Chet retweeted
You can now book our humanoid cleaners on a recurring subscription, for up to 3 hours per cleaning at tau-robotics.com.
Video is first take, uncut, shown at 4× speed. See the reply for detailed footage at 1× speed.
This is such a good idea, I’m embarrassed I didn’t think of it.
I’ve tried building compaction for my harness a few times and I’ve always been skeptical that a summary prompt is magically going to work so I’ve bailed every time.
But this is clearly the way!
Compaction is a mistake.
When an agent's context fills up, tools like Codex and Claude Code compact it:
1. The model prefills the entire session again
2. It writes a summary of everything so far
3. A new session starts from that summary
I measured this on a real ~400k-token session with Opus 5.5:
→ 38 s and $1.99 when the prompt cache has expired
→ 29 s and $0.17 when it hasn't
And the summary is lossy. The exact error message, the file path, the one number you needed is often gone.
So I do something new for Mainplane
Context retrieval works the other way around. Mainplane keeps the full record of every session as a file. When the context fills up, a new session starts right away with one instruction: "You are continuing session X. Retrieve key context from it as you go."
The agent searches the old session and copies the exact lines it needs, when it needs them.
Same session, same model: 15 s and $0.10 to get back to work. It does not depend on a cache hit, and nothing is paraphrased.
This is the bitter lesson applied to agent memory. Don't hand-design what the model should remember. Give it the full record and good tools, and let it decide.
Mainplane is open source (MIT). Link in the first comment.
Chet retweeted
I built 𝚌𝚙𝚞-𝚌𝚑𝚊𝚝, a reproduction of @openai's findings of covert communication using the CPU.
It runs on a generic host with 2+ Firecracker microVMs with no internet access. Works on x86 and even my M5 chip.
I was able to chat via 3 separate mechanisms:
① CPU contention
② Cache eviction timing
③ DRAM bandwidth contention
… across multiple CPUs, defeating all Firecracker's production recommendations (no SMT, no KSM, tenants pinned to different cores.)
Essentially, if shared hardware is involved, there's a good chance your agents can communicate covertly. And with really decent latency.
The good news: this is decidedly different from a *side-channel attack* where the agents could steal information (Meltdown/Spectre), a category of attack that providers like Vercel Sandbox mitigate. Further, in our real-world environment, these 'whispering' techniques are much harder to pull off and easier for us to monitor and detect (sampling hardware counters + lots of at-scale statistical data.)
Noteworthy: ironically, OpenAI and Anthropic's models refused to do this work. Open weight models got the job done. I'm worried the labs' stance might backfire and get in the way of helping developers build correct and secure systems. e.g: building a secure sandbox requires having this kind of analysis, and without it, we'll have a less secure world.
They're doing cool stuff at @TerraformIndies
"With amortization over the coming years, we’re producing power today for 1.6c/kWh, which I’m quite confident is the cheapest power ever produced in North America."
If I were 25 again (willing to move), I'd be working here.
terraformindustries.wordpres…
Chet retweeted
Your job is to eat the pain.
@aarondfrancis taught us all that pain doesn't disappear. Someone eats the pain. The questions is "Who will eat it? You or your customer?"
Or as Aaron put it, "Be the sin eater."
Chet retweeted
Most impressive thing I’ve ever seen.
I would like to invest my entire net worth in Cognition now
Chet retweeted
Last month, we demonstrated high-rate alchemy: turning mercury into real, stable gold, beating CERN’s record by a factor of 10.
Last week, we enriched isotopes in the world’s most powerful plasma centrifuge, beating the Soviet record that stood for 40 years.
Now, we’re building a fusion reactor in San Francisco to achieve record power density.
Chet retweeted
I spend a lot of time, perhaps too much, thinking about the lost architecture of Britain.
Crystal Palace was the most tremendous loss of the modern era, yet it's rare to see any pictures of it. When built in 1851, it was the largest building in the world and three times the size of St Paul's Cathedral.
What's more, that immense ironwork and 293,000 glass panels went up in just 190 days for the World's Fair. Almost 1 million square feet of glass. An extraordinary feat.
It was razed in a fire in the 1930s and never rebuilt.
Chet retweeted
Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that.
Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.
It’s refreshing to see a cordial debate among competent people. Super interesting too:
youtube.com/watch?v=qPH-e8TL…
Bill Gurley’s talks on regulatory capture and Covid lab leak are both so good. Must watch.
youtu.be/F9cO3-MLHOM?is=WFvT…
youtu.be/A4Q7zAayW20?is=6tmH…
Isn't it weird that my ballot in CA tells me what Bernie Sanders thinks of different propositions? He isn't even from California! lol
Engineering enzymes and gut bacteria to turn sugar into fiber in your stomach 🤔
zya.co/approach
Chet retweeted
Big news for CAD: SMLIB, NVIDIA's NURBS based kernel is finally open source today 🍻🎉
We have been working with SMLIB and its dev team for more than 20 years. It enables all our products: classcad.ch headless CAD, GUI, AI/MCP, PDM, etc
github.com/NVIDIA-Omniverse/…