founding eng @ https://nitter.gq/t.co/VvOl7DRKx4 - previously @ https://nitter.gq/t.co/8m8weVc4gO. Also studying 中文 and trying my best to travel.
Joined June 2010
- Tweets6.3K
- Following2K
- Followers299
- Likes50.3K
Peter retweeted
Everyone and everything is narrowed in on Democrats winning about 240~ seats in the House
Unless your ConET and still under the delusionals that Republicans will keep the House lmaooo
Imagine getting forced to use GLM for day to day work. I also have serious doubts about it improving "lead time" by 70%.
Clearly just bean counters running that engineering org
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
One of the my deepest joys with this new era of engineering is kicking an agent off (from my phone nonetheless) with an idea and a general mission and letting the model hyper-optimise things.
Everyone always talking about how to get PR CI passing faster, I honestly think this is kind of straightforward and easy to hillclimb. But the real bottleneck, for us at least, is deploying. When you're shipping 50 PRs a day how can you realistically get changes into prod in a safe state? We have (what I think) are excellent integration tests which run as part of CI. We then have a canary on which Devin live tests our changes, then we promote, but this inevitably ends up with a backlog as well. main -> staging -> prod pretty much the same issue.
Who is working on this?
Peter retweeted
now wishing i took more raw. i’m in love
So I recently badgered my cinematographer @steveyedlin into making my ideal iPhone still camera app. It shoots RAW, bypasses the iPhone's look processing and instead uses the same image processing we use on our movies. apps.apple.com/app/apple-sto…
I don’t think I could have timed this any worse if I tried. Literally nothing I can do but sit and wait, can’t even arrange anything else. Will always buy a refundable ticket in future.
Not sure what sort of things people are working on with open models, but this benchmark exactly matches my experience. As much as I'd love to use them, they're not even close to where I would them need to be for any sort of serious work.
Introducing FrontierCode 2.0: our new coding benchmark that measures a model’s ability to go beyond the spec.
Most coding evals measure a model’s ability to write code that meets a spec. With FrontierCode 1.0, we asked a new question: would you actually merge the PR?
FrontierCode 2.0 is comprised of more diverse tasks, featuring larger scopes and higher mergeability standards. We are now asking: not only would you merge this PR, but should you?
One key feature of FrontierCode 2.0 is underspecified prompts. Rather than including every engineering decision, prompts are deliberately ambiguous, giving the models room to demonstrate their judgment.
The result is a benchmark that is far from saturated, with the top model scoring just 45%.
Anyone had recent experience in trying to get an app through the iOS app review process? It's been nearly three weeks now with absolutely zero updates, totally opaque experience, tried contacting support multiple times now to no avail.
Maybe this model will actually be useful for knowledge work
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
You doubted the Zuck?
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
Amazing research, but I find it difficult to comprehend both the content itself as well as the level of intelligence needed for this sort of work. Constantly in awe of what people are capable of.
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Not quite sure how it’s possible @LangChain thought it was a good idea to ship the Langsmith UI without OTEL support?