Martin Singh-Blom
@singhblom.com
Head of ML at Embark Studios. Working on RL for physics based animation, LLM applications in games, generative models, etc. I like it when ML makes what previously was impossible possible.
Ah, Patrick Stewart. So many, many moments to remember you for. So many people to whom your work left a lasting impact. So long.
What a piece of work is man
YouTube video by Brian Focht
youtu.be
Forever in awe of young people who stood up for democracy, relentlessly protesting and mobilising for years against a corrupt, authoritarian regime. #Serbia
This has been really fun to be part of. Seeing the components come together slowly one by one and seeing the gameplay guys pick it up and start to play with it has been awesome.
THE FINALS@the-finals.bsky.social · 2w agoDYNAMIC COMMENTARY | THE FINALS #TheFinals #ReachTheFinals youtu.be/eA9_Q6d2H-g?...
uh oh "We show that the formalised Lean proof does not correspond to the NL proof of blow-up of solutions to the Navier-Stokes equations."
Pedro Monkey Puppet Meme
ALT: Pedro Monkey Puppet Meme
static.klipy.com
arXiv math.AP Analysis of PDEs@mathap-bot.bsky.social · 4w agoAlexander Bastounis, Fabian Circelli, Anders C. Hansen: Navier-Stokes lost in translation: Why Lean verification of AI autoformalisation does not guarantee correct natural language proofs https://arxiv.org/abs/2610.08144 https://arxiv.org/pdf/2610.08144 https://arxiv.org/html/2610.08144
Thank you to everyone sharing in this moment! I’m so grateful to all my trainees & collaborators… dlab.stanford.edu/group-members (pic: Peter and I together in 2017) Reading for all those asking: for technical details: dlab.stanford.edu for everyone: deisseroth.org/projections
This week's #PaperILike is "Latent Programming Horizons in Coding Agents" (Silva et al., 2026). Now seems like a good time to better understand what coding agents are doing. Here's a nice example of the kind of analysis one can do. PDF: arxiv.org/abs/2607.05188
Latent Programming Horizons in Coding Agents
A coding agent solving a software-engineering task spends dozens of steps reasoning, editing code, and running tests, yet little is known about what the underlying language model internally represents...
arxiv.org
I see that OpenAI's messaging of "if you give us all the money on Earth we will annoy some mathematicians" is being somewhat overshadowed by Anthropic's messaging of "if you give us all the money on Earth we will kill everyone". I'm slightly worried about what a compromise position might look like.
it’s official: OpenAI claims they solved Navier-Stokes 10,000 concurrent agents coordinated by an internal model (that’s currently still in training) worked for 88 hours openai.com/index/navier...
mr. TIM@timkellogg.me · 2mo agoNavier-Stokes — huge drama around this truth is more like: 1. critical insight produced by human mathematicians 2. massive acceleration from AI 3. Ant employee worked with NYU prof to carry out the research program statement by NYU prof: cims.nyu.edu/~tristanb/st...
The OpenAI Navier-Stokes drama is good drama. Did OpenAI train on conversations where mathematicians were working on Clay prize proofs? What should that mean for credit assignment ?https://news.ycombinator.com/item?id=49605915
Navier-Stokes – Tristan Buckmaster [pdf] | Hacker News
news.ycombinator.com
This was really nice. Pair programming during the planning phase with a colleague and with the LLM seems like something that should be done much more. Align on architecture before the clanker starts clanking.
Kyle@kylestratis.com · 2mo agoAt @martinfowler.com's site, a great piece by Rachel Laycock on how agentic development has changed the code review process and how we can get the things we used code review for (hint: not finding bugs) by rethinking our process
GLM-5.3-Flash rips the doors off and is under a full MIT license we have, and I mean this literally, a Smart Enough model for almost any task a normal person and many AI perverts can envision, and it is free to use at the same time, inference hardware will only get more accessible to you
absolute horses@jcoglan.com · 2mo agocould a handful of billionaires owning the means of production of all knowledge/creative work revive socialism
Trying to maximize the science time with the kids this evening. The eclipse had them super excited.
There's growing evidence the surge into Ceuta (🇪🇸PM Sanchez has called it an "attack") has, like in 2021, been orchestrated by the Moroccan government in retaliation for Sanchez visiting Algeria Yet it's Spain being criticised by European politicians, not Morocco. So much for solidarity.
About 60,000 migrants arrive in Ceuta in 24 hours, Spanish territory's leader says
The estimated number of arrivals is more than 70% of the population of the city, which is about 83,600.
bbc.com
AI safety folks should be way more worried about concentration of power than the proliferation of powerful models
It's wayyyyy better for society if we have high-quality open weight models where you choose to run inference. And if China's the one to do it, so be it.
Regime Accountant, CPA ☠️🎢@regimecpa.bsky.social · 3mo agoI don’t care if China beats us on AI with open weight models. I’ll happily buy the higher quality model for cheap, that’s a consumer win.
Here's something I post from time to time. My answer to a reader who asked me: what could journalists do NOW to break with some of their more corrosive habits.
Unfortunately, a stunned Paris is how they all got into this whole mess to begin with
Pop Crave@popcrave.com · 4mo ago‘The Odyssey’ women Zendaya, Anne Hathaway, Lupita Nyong'o, and Charlize Theron stun in Paris.
Human intelligence is fundamentally a collective intelligence. We solve complex problems by participating in a vast cultural network that builds upon ideas across generations. I believe the strongest AI systems will become a collective intelligence, too.
Sakana AI@sakanaai.bsky.social · 5mo agoIntroducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API. Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls. Try it: sakana.ai/fugu 🐡
How do brains plan actions towards goals? To get at this question we studied mice navigating complex mazes as goals changed on every trial 🧵 Work with @thomasakam.bsky.social @behrenstimb.bsky.social @kristorpjensen.bsky.social now on BioRxiv: www.biorxiv.org/content/10.6...
15 best movies quickly: In the mood for love The seven samurai Amelie No country for old men Inglorious Basterds Mother of Mine Raise the red lantern Bridge over river Kwai The Martian Good Will Hunting Die Hard Crouching Tiger, Hidden Dragon Life of Brian La Vita e Bella Death of Stalin
Mark Copelovitch@mcopelov.bsky.social · 6mo ago15 best movies; don't think about it too hard! (not in order) Lives of Others Inglourious Basterds Shawshank Redemption Best in Show LOTR: RoTK Princess Bride Godfather Casablanca Amadeus Schindler's List Blazing Saddles L.A. Confidential Big Lebowski Silence of the Lambs Raiders of the Lost Ark
Generational tweet that might even spawn a school of architecture.
It is with great sadness that we announce J. Craig Venter has died. 🧪🧬 www.jcvi.org/media-center...
J. Craig Venter, genomics pioneer and founder of JCVI and Diploid Genomics, Inc., dies at 79
La Jolla, California—April 29, 2026—The J. Craig Venter Institute (JCVI) announced that J. Craig Venter, Ph.D., the Institute’s founder, board...
jcvi.org
This week's #PaperILike is "AssistanceZero: Scalably Solving Assistance Games" (Laidlaw et al., ICML 2025). AlphaZero-like combo of learning & planning for assistance games, where robot & human share reward fn that robot doesn't know. + Minecraft! PDF: arxiv.org/abs/2504.07091
AssistanceZero: Scalably Solving Assistance Games
Assistance games are a promising alternative to reinforcement learning from human feedback (RLHF) for training AI assistants. Assistance games resolve key drawbacks of RLHF, such as incentives for dec...
arxiv.org
have been researching bestiaries for a project and found this one of a griffin about to tear off some poor soul's entire head very relatable
Global solar additions reached record 511 GW [613 GWp with 1.2 Wp/W,ac ratio] in 2025, says IRENA www.pv-magazine.com/2026/04/02/g... - and much more ahead: doi.org/10.1016/j.en... ; doi.org/10.1016/j.rs...
Global solar additions reached record 511 GW in 2025, says IRENA
Latest figures from the International Renewable Energy Agency find solar contributed the majority of a record 692 GW of renewables capacity added worldwide last year.
pv-magazine.com