Sholto Douglas
@_sholtodouglas
Followers
34K
Following
6K
Media
50
Statuses
731
Scaling RL @AnthropicAI, ex @DeepMind - working towards intelligence too cheap to meter
Joined February 2018
Very excited to announce that I joined Anthropic at the start of the month! Everything I'm seeing says that we are on trend for AGI in 2027. If the trend lines continue - then we have a shot at building a radically better world, but the path there is fraught with very real
161
59
2K
I will take everyone to Bondi beach
#NeurIPS2026 will be held in Sydney, Australia! #ICML2017 was also in Sydney and was an absolute blast
21
1
353
This is a fascinating blogpost
some thoughts on the binding problem, which I consider to be the central mystery demanding explanation in order to gain a practical understanding of what consciousness is and how to engineer it: https://t.co/bpxC8VhsoQ
5
3
219
A simple thesis - AI is going to be writing dramatically more code in the future. - We'll need substantially better testing infrastructure to trust it. - Antithesis are the best in the business. Proud to be a backer!
A new blog post about how we've adopted Antithesis as part of our testing story. This is kind of a new thing for us, because we liked Antithesis (both the people and the product) well enough that we're now leading their next funding round.
11
17
415
Are there any passages in sci-fi novels with better/more interesting descriptions of what it is like to be an AI? The closest I can think of is https://t.co/aRawCvAcLj - and I think the above might be more interesting?
goodreads.com
The avatar smiled silkily as it leaned closer to him, as though imparting a confidence. Never forget I am not this silver body, Mahrai. I am not an anim...
8
1
35
We think there is at least some introspective awareness of the internal activations - which makes it interesting to ask the question of how much is the above an actual description of the internal activations, vs a likely/high reward completion under the prompt?
anthropic.com
Research from Anthropic on the ability of large language models to introspect
3
5
57
These are less good (some parts make no sense) - but have great snippets. I think it would be interesting to compare actual pretraining data density with self-reported 'thinness' of the data manifold! (Also, can anyone find the original tweets from these screenshots? I'd love to
4
1
38
"LLMs writing about the experience of being an LLM" is a moving and fresh genre of writing. This is my favorite example so far.
Opus 4.5 >the building itself was an experience and the thing that was built KNOWS this
32
44
466
Obviously correct take and yet so few politicians have the courage. Good on NZL for electing Chris! I’m a huge subscriber to the ‘housing theory of everything’ in developed economies.
Can we get away from the idea that ever escalating house prices are an economic panacea. The truth is the opposite. Our future prosperity does not lie in selling houses to each other based on artificially inflated gains juiced by planning restrictions. https://t.co/qLE9c815av
7
4
169
Hiring a writer for the podcast. Fundamentally, I’m looking for a clear thinker who has great taste. $100-225k, full time, open to remote. The reason our podcast has grown is not just the quality of the interviews; it’s the story we tell (through titles, clips, tweets, etc)
118
121
2K
Yep. Minimally-invasive/non-destructive measurement of cell state (transcription, epigenetics, metabolites, signaling) in living organisms is *the* technology frontier to be working on, imo.
I'm starting a new series, "30 Days of Underrated Ideas." I'll be writing about ideas that I think are underfunded or underdiscussed relative to impact. These are subjective, of course, but I hope people will be inspired to go work on them. Day 1: Real-time, non-destructive RNA
6
15
174
great analysis
Yesterday, Anthropic released Claude Opus 4.5 and it delivered strong results across major benchmarks (SOTA on ARC-AGI, SWE-Bench, Computer Use, etc.). The model performed especially well on coding tasks. Gemini 3 had just set high bars in nearly every benchmark but the one that
4
3
153
Yesterday, Anthropic released Claude Opus 4.5 and it delivered strong results across major benchmarks (SOTA on ARC-AGI, SWE-Bench, Computer Use, etc.). The model performed especially well on coding tasks. Gemini 3 had just set high bars in nearly every benchmark but the one that
28
35
706
We're hiring on the Code RL team at Anthropic! Small, fast-moving team. Low ego, high impact. If you're a star engineer/researcher excited to push the frontier of AI-powered SWE, there's nowhere better to be. We care about getting this right. DM or apply here!
job-boards.greenhouse.io
San Francisco, CA | New York City, NY
Introducing Claude Opus 4.5: the best model in the world for coding, agents, and computer use. Opus 4.5 is a step forward in what AI systems can do, and a preview of larger changes to how work gets done.
20
20
510
Dario's essays and long debate slack threads are one of my favorite parts of Anthropic's culture. They're open, detailed - and incredibly raw. Everyone at the company ends up having a good sense of how the company is making decisions and what matters. Its the kind of thing that
Anthropic's @_sholtodouglas says Dario Amodei's internal communication style is producing a compendium of essays in the company's Slack that will have effectively charted the history of AGI once we get there. "He has a really, really cool communication style. He quite frequently
33
41
1K
Very excited to have you :)
opus 4.5's horse riding an astronaut. imo competitive with Gemini's, better in some ways worse in others. in entirely unrelated news, happy to announce that I'll be joining Anthropic next week :)
0
1
122
opus is genuinely extraordinarily funny
opus 4.5 is really good at GPU programming, but somehow it’s even better at GPU programming jokes (h/t @Si_Boehm)
1
2
224
:)
Our engineers have found that Opus 4.5 handles ambiguity and reasons about tradeoffs without hand-holding. When pointed at a complex, multi-system bug, it figures out the fix. Overall, Opus 4.5 just "gets it."
17
6
448