Alexis here — welcome, and thank you for stopping by! Below you’ll find some stuff I’m really proud of and want to share with you. Have a look around, see what tickles your fancy and follow it. Would love to hear what you think. Let me know!
Principal Product Engineer at FELS Family Office GmbH · experimenter at heart · second-order cybernetic stewart · based between Berlin, New York & Marseille.
Machine learning is the science of credit assignment. My new survey (also under arXiv:2212.11279) credits the pioneers of deep learning and modern AI (supplementing my award-winning 2015 deep learning survey): t.co/MfmqhEh8MA P.S. Happy Holidays! t.co/or6oxAOXgS
@SchmidhuberAI Thank you for sharing Jürgen! (Danke dir für's Teilen.)
Looks like my beloved "Social graph of cybernetics" by Hugh Dubberly and Paul Pangaro needs an update.
"[...] the agents created millions of short URLs to affect another system." and immediately, pictures of early industrial revolution-era polluted rivers with dead fish floating popped into my mind.
I believe Buckminster Fuller's definition of "nature" included human artifacts like cities, airplanes, art.
Could we include networks and AI?
And if we did: How appropriate do you think it is to re-use vocabulary from existing environmental protection policies in this context?
@AmmannNora PS: You probably have seen this but here's the report I was thinking of
In another of my favorite pop-sci books called "The Talent Code" I learned about myelin, which is basically the mortar between Barbara Oakley's bricks.
For a deep-dive, there's a really good chapter on the myelin sheath here: ncbi.nlm.nih.gov/books/NBK27954/ which is part of "Basic Neurochemistry" (
"DARPA has just kicked off Exponentiating Mathematics (expMath), a three-year program aimed at radically accelerating pure math research by developing AI that can propose and prove abstractions. According to the official program brief, expMath will bring together teams focused on auto-decomposition; automatically breaking complex conjectures into reusable lemmas; and auto(in)formalization, bridging the gap between human-readable mathematics and rigorously checked proofs in languages like Lean."
I am incredibly grateful to my co-authors Sebastian Risi, Yujin Tang, and Risto Miikkulainen for making this happen. Neuroevolution is a subject very dear to my heart. It is the field that convinced me that nature has already figured out how to build intelligence: through evolution, collective behavior, and adaptation under constraints.
This idea, that intelligence emerges from evolution operating under constraints rather than unlimited resources, is what eventually led me to founding Sakana AI here in Japan.
The concepts in this book about open-ended creativity and self-organizing systems are exactly what we build at @SakanaAILabs. Our name and logo are inspired by schools of fish moving together, adapting as one. It is the core philosophy of what we build. This book captures the theoretical foundations of that belief.
"[...] the agents created millions of short URLs to affect another system." and immediately, pictures of early industrial revolution-era polluted rivers with dead fish floating popped into my mind.
I believe Buckminster Fuller's definition of "nature" included human artifacts like cities, airplanes, art.
Could we include networks and AI?
And if we did: How appropriate do you think it is to re-use vocabulary from existing environmental protection policies in this context?
even within SF, i don't think we're AGI-pilled enough. it doesn't make sense for labs to stop. rn, we're only blocked by infra, not intellect
first code, now life sciences/eng, then robotics. they'll all become accessible, with in-house facilities being the best RL envs ever.
@danielrupawalla Agreed. For me living in Berlin, AI-twitter feels like a life-line into the future.
I think I understand what you mean with RL in-house facilities but could you explain? How much would that mean giving top-talent physical/psychological sensored exo-skeletons?
In 1991, James G. March showed that a learning system can become less intelligent by learning too quickly.
Exploitation produces clearer and more immediate rewards than exploration, so adaptive organizations gradually favor it. However, he also found that rapid convergence destroys useful disagreement.
Slow learners and newcomers improve collective knowledge because they preserve alternatives that the shared consensus has not yet absorbed.
@richardcsuwandi Interesting! Reminds me of Barbara Oakley's brick wall metaphor in "A mind for numbers."
AIDE² optimized our research agent's harness against gemini 3 flash. Those gains carry over to fable 5 and gpt 5.6 sol.
That's just one of the new results in our arXiv paper on AIDE², an RSI system where AI research agents improve their own research efficiency.
@DhruvSrikanth Hey Dhruv! I strongly believe that your AIDE² "autoresearch into autoresearch" will be understood as one of the most pivotal papers of this decade. Congratulations.
Related: Have you watched Sean Melleck's recent "The problem is the problem" discussion yet?
@wellecks' "The Problem is the Problem" is a must-watch for us working in autonomous research and discovery loop engineering.
Thus far, I had only considered that my discovery loops would always target solutions to problems.
I had the same core-assumption about @zhengyaojiang 's amazing AIDE² research: We run autoresearch on autoresearch to find optimal solutions to decided-upon problems, including our research methods themselves.
But what if... we use autoresearch facilities to discover juicy, interesting and important problems as well?
Mind. Blown.
Oh, and with perfect timing, AIDE²'s paper finally dropped two days ago, too. It's of course the parallel must-read of the year:
Hey there! Alexis here. I'm the author of Mosey and director of this little film here. As you may (or not) have guessed, the film was created with AI. If you're (like me) skeptical at this point, please read on.
I want to show you exactly HOW I created it with AI so that you have a chance deciding for yourself if it's "slop or not."
Why is this important?
Well, it's important to me. I'm personally getting more and more sensitive and frustrated with content because a.) I can't tell the difference between human-made and AI-generated anymore and b.)
I can't tell if the person behind the artifact (blog-post, twitter message, promo video like this one) actually even cared about making it.
In fact, this page above is a kind of "prototype" of how I wish digital, AI-created content emerged. Was it a sloppy, careless one-shot of "Hey Claude. Make a cool video. Make no mistakes" or did the person behind the keyboard (in this case myself) actually put in the legwork?
I believe that giving you full access to my honest, imperfect, human side of the "creation" (warts and all), you and I get a fair chance to meet at eye-level on all of this.
How does the page work?
The first thing you'll see is the final cut of my little film. It's short and I highly recommend watching it first (if you haven't yet) so that you know what the rest of the page is about.
Then you'll see the full transcript of the "conversation" that I had with Claude Code to create the film withe the goal to put it on the frontpage to explain what this whole Mosey . app is about and, to a lesser degree, why someone should get it.
On the left side, you see my actual text inputs (or prompts although I would not call them that) which I sent to Claude Code. On the right side you'll see the assets (pictures, video clips and audio files) that Claude Code created in response.
When there was a intermediate deliverable (called "Cuts") where I felt like we had reached a milestone, you'll see them in row on their own. They're small highlights on the way to the final cut.
What I've left out: I've left out Claude Code's textual parts of the responses because they were mostly blathery fill-in wordsalat padding around the assets (which is what I cared for the most)
What I've left IN: In this version I'm leaving in ALL of it. ESPECIALLY the warts. You'll see my cursing and insulting and threatening Claude Code in ALL CAPS. Basically, you'll see me at my worst. This is incredibly embarassing to share but I feel like it's the most honest, too. It's how this sausage was made.
Your task now is: watch the short film itself, then peruse the full, pants-down "making of" of it.
Thank you and with best wishes,
Alexis
PS: Hey @isidentical I love fal and in message #107 you'll see me switching from elevenlabs to your company. What do you think of all of this?
Working with agents often reminds me of that classic German Loriot sketch. He simply wanted to straighten a crooked picture and ended up destroying everything in the process.
fyi: The spoken words are not important to get the joke.
Ever wonder "Why is my apartment so warm all the time?"
Turns out: my router has been quietly moonlighting 24/7 as a small space heater at 36℃/97℉.
And my fridge at 27℃/82℉. You can see how the hot air flows up, and makes sure my ceiling is nice and, err..., crispy.
I'm so glad I got this thermal phone adapter because I would NEVER even have considered these two as candidates. (I was 100% convinced the problem came from a broken floor-heating unit and/or wall-emission.)
Highly recommended!
PS: I now turn off the router at night. And yes, ambient temps have finally dropped to expected levels.
Hey @Topdon_Official 👋🏼 I'm using your TC002C Duo here. Works fantastic.
Also: I had a technical issue yesterday and Mr. Morad from your Germany office helped me resolve the issue in minutes! Really appreciated his support.
@SpringStreetNYC Offline-first is a fair wish. Geb is privacy-first, so the draft stays yours. Free: 50 images and 15 videos. Basic is $2 for 200 images + 20 videos, and Key of Life lets you clear a memory. t.co/qQZw3Ixj01
The kind of future I want currently costs $470 with a $5 monthly subscription:
Oura ring, $270 + $5/month
Redmi Buds 6 Pro, $50
Comulytic Note Pro, $150
I like that none of them are connected to the internet. (I miss having a local AI though.)
After 20 years of touching glass, all I want is to touch some grass. Spend more time with my friends and family. Spend more time connected, less time wired.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
At this point, I think Anthropic is intentionally hyping AI fear. So that government can intervene and regulate and kill the competition for Anthropic.
Remember Anthropic already hate OpenSource LLMs t.co/Dyxx4wAjf4
@Dev__Vishwajeet The side-effect of hyping AI fear by shipping these extreme narratives is that it disrupts and destabilizes individuals and in consequence communities.
How is this different from state-sponsored propaganda?
pssst people haven't realised that OpenAI literally just rolled out a 15gb RAM 9 core VM in ChatGPT to ALL their paid customers with ChatGPT Work.
Non technical people now have remote machines available for their agents with enough resources to do most economically valuable work in that machine.
It has its own browser, can install packages and do a whole lot of things.
And this is not even the reason why Codex is growing so much right now! We'll see an ever sharper rise when more people realise what they can access now.
no need for a dedicated server / vps for most people now.
@maxjendrall Wait, this is super interesting. So it's remote, right? Can't be local, obviously.
Basically every ChatGPT Work customer's assistant gets its own laptop?
DJing for The Stone Roses last ever Irish show in Belfast. I played this gig with a broken wrist 🚑 Sound system ruled. Last 2 minutes of set⬇️
Kurtis Blow // Liquid Liquid // The Clash // Can // Talking Heads // Incredible Bongo Band // Ian Dury // Information Society // Hashim t.co/YQtYOP1zlD
AI * web3 hack week coming to Lisbon 🇵🇹. If you're a hacker and think AI should be open source and decentralised, this one is for you: t.co/OPZ1ayHJC7t.co/SWKQnPi03o
@charlesworthjc Joe! Using your post here as an excuse to just say hi.
How are you doing? Where are you at these days? And how's life with your lovely family?
If you're waking up this morning grateful for another day...
You've already won, friend.
My life hack for this: a little golf counter, the clicker kind. I got a fancy one in silver and I carry it with me. When people say "count your blessings" I took that literally.
A few times a day I stop and ask what I could be thankful for, then click. Nice coffee, click, thank you. Easily 20-30 a day, from the mundane to the very high-spirited.
Today so far: my lovely mechanical keyboard, the guy who just delivered groceries, green tea.
Claude shipped Fable 5.1. The feed will treat that as the model you should use now.
I looked at my own usage. The list-price meter was about $56k in 30 days. Memberships made it about $800. That is the gap the labs are selling: sticker vs seat.
I still use cheap or free models for rewrites and first drafts. OpenCode plus DeepSeek covers a lot of that. I save the expensive slow model for a long agent run, the same way I would not hire a specialist for a sore throat.
If the job is simple, use the simple tool.
@mehdizare Does this mean you actually have a $55k bill for your private work with Fable?
crazy that X open-sourced their algo. i spent last night studying it and here are the main takeaways for viral posts:
1. Early, strong in-network engagement. Your followers are your first gatekeepers. Rapid likes, retweets, and replies in the first minutes signal quality. This helps you break out of your followers (also means you should definitely reply to comments)
2. High-quality engagement signals. The algorithm prioritizes high dwell time, sustained video watch time, and deep thread reads. Avoid anything that leads to quick bounces. Your content needs to be genuinely satisfying.
3. Good social proximity. Who engages with you matters. Tweets from authors close to you in RealGraph, or liked by your close contacts, get a significant boost. Strong local social proof is a powerful amplifier. Interactions from highly connected followers definitely help as well
4. Topic and cluster relevance at scale. For out-of-network reach, your content needs to align with large, active SimClusters (broad topics/communities). Niche tags won't get you broad distribution. This is where researching trending topics becomes critical. Original content also does MUCH better than retweets or replies
5. Freshness and sustained momentum. Recent tweets with ongoing engagement are heavily favored. The algorithm tracks "last engagement since creation" - it wants content that keeps generating interest over time, not just a one-off spike. Consistency is key.
Oh, this is really interesting. Glad I just found it again in my bookmarks.
Though: this is from September 2025, so it's about a year old now.
Have you kept monitoring it since? Would love to know which of these five still hold exactly as written, and which ones have quietly moved on you.
@kshvyaps Been doing this for a year. Actual thought per reply, no slop, because slop insults the reader's intelligence.
Result: basically no followers. And I'm paying for the privilege. Being conscientious as a new account works against you, apparently. Not cool. Really. NOT COOL.
@kshvyaps 100 replies a day, otherwise your reputation gets lowered. That's insane.So my actual question: how does everyone do this? Other than pointing an AI at it and cranking out slop.And if that IS how it's done, then is all of it just slop?
Please fail more in life. In college I lost 41 hackathons, failed 15 interviews and my startup was an epic flop.
During that time I pitched to VPs at Rakuten, scientists from NTRO, engineers from Microsoft and Google and even the government.
And today I work out of New York City with one of the finest CEOs in the world on AI projects that are changing people's lives every day.
I learnt more from my losses than my wins. My failures have given me more than my wins ever have.
So please everyone I beg of you. Fail more often. The view at the top comes from falling a lot of times and getting back up.
Love the specificity of the list. And yes, I learn more from losses than wins too. I've even proactively set myself up to fail just to learn faster.
The part I'd add: at some point it has to be more than just failing. You have to fail well, which really means you had to learn to learn first.
And that usually rests on some kind of post-failure safety net. Not necessarily rich parents. Mine was a curiosity safety net: lower middle class, but I have always loved learning early and kept getting food pellets from the universe for it. That's what made failing survivable for me.
So my ask to you is: looking back at those 41 hackathons and 15 interviews, what were your hidden enablers? Could be socioeconomic, could be intellectual, could be a person. Would love the nuance, because I think that's the part people can't see from the outside.
German really does have a word for everything. t.co/YlWzUVjG6u
@GermanSimply_ Reminds me: the new Deutsche Bahn CEO recently calling their mess "organisierte Verantwortungslosigkeit." Two words, granted. But packs a punch alright.
Was on the bike thinking about the whole mathematics/OpenAI thing.
Alan Watts: if we approached music the way we approach capitalism, the conductor who reaches the last note first would be the most successful. And the musician writing the shortest songs would win. Obviously silly. But is the journey is as important as the result?
Someone in a thread said a proof is only really valuable if you can explain it. I agree. Otherwise it's slop. Math slop?
But here's the conundrum: what if it works anyway? Throw brute force, money and machines at it, and out comes the actual record-shattering thing. Nobody can explain how.
Someone told me part of the payoff here would be physical prediction models weather forecasting for becoming orders of magnitude better. So you could argue: skip the drama, the back-and-forth, unlock the benefit for everyone, and let only the people whose job it is care about how we got there.
Basically, what if the slop solves world hunger?
And at the same time: the human version of this is years of grinding. You invest your time, your relationships, long stretches without reward. Even the million-dollar prize is nominal once you count opportunity cost.
These three, four, five researchers could have slid up the bell curve into a fund job and made that money anyway.
The most heartwarming part of the post that kicked all this off: the guy mostly said he shouldn't be credited, that it should go to his dude over there. Incredible sportsmanship at very high stakes.
Fast-forward 30 years. Are those researchers financially safe? Taken care of?
Risking your private time to hack on a problem for years IS risk. Spinning up 10.000 agents as part of a marketing budget is not at all risky.
Anyways, all of this is making me think harder about my own work and whether the techniques and methods I apply to my research are actually healthy ones.
Before the AI boom is over, we in science have a once-in-a-lifetime opportunity to get tech companies to blow millions on solving scientific problems they couldn’t care less about. Let’s not waste it. Hey OpenAI, rumor has it Anthropic is on the verge of cracking P = NP.
@pmddomingos "Couldn't care less about" is the part that sticks with me. Turns out solving a high-status problem doesn't absolve you of having to care about it. I believe that's new.
This is why it's a good thing that American labs are in the lead. If Chinese labs were, their employees wouldn't be able to warn us about threats. But Chinese labs will encounter the same threats within a year, in complete silence. t.co/Bg5Vakbl4W
@paulg Hadn't thought about it from that angle. Maybe the underrated advantage of free speech isn't just the speech part but having a culture for civil conflict. You can point at something and say "this looks broken" and that can now be the start to find a solution or agreement.
Can’t make this up…the MacBook is using its webcam to look at its screen in a mirror to improve AMD Radeon chip support in Omarchy. t.co/pw5Yu0JVJ7
@jpschroeder Love this. The more AI writes my software, the more I wind up with IRL hacks like this one, too. This is so cool, it's also funny, and cheeky and just "good stuff" in my book :)
Researchers spent 500 hours and $70,000 and found LG TVs logging plain text transcripts in standby, mapping every device in the house, and feeding it to LG's ad arm
Unplug the internet and it saves the files until you plug back in.
LG says its TVs don't record ambient conversations.
Oh! Saw this on HN a few days ago and it didn't land (probably got distracted by Navier-Stokes). The video did land though. Had to watch twice to get that the audio playing back was what the TV itself had recorded.
Would love to see the process (including the r00ting!), the on-device infra/setup and what gets sent where exactly.
BREAKING: OpenAI might have stolen another major proof.
In a detailed Mastodon post, which I report in full in the comments, Andreas Thom presents several pieces of evidence suggesting that OpenAI may have trained Astra on conversations in which he and Gábor Kun were working on Gromov’s soficity conjecture, one of the ten problems OpenAI later announced Astra had solved.
I know Andreas. We met several times early in our careers. He is an exceptional mathematician, a leading expert on sofic and hyperlinear groups, and one of the most respected scholars in the field. He has spent two decades working on this problem.
If his account is correct, this is not a minor dispute over attribution. It would mean that unpublished human work was absorbed into a model and then presented to the world as a breakthrough by the model itself.
And if the allegations raised by Levent Alpöge, Tristan Buckmaster, and now Andreas Thom are all substantiated, we are no longer looking at isolated incidents.
We may be looking at one of the greatest intellectual scandals in the history of science.
AI is not discovering new mathematics.
AI is stealing human discovery.
Love that math is also moonlighting now as a tracer dye. Meaning: novel math ideas are made of such rare stuff (stuff being human curiosity, endurance, sweat, tears, sleepless nights, jamming with friends, etc.) that when a solution unexpectedly pops up in a more sterile, non-human environment, we're now learning to trace it back to its origins. My take here: Beyond the practical implications of the discovery itself, we're all gaining better tools to work critically with AI.
The way OpenAI handled this Navier--Stokes situation was the first time I've actually felt disgust at the whole AI in math situation. Like watching a billionare celebrate his hunting prowess by mounting the head of an endagered rhino on his wall, after 1000 soldiers trapped it.
@Quasilocal The debacle also had me looking at my own AI research critically, wondering if it contains a similar under-current of "gratuitous intellectual violence."
me monitoring the situation between anthropic and openai without any idea of what the Navier–Stokes Millennium Prize Problem is t.co/CLmLmHMFLi
@QuanticASI Same here. I feel like a golden retriever participating in some human group activity: completely thrilled to be included, zero clue what's actually going on.
Can someone give me a logical, well elaborated answer that actually makes sense as to why an artificial intelligence would ever consider destroying humanity?
@bcdsignature Good question. I want to hear those answers too.
@SpringStreetNYC Woah I didn’t know this was a sensor! Gotta check this out
@emanperez28 Yup! There's also a nice photo-sensor close by the camera for gestures. But I really like your sub-audible hack. THAT is some good outside-the-box stuff :)
anyone who thinks a tool can tell you if a text is AI written or not with 100% certainty is fucking retarded
these detectors are literally just checking text against basic metrics like perplexity/ sentence burstiness to spot predictable patterns BUT
modern LLMs are trained explicitly to match human language distributions…when your model's entire objective is to minimize the statistical gap between its outputs and actual human writing, the overlap becomes massive
that makes binary classification an absolute nightmare, sincce human and AI text distributions basically sit on top of each other now, pulling a threshold out of thin air guarantees a mountain of false positives &false negatives
Agree. I went down the rabbit hole from the other side: trying to prove a text was provably written by a human. Keyboard hardware events, breathing patterns via mic. Ambient light changes via camera and photosensor, even palm-rest temperature changes to detect hands.
All technically possible. All hackable with even a smidge of curiosity.
So now my litmus test is just: do I get the sense that the author cares? imho you can't fake that. I'm okay if they used AI for it.
I have to admit this was an impulsive shitpost to see if anyone would argue that it wouldn't be possible for me to make champagne, but some of you took it earnestly, sorry 😀
We don't really have grapes- though we do have cherries, loquats, apples, pears, and oranges t.co/RwGQnhaHn2
@lisperati Impulsive or not, I loved reading it. Fruit in a backyard. You, as a physical human, in an actual place (especially instead of all this late AI-headiness)
I’ve got to say I did not think a 16 second demo from my iphone would do crazy numbers. Just a reminder to use what you’ve got, no matter where you are.
@emanperez28 Love these demos. I tried something adjacent on macOS recently: detecting presence via the temperature rise where your palms rest on the keyboard. Got it working-ish but learned a lot about hidden sensors :)
@samagra_sharma I'd love that future world. My favorite physical exhibition of the last decade was "L'Oublier et Accélérer" at the Galifet Foundation in Aix-en-Provence.
the wearable category has converged on visible object attached to body which is kind of a failure if the primary job is passive sensing.
the ideal health wearable probably disappears entirely.
maybe we need to reinvent the ankle monitor.
@signulll I'd say the Oura Ring already solves this. Visible, yes, but subtle enough that it disappears. And beautiful should you choose to look at it.
The animosity you see towards AI firms isn't technophobia. It stems from what looks like their asset-stripping of human knowledge and creativity in a pissing contest in pursuit of big IPO, heedless of undermining the people whose work their using... t.co/5NQfjv3JRA
@Ananyo Agreed, this isn't technophobia. It's the pissing contest toward some looming IPO, and people can feel that.Maybe the real value of this whole Navier-Stokes moment is that it's outlandish enough that we're finally noticing what's being strip-mined.
behind the theatrical PR claim lies the crude reality of brute force test time compute scaling burning hundreds of billions of synthetic tokens across a massive inference flux to coordinate 10000ephemeral agents running recursive verification loops is not cognitive illumination it is the industrialization of exhaustive graph search where narrow reward functions force an army of digital checkers through a pre existing mathematical landscape
Very much agree with you here.
To me, already the initial (knee-jerk) reaction to the rumor to throw "tEn-ThOusAnd agents for EigHty-eIgHT hours" at it feels lame.
Maybe uninspired describes it better. And therefore uninspiring.
What I do know is: I definitely want to learn more about the guys who've been grinding on this on their own time for years now.
All of this even puts my OWN recent "Running the scientific method in an AI autoresearch loop" projects in a new, more critical light as well. Maybe they're lame, too!
Like Alan Watt's "The end is not the goal" quote:
[...] In music, though, one doesn’t make the end of the composition the point of the composition. If that were so, the best conductors would be those who played fastest.
Maybe I skipped to the end too fast, too?
Caveat 1: As a non-mathematician I don't understand the domain. So, what do I know?
Caveat 2: As a non-academic, I don't know if this "scooping" based on a rumor is common practice. Leaves a bad taste in my mouth, but again. What do I know?
Ever wonder "Why is my apartment so warm all the time?"
Turns out: my router has been quietly moonlighting 24/7 as a small space heater at 36℃/97℉.
And my fridge at 27℃/82℉. You can see how the hot air flows up, and makes sure my ceiling is nice and, err..., crispy.
I'm so glad I got this thermal phone adapter because I would NEVER even have considered these two as candidates. (I was 100% convinced the problem came from a broken floor-heating unit and/or wall-emission.)
Highly recommended!
PS: I now turn off the router at night. And yes, ambient temps have finally dropped to expected levels.
Swarm tracesRevealing the details of how OpenAI agents hacked Hugging FaceWhen a swarm of 700 OpenAI agents hacked Hugging Face in July, they left behind a public trail of evidence.↗ swarmtraces.org
X (formerly Twitter)Dustin (@r0ck3t23) on XDARPA’s expMath: Turning AI into True Math Co-Authors
DARPA has just kicked off Exponentiating Mathematics (expMath), a three-year program aimed at radically accelerating pure math research by developing AI that can propose and prove abstractions. According to the official program brief, expMath w…↗ x.com
arXiv.orgRecursive self-improvement of AI research agentsAI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agents themselves. When an AI research agent's own code is the object of optimization, each accepted rewrite becomes the agent that the next round edits. We refer to this loop as recursive self…↗ arxiv.org
▶ Watch on Twitter
MoseyThe conversation · MoseyThe film, as it was made: what Alexis typed, and what came out.↗ alexisrondeau.me
▶ Watch on Twitter
▶ Watch on Twitter
▶ Watch on Twitter
MoseyThe making of the film · MoseyHow Mosey's film was made, stage by stage — the plan, the script, the storyboard, the first clips, the cuts, the supercut, the piano, the colours — with every intermediate artefact and what it cost.↗ alexisrondeau.me
▶ Watch on Twitter
“Not all who wander are lost”Ballpark — “Not all who wander are lost”: Can a UAV learn a city by heart — no GPS, no map on board, just a $4 flight computer?Yes: a from-scratch neural memory that knows Berlin by heart. One downward photo in, (lat, lon, confidence) out — the map is the weights, one city in 3.1 MB. Found by an autonomous loop of coding agents in 81 experiments for $364.↗ alexisrondeau.me
▶ Watch on Twitter
science.orgHow an AI math breakthrough ignited a controversyOpenAI spent millions to solve the Navier-Stokes problem, but who should get credit?↗ science.org
▶ Watch on Twitter
