DJ Strouse (@djstrouse) header
DJ Strouse (@djstrouse) profile picture

Research @AnthropicAI

San Francisco, CA
Joined August 2007
Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)Photo by DJ Strouse (@djstrouse)
.@Amazon Mechanical Turk really should’ve been called the Auto-Man Empire.
2
4
1
45
Cool resource if you want to learn more about the challenges of frontier async RL!
At Proximal, we believe that building our own research infrastructure is crucial to design the highest-quality training data Today, we are sharing more information about our internal post-training stack
3
23
1,590
Honest take up front: it’s not just smarter than before — it’s a better writer. And that’s worth stating plainly.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
7
1
86
4,760
DJ Strouse retweeted
Opus 5.5 is way, way, way better than Opus 5. Sorry about that model, please try this one.
322
313
170
11,099
1,105,217
also important news we fixed the writing
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
172
113
59
4,208
350,230
Shouldn’t a podcast about weightlifting with entrepreneurs be called Raising Stacks?
the legends at @soontechnology and I have been on a mission to find the most jacked AI researcher and we think that @LiamFedus, co-creator of ChatGPT and co-founder of @periodiclabs might be it. check out the latest episode of Swole as a Service where he totally framemogs me while discussing whether or not AI will replace scientists 00:21 Who is Liam Fedus? 01:21 Lying skull crushers & will AI replace scientists? 03:05 Barbell curls, Jayden’s fitness tips, & why build an AI scientist now? 05:17 Hammer curls & is the singularity imminent? 07:43 Sleds & is the permanent underclass meme real? 08:37 Chestmaxxer Challenge & why hasn’t AI cured all diseases?
11
963
nobody wants to be in a code-pendant relationship
3
1
16
2,973
Neat paper - a lot to learn from learned token routings!
New paper! Collaboration with @TianyuHe_ and Aditya Cowsik. Thread.🧵
7
1,084
The lengths they’ll go for SEO…
It’s a day late, but I guess now we know this year’s version of OpenAI upstaging Google I/O?
4
758
I was very lucky to work with Behnam and Ethan at Google, and the team culture they built there was really something special. This is an amazing opportunity if you’re on the job market!
Replying to @ethansdyer
@ethansdyer and I have started a new team at @AnthropicAI — and we’re hiring! Our team is organized around the north star goal of building an AI scientist: a system capable of solving the long-term reasoning challenges and core capabilities needed to push the scientific frontier. We're thinking across the full model stack. Right now, we're focused on improving models' ability to use computers — both as labs for long-horizon tasks and as a key bottleneck in scientific workflows. Staff Infrastructure Engineer: job-boards.greenhouse.io/ant… Senior Research Engineer: job-boards.greenhouse.io/ant…
1
1
17
2,404
It’s because they’re minors
Minecraft proves that abolishing child labour was a mistake. The children yearn for the mines.
10
785
Sorry, there was confusion about whether 4.1 is larger than 4.5
what the hell is gpt 4.1 are just collectively accepting gpt4 -> gpt4.5 -> gpt4.1 ????
2
4
1
52
9,078
Today I’m launching @reflection_ai with my friend and co-founder @real_ioannis. Our team pioneered major advances in RL and LLMs, including AlphaGo and Gemini. At Reflection, we're building superintelligent autonomous systems. Starting with autonomous coding.
172
208
97
1,935
501,331
This is as good of a time as any to mention: As of late last year, I’ve joined OpenAI and moved to SF to work on pushing the frontier of RL for reasoning!
Looking forward to The Return of the Reasoning Models
2
43
3,518
Looking forward to The Return of the Reasoning Models
Pretraining maximalists strike back.
1
1
1
14
5,289
A lot of grad students have asked me how they can best contribute to the field of AI when they are short on GPUs and making better evals is one thing I consistently point to.
21
18
5
453
43,341
Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n
25
386
77
1,913
468,636
best thread on o3 FrontierMath, from one of the creators
1/11 I’m genuinely impressed by OpenAI’s 25.2% Pass@1 performance on FrontierMath—this marks a major leap from prior results and arrives about a year ahead of my median expectations.
4
6
84
11,105
The only wall models are hitting is saturated evals - hopefully our MATH expansion pack will keep you busy for a few months…
Math reasoning benchmarks keep getting saturated… Excited to introduce HARD-Math: Human-Annotated Reasoning Dataset for Math. Consisting of 4,780 short answer problems, based on the AHSME, AMC, & AIME contests, HARD-Math still poses a challenge for frontier LLMs. Read on 🔎⏬
1
26
3,288
DJ Strouse retweeted
Fun interpretation of a candidate's final absolute percentage point lead (e.g. Biden +1 in PA) - its the probability that, when counting votes in a random order, the winner leads the entire time. Horse races are inevitable! en.wikipedia.org/wiki/Bertra…
Key point by @dhopkins1776 on the @FiveThirtyEight liveblog: Biden isn't "pulling ahead at the last second", these votes were all cast on/before Election Day. "The illusion of a dramatic horse race is simply a product of the order in which we count votes." fivethirtyeight.com/live-blo…
1
3
2
16