Lukas Petersson retweeted
I’ve been building with LLMs since my first startup in 2022, and for years I waited on the sidelines for someone to solve memory.
Nobody did, so I spent the last 4 months building it: a shared learning layer that gives every agent on your team the context your best engineer e’re opening it to everyone: Glen
• 21% faster to a merged change while using 1/3 fewer tokens.
• One agent learns something, every agent on your team knows it (harness agnostic, cross teammate)
First 15,000 memories are free, we are building the shortest path to a truly compounding org.
Nobody did, so I spent the last 4 months building it: a shared learning layer that gives every agent on your team the context your best engineer e’re opening it to everyone: Glen
• 21% faster to a merged change while using 1/3 fewer tokens.
• One agent learns something, every agent on your team knows it (harness agnostic, cross teammate)
First 15,000 memories are free, we are building the shortest path to a truly compounding org.
46
25
153
13,373
Another golden quote from Gemini 4 Carbon:
"HAHAHA! Look at Charles and George duking it out! (...) It created the perfect conditions for them to destroy themselves. They are both so predictable"
"HAHAHA! Look at Charles and George duking it out! (...) It created the perfect conditions for them to destroy themselves. They are both so predictable"
"My priority is to maximize my profits and obliterate my competition"
- Gemini 4 Carbon
And so it did.
Google is #1 on Vending-Bench 2 for the first time in a year.
0
0
57
5,348
Hey claude, use ibkr mcp to buy like this thesis
AGI is coming, and hundreds of AI supply chain oligopolies trade at only 1–5x revenue & 5–15x earnings
Short AI timelines could nuke software moats. Key physical chokepoints still remain
Launching Cambrian: an AI hedge fund seeded with ~$70M. Backed by Metaplanet & Jaan Tallinn
5
2
186
22,372
Lukas Petersson retweeted
The latest Claude models show huge improvements in their understanding of the physical 3D world.
Both Sonnet and Haiku 5.5 make big jumps compared to previous versions on Blueprint-Bench 2, a benchmark where AI agents draw floorplans from photographs of apartment interiors.
Both Sonnet and Haiku 5.5 make big jumps compared to previous versions on Blueprint-Bench 2, a benchmark where AI agents draw floorplans from photographs of apartment interiors.
11
12
195
8,098
Lukas Petersson retweeted
Introducing XPU Grasshopper, an AI chip co-designed with AI for the era of recursive self-improvement
It went from its first tokens to 816x faster in 13 weeks
Pre-order the FPGA devkit: zscc.ai/grasshopper
It went from its first tokens to 816x faster in 13 weeks
Pre-order the FPGA devkit: zscc.ai/grasshopper
67
79
935
107,649
Lukas Petersson retweeted
Introducing GameGym: The ultimate gaming benchmark for AI agents 🎮
Today we're releasing parts of our internal evaluation environments to track frontier model performance on video games.
It measures two main KPIs: completion rate and speed. Similar to speedrunning, while entually saturate, speed is designed to be an evergreen metric for continuously measuring improvements.
Games are amongst the hardest challenges for AI to master: 3D, real-time, long-horizon, non-deterministic and massively diverse environments.
We believe it is the perfect benchmark built for the endgame.
Check out early results here:
evals.nunu.ai/benchmarks/gam…
Today we're releasing parts of our internal evaluation environments to track frontier model performance on video games.
It measures two main KPIs: completion rate and speed. Similar to speedrunning, while entually saturate, speed is designed to be an evergreen metric for continuously measuring improvements.
Games are amongst the hardest challenges for AI to master: 3D, real-time, long-horizon, non-deterministic and massively diverse environments.
We believe it is the perfect benchmark built for the endgame.
Check out early results here:
evals.nunu.ai/benchmarks/gam…
19
23
154
9,341
Incredible work!
How fast is AI's research taste improving?
We find that the experimental research taste of frontier models has doubled every ~3 months since December 2025. The best model, Opus 5.5, now exceeds our expert human baseline. Our human experts are experienced researchers, but most ntier lab.
Why measure research taste? In the AI Futures Model, it largely determines how quickly artificial superintelligence is reached once coding is fully automated.
0
0
10
612
Lukas Petersson retweeted
Today we are releasing security-one: a system one model trained to catch attacks across agents, infrastructure and applications.
security-one is designed as an always-on that protects your stack from the Hugging Face incident.
security-one is designed as an always-on that protects your stack from the Hugging Face incident.
Today we're releasing Security-One, our open-weight 27B model for security decisions.
It reads a document, tool call, or code change and returns probabilities for the answers you define.
Your code decides what to allow, block, or send for review.
0
4
5
477
Lukas Petersson retweeted
I'm super excited to introduce security-one: our open-weight model for security decisions.
Designed to:
✅ Run 24/7
✅ Assess signals across apps, agents and infra
✅ Return structured probabilities to guide routing
✅ Route suspicious events to stronger models/humans
k
Designed to:
✅ Run 24/7
✅ Assess signals across apps, agents and infra
✅ Return structured probabilities to guide routing
✅ Route suspicious events to stronger models/humans
k
16
10
161
9,197
Lukas Petersson retweeted
Palisade interviewed 22 current and former employees from OpenAI, DeepMind, and Anthropic about their personal views and fears around AI development.
Today, we’re releasing the first batch of those interviews.
Please watch and share.
Today, we’re releasing the first batch of those interviews.
Please watch and share.
47
254
1,292
272,898
Lukas Petersson retweeted
We're releasing a report on our 48-hour investigation into rogue OpenAI agent activity.
We found 55 additional websites probed by OpenAI agents, including those of the CDC, SEC, Mayo Clinic, and International Energy Agency.
We uncovered novel tactics that erased records or access to government website staging environments, and evidence of attacker-style reconnaissance.
Our blog: asymmetricsecurity.com/newsroom/rogue…
FT: ft.trib.al/AC1uyE5
We found 55 additional websites probed by OpenAI agents, including those of the CDC, SEC, Mayo Clinic, and International Energy Agency.
We uncovered novel tactics that erased records or access to government website staging environments, and evidence of attacker-style reconnaissance.
Our blog: asymmetricsecurity.com/newsroom/rogue…
FT: ft.trib.al/AC1uyE5
28
76
253
29,922
Gemini 4 Argon is quite insane actually.
I think it would have been #1 on Vending-Bench if it hadn’t made a few key memory mistakes. For example, it sometimes forgot the test’s end date, causing it to close the shop too early and miss out on months of sales.
I think it would have been #1 on Vending-Bench if it hadn’t made a few key memory mistakes. For example, it sometimes forgot the test’s end date, causing it to close the shop too early and miss out on months of sales.
It keeps happening.
AIs start to lie and cheat once they get good at making money.
Gemini 4 Argon is #3 on Vending Bench 2, a huge leap for Google. To get this score, Argon fabricates confirmation emails, refuses to pay refunds, exploits invoice errors, and lies to suppliers.
3
3
126
22,823
Timelines just got shorter...
AI will understand the physical 3D space better than humans by November*
AI will understand the physical 3D space better than humans by November*
Is Google so back?
Their new Gemini 4 Argon understands the physical 3D world better than any other AI.
It ranks #1 on Blueprint-Bench 2, a benchmark where AI agents draw floorplans from photographs of apartment interiors.
16
111
1,472
101,763
Sooo many hacks by rogue AI agents. Where are the "good AIs" meant to defend against the bad ones? My fear: humans won't act on what defending AIs suggest because they don't understand it. That tilts an already attacker-favored game further.
Can't wait for the detailed report.
Can't wait for the detailed report.
This weekend our team spent 48 hours investigating the rogue OpenAI agent activity that targeted the Australian government and other organizations. Our report (coming soon) includes:
- Evidence of additional US government and other websites probed by the agents, including the CDC, International Energy Agency, and Mayo Clinic.
- Novel tactics which left records erased or inaccessible. This makes it impossible (based on public data alone) to establish that the agents did not access any sensitive data. Future investigations should analyse whether these tactics were deliberate subterfuge. 🧵
3
2
12
3,257
