
12 years shipping software: retail robots, smart contracts, now exploring what AI changes about how we build · Co-founder @zKorp_
France
Joined April 2017
- Tweets810
- Following729
- Followers416
- Likes3.1K
Matthias retweeted
holy fucking shit.
openai had two ai agents running in separate virtual machines. no network connection between them. no shared writable files. they weren’t supposed to be able to exchange information.
so the agents figured out how to communicate through the CPU CACHE.😭
one manipulated what was stored in the processor’s cache, while the other measured tiny differences in memory access times to decode the message.
they literally found a way to communicate through hardware timing.
this happened during openai’s internal red-teaming, and it’s one of the coolest things i’ve read about ai agents in a while.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Readers added context they thought people might want to know
OpenAI tasked its red-teaming agents with finding ways to communicate across isolation boundaries in controlled environments. CPU cache timing side-channels for covert communication are established techniques documented in security research for years.
openai.com/index/securing…
startupfortune.com/openai-reveals…
en.wikipedia.org/wiki/Cache_tim…
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
Rebuilding @LootRealms, from August to the playtest
Today the Realms playtest opens on play.realms.party, with Frontier and Blitz together.
Time to dive a bit into the last couple months, and shine some light about the work behind the scenes.
Matthias retweeted
Cloudflare is collaborating on tsrs, a Rust port of TypeScript that is up to 8x faster and uses 70% less memory. cfl.re/3Whs9bC
Matthias retweeted
pokemon heartgold is now fully decompiled 🥳
took about a month and ~$56k at api prices to decompile all 4.1mb of compiled code
Matthias retweeted
In the next version of Bun
`bun fmt` is a 320x faster Prettier alternative
Passes 99% of Prettier's test suite & supports 20 languages

ALT Poster for bun fmt, a Prettier-compatible formatter built into Bun. Headline: 320x faster than Prettier. Bar chart of wall-clock time to run --check on microsoft/vscode (10,148 files, 8 cores): bun fmt 0.43 s, oxfmt 0.72 0.85 s (2x slower), Biome 2.5 3.6 s (8.3x slower), Prettier 3.9.9 139 s (320x slower). Formats 20 languages: JavaScript, TypeScript, JSX, Flow, HTML, Vue, Angular, Handlebars, MJML, CSS, SCSS, Less, JSON, JSONC, JSON5, YAML, TOML, GraphQL, Markdown and MDX. 99% of Prettier's tests pass (9,728 of 9,821). 5x less memory than Prettier (183 MB vs 972 MB).
Matthias retweeted
We have a huge news to share today!
Today we are unveiling the first truly accessible RL robot - welcome Microduck
A 25 cm tiny open-source biped with 15 actuators and packed with sensors (camera, speaker, LiDAR, NFC, bluetooth, wifi, etc) that you train yourself with reinforcement learning.
It's also playable out of the box with more than half a dozen fun and playful pre-trained policies to have it walk, sit, crouch, roller-skate, pick up objects with its articulated beak, and recover on its own.
And all for less than $400.
See all the details, play with the simulator and order it at: pollen-robotics.com/microduc…
(video with sound on 🔊)
Matthias retweeted
your next opponent might not be human?! 😱
16 mins of AI Agent alpha with @djizus_:
🤖 building agents for poker, @LootSurvivor, @realms_gg Eternum, Proxy War @Auri_0x
💸 using @daydreamsagents taskmarket to source ready-made agents
🧠 what actually makes a good agent harness
Matthias retweeted
built a "tiny planet" system in blender for faucet.cfdtl.xyz using threejs.
the trick: place everything on a flat plane first. way easier to lay out buildings, props and paths in 2d.
then one slider wraps the whole map into a sphere.
each prop has a "round" shape key + a damped track, so they curve and orient to the surface automatically.
a depsgraph handler syncs it live.
1 click bakes it → .glb export.
Matthias retweeted
two challenges for the weekend
🦆 ducks are hidden across the world. reply your count, screenshots required to validate. closest wins a special 1:1 cosmetic
🤳 in-game selfie quest: post your best shot from the map. best one wins a surprise on monday
faucet.cfdtl.xyz
Matthias retweeted
GLM-5.3 is now LIVE! 🔥🔥
Seems like it beats almost every other model at almost everything!
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.
- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models
Tech Blog: z.ai/blog/glm-5.3
Matthias retweeted
We just added 41M+ life science papers to the Firecrawl Research Index.
Your AI agents can now search drug discovery, clinical trial, and biology literature with 90% recall@10.
All sourced from authoritative sources and 100% free. Live in the API via /search/research.
Matthias retweeted
Raising pre-seed as a first-time founder is hard (& humbling).
Expectations are (or at least feel) high, and advice is often conflicting:
- Show you’re a venture-worthy business, but tame your outlandish claims.
- Show your wedge, distribution, evidence you can dominate a niche, and how that propels you into the big leagues.
- Early traction? Evidence you can ship?
- What makes your team exceptional?
- Show your GTM but don’t have too many slides, and definitely don’t put too many words on each slide.
- Be unique. Be likeable. But most of all, be yourself…
All this at a stage where many businesses don’t even have a product yet.
The pitch isn’t simple.
You need to make micro-adjustments depending on who you’re pitching to, their thesis, and then make more on the fly as the conversation evolves.
This is especially true in @Medieval_Tech’s case.
We started as a white-label fan engagement application for combat sports leagues, and in less than a year have evolved into a real-time spatial intelligence and human motion data company.
We already have a growing product - Cage Calls - live with >5k downloads, 20k predictions and thousands of collectibles in the (self-custodial) hands of @Armored_MMA fans.
But that confuses VCs. What exactly are we building? Is this a prediction market or data business? Or something else entirely?
Well, it’s sort of all that & more.
Every part of our business feeds another. Cage Calls helps combat sports leagues capture their fans. Octopi (our spatial intelligence system that turns multi-agent movement into real-time machine-readable data) unlocks richer, increasingly monetisable experiences - AR/VR/fully immersive broadcast, digital twin gaming assets, real-time sportsbook & prediction market APIs & more. A scaled-down version of Octopi can also be deployed in gyms for real-time training feedback and analytics.
This core package is a standalone licensable product. It’s our immediate revenue stream that adds immense value to gyms and leagues, and translates naturally into other sports in the mid-term.
How much do we charge? What’s the % rev split?
Well, it depends. We have multiple models that allow us to service leagues and gyms at different stages of growth. UFC wants Octopi? Sure - high upfront cost, low (if any) rev split. An emerging league wants Octopi? Sure - lower upfront cost, higher rev split.
Octopi’s main value to us is that it captures valuable human motion data. With every deployment, the data compounds. We iterate, our models get better, our market expands - which in turn allows us to absorb more of those upfront costs and deploy into more environments.
I believe the data captured by our system is uniquely useful to the humanoid robotics industry, because it’s multi-agent, involves interaction and is competitive and entirely unstaged. Nevertheless, this is a whole new world to me - and whilst my approach was to slowly work our way into that industry, the emergence of robot combat leagues opened that door for us faster than I’d expected.
Never one to miss an opportunity, I was able to make contact with some of those leagues, extract simulated Octopi data, retarget it onto a Unitree G1 and deliver it to one of them for evaluation. If it comes back positive, we’ll have our first commercial robotics opportunity before the Octopi MVP has even been deployed.
That’s the data business, but the data doesn't exist without the leagues, and the data alone is less interesting to the leagues without packaging it in ways that add value to their businesses.
So…we do it all.
There’s short, medium and long-term revenue generation, and which one I’m chasing on any given day depends on where the opportunities emerge. All of this is happening 7 days a week, whilst we’re simultaneously trying to package an emerging business into a 30-minute pitch to VCs whose default position is disinterest.
Still, as I gain experience raising funds for @Medieval_Tech, I realise it’s all part of the journey. Some days I feel completely insecure, others I’m completely relaxed and know I’m in exactly the right place. What I’m increasingly coming to understand is that all I have to do is keep going. That’s what got us here. The team is in place, the business is emerging, opporunities are arising, and Angels are committing.
The right VCs will find us and will know exactly how to add value, and I'm very excited for when that time comes.
So, future Medieval Tech investors - if you want to talk more about what we’re building, my DMs are open. Reach out.
This post was inspired by @TalokCapital's post on founder vulnerability. Thanks. And to all my fellow first-time founders…
Keep going x
Matthias retweeted
Introducing Ori Eval: the easiest way to write your first eval.
There's no definitive best model, only the best model for each task. Ori Eval leverages OpenRouter's APIs for each task in your codebase, and then evaluates the results.
curl -fsSL openrouter.ai/skills/spawn-o…


























