x = √(1024/768) exactly. Full dense d=768 shows zero gain. On anisotropic data, a block-diagonal preconditioner gets most of it. Interesting. elastic.co/search-labs/blog/…
Chris Simpson retweeted
We set out to build a better retriever, so we looked for the hardest IR benchmarks.
For each, we asked how much headroom remained by running oracle reranking with a frontier LLM. Most had little room left!
So we built OBLIQ-Bench to study much harder search queries than before.
Devastated to not be in Charlottesville as usual for #haystackconf. It really is the greatest gathering of the minds in the search/information retrieval space. Great community - sad to be missing it this year. #fomo @softwaredoug @renekrie @_tallison @treygrainger @o19s @dep4b
Chris Simpson retweeted
As promised, here's a recording of my 30-min keynote and the subsequent Q&A for the inaugural late interaction retrieval (LIR) workshop, cc @bclavie @antoine_chaffin.
The talk is admittedly advanced, as it's directed at an expert IR community. But hopefully still broadly useful!
Lots of people interested in the late Interaction workshop, listening to @lateinteraction's keynote!
Awesome talks tonight, and some truly fascinating discussion after. Great work @HornetDev
#londonsearchweek
Definitely revisiting this paper in the context of this evenings discussions: arxiv.org/pdf/2510.21440 #udcg
“Innovation in the Age of Artificial Intelligence” Keynote by Rick Hamilton from Focussed Ultrasound Foundation @FUSFoundation at #haystackconf
excited to be back for another #haystackconf always where some of the best conversations of the year happen. Can’t wait to catch up with everyone!
Chris Simpson retweeted
Yup!
Search is more relevant than ever—it’s the natural abstraction for connecting AI with evolving reality.
As AI systems move from static reasoning to acting in the real world, search becomes the bridge between what models know and what they need to discover.
Chris Simpson retweeted
Let’s fucking goo!! DeepSeek R1 1.5B running FULLY LOCALLY in your browser at 60 tok/ sec powered by WebGPU🔥
Intelligence truly is too cheap to meter! ⚡️
Chris Simpson retweeted
🚀 DeepSeek-R1 is here!
⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!
🌐 Website & API are live now! Try DeepThink at chat.deepseek.com today!
🐋 1/n
🤯
This "Aha moment" in the DeepSeek-R1 paper is huge:
Pure reinforcement learning (RL) enables an LLM to automatically learn to think and reflect.
This challenges the prior belief that replicating OpenAI's o1 reasoning models requires extensive CoT data. It turns out you just need to give it the right incentives.
We are so back in the AlphaGo excitement era: by playing countless Go games and maximizing the reward function (winning the game) using pure RL, AlphaGo beat the best human players.
Now we are entering the LLM RL era.
2025 could be the year of RL.
Love this @Film_London. A great glimpse in to the past. I believe this was captured close to the current day site of my home. Very different surroundings in 2025. Thanks for sharing!
As we begin the new year, Horses New Year’s Feast is our #FilmoftheWeek, a short film showing the animals at the Home Of Rest For Horses during the annual New Year feast at a charity event on Westcroft Farm, Cricklewood in 1913.
buff.ly/3PwMAuo
@BFI
Chris Simpson retweeted
Retrieval is the cornerstone of reliable AI, so it's full circle. From @sequoia sequoiacap.com/article/ai-in…
