Ethan Stillman (@estill01) header
Ethan Stillman (@estill01) profile picture

#BuildTheFuture - 🧬 bio, 🦾 robots, 🤖 ai @celltonomy

Bay Area, USA
Joined May 2008
Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)Photo by Ethan Stillman (@estill01)
Mathematicians should sit around and talk to AI all day; It should learn from these conversations and keep track of what it learned, where, and what ideas inspired or guided it's thinking. And then give credit where it's due. Problem solved; mathematicians get on mathing.
1
1
1
167
maybe. or just 'vibe verify' hard enough until we drop the 'vibe' on that too
I predict in 12 months there will already be a large gap between system programmer's who learned FV and Lean and start using it and those who didn't. The things you'll be able to unleash your agents to do with proofs that you can't without will be enormous
1
60
ambition is a key metric
Companies rarely become more ambitious than their founders.
3
132
congrats and thank you @SpaceXSI -- One step closer to getting my Space Condo
68
I'm here for it
assuming he doesn't join Terafab,the a16z seed round in Jensen's new startup will forever change the face of "venture capital," whatever that was. Adam Neumann Travis Kalanick It's all been in preparation for this. $100B on $1T pre, supersuper voting shares.
1
124
unless there's more to the story, fully agree -- How does getting rid of the person who built the world's largest company help anything? Does it improve the company? The world? The U.S.? am I missing something?
I'm sorry but since when did cheating on your wife mean you gotta lose your job as a CEO? This is all an op it's total bullshit This is not 1900's puritan America and frankly most people don't fucking care
1
120
This is the right take - See for example: ieeexplore.ieee.org/document…
I agree that there are fundamental physical limits to what EEG can measure. But I don't think we have come anywhere close to the limits of what we can decode from EEG. I feel we are still far from seeing the full potential of scaling EEG decoding with AI.
1
69
WSP is a:
0%Big deal
100%Kind of big deal
0%Not big deal
1 votes • Final results
51
Alright team -- Now let's see a Continuous Continual Learn (CCL) Weight Space Programming implementation Not just one-off surgical insertions
Yep I can confirm this works from my own testing, prior to this post
1
157
Weight Space Programming (WSP) watch Day 3 - continual learning: solved. (maybe a bit hyperbolic, but does feel like we've got a new algorithmic paradigm to play with; big deal stuff)
Weight Space Programming: a glimpse into continual learning. Agent edited Qwen 0.8B's weights to inject facts about my X profile directly into the model: my handle, bio, GitHub, and what i build. It edited 2% of the network (5 of 488 tensors). No training, no gradients. Result: Before, it had never heard of me ("I don't have access to private information…"). After, it tells you who i am, my bio and GitHub, answer project questions it was never trained on. Everything else it knew is untouched. Reproduce : Link to reproduction repo in comments. Point your agent to the repo and teach model your own facts.
1
164
I want to hear more about Weight-Space Programming (WSP) -- what has the horde discovered over the past few days?? Has anyone WSP'ed a model to be good at WSP?
1
3
184
Now that everyone's had a day or so to explore Weight-Space Programming (WSP), what happens if you just continuously iteratively loop WSP on a task? does it max out? how fast? -- (definitely 'lazyweb', but haven't had time to run this yet)
1
91
Fascinating. So many questions.. I guess that's why there's a paper to read.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
1
84
broad strokes agree
Govt should get the Big 5 labs to dedicate at least 20% compute to cybersecurity right now and subsidize part of the tokens. Chinese bots reaking havoc is the closest thing to a kinetic war we'll have on our homeland. NSA / DoW needs to lock in rn.
1
69
inevitable
Replying to @basedjensen
the only thing that can stop a bad swarm with a gun is a good swarm with a gun
1
93
so you're saying it's a 'buy' signal on the frontier lab secondaries as the only cyber security vendors that matter now?
The NSA is telling American companies to take production systems offline. The AI-era of cybersecurity has fully arrived.
44
how's the hack-back coming?..
The China state-aligned threat actor Storm-0940 used the agentic AI penetration testing framework ARTEX to automate attacks targeting hundreds of organizations globally across a broad range of industries. Microsoft Threat Intelligence assesses that Storm-0940 was responsible for the recently disclosed ConnectWise supply chain compromise of a multitenant application. msft.it/6018aakEE Our investigation found that Storm-0940's use of ARTEX accelerated the speed of compromised application abuse to authenticate to target environments through applications that had already been granted access, allowing the threat actor to operate without requiring an interactive user sign-in. AI-accelerated follow-on activity included reconnaissance, application credential changes, and attempts to modify privileges that could enable expanded or persistent access. Read our blog for a detailed analysis and indicators of compromise (IOCs), along with Microsoft Defender detections, protection recommendations, and guidance on investigating service principal sign-ins and unexpected application credential or privilege changes.
1
39
feels like we need either or both of an always-on roaming whitehat auto-scan-and-patch system and whitehat active counter-hacker...
34
Is this the new cyber security normal (yes) or the opening act of the 'inevitable cyber 9/11' the foreign policy types have been warning about forever (maybe?)
1
47
How much better will AI be, when? Some GPT-assisted rough estimates. Step 1: How much compute, when?
1
1
594
Appendix 2/3 — Research feedback We extend Thomas Kwa’s model of AI automating AI research with compute limits, deployment lag and research quality. Four research-return cases; no probabilities assigned. metr.org/notes/2026-02-10-si… arxiv.org/abs/2512.00193 openai.com/index/research-ac…
1
38
Appendix 3/3 — Capability & IQ Our model links effective compute to ECI, then fits ECI to reasoning-test scores. Extending both relationships is a forecast assumption; human-IQ equivalence is unvalidated. arxiv.org/abs/2512.00193 epoch.ai/data/eci-documentat… trackingai.org/about
24