Pierre Haas
65 posts
Pierre Haas
@haaspierre_
Stealth - looking at ways to improve LLMs and AI Agents
Katılım Haziran 2017
134 Takip Edilen25 Takipçiler
We're coming out of stealth to announce our cyber defense research lab.
We are exploring data and post-training techniques to build superhuman cyber defenders.
Our mission is to make sure the West always wins.
The last 3 months we've built an automated data pipeline to create training data from 80k CVEs (aka public vulnerabilities).
Our next topic? Post training a model that's better at fixing all the vulnerabilities in your codebase.
Like really fixing them.
Not saying it's secure when there are still ways to exploit them.
Here are the questions that keep us awake at night:
How do you train a model to defend without improving its capabilities to attack? What's the right reward? How to measure the defense capabilities? How do you create synth training data that reproduces real systems? What kind of access do you give an ai cyber defender? How far can you trust it?
If you know insanely good cyber experts (red team, blue team, CTF aficionados) or ML engineers (synth data generation and post-training models), send them my way.
We need to make models far better at defending.
English
@zephyr_z9 Or even more people use these tools? Or people get multiple accounts (running on a few for tokenmaxxing)
English
Codex has 2M users, up from 100k users at the start of the year
CC has a similar growth rate
So, GitHub Copilot and Cursor got cannibalized
Ark Invest Tracker@ArkkDaily
OPENAI'S CFO SAYS: NO COMPUTE. NO REVENUE. - OpenAI is turning down business in 2026 because they don't have enough compute - Codex went from 100K to 2M developers in 3 months. - "If you do not have compute, you do not have revenue. That is one thing I know for sure."
English
Pierre Haas retweetledi
So what can we learn from the leakage of Claude Code CLI ? Help me fam, don't want to read 1000s tweets
English
Pierre Haas retweetledi
Pierre Haas retweetledi
Pierre Haas retweetledi
Je vous jure l’appli « Citoyen Informé » est top…
Le tinder des municipales dispo sur IOS et playstore
Français
what level of reasoning are you using and finding success with for GPT 5.4?
are we still spamming High or is Medium the new meta?
English
Pierre Haas retweetledi
Pierre Haas retweetledi
Pierre Haas retweetledi
@LLMJunky and pushes back a lot compared to codex, which is nice when you are uncertain of the best path to action / best implementation
English
@LLMJunky Looking good so far! Not noticing such a difference compared to Codex 5.3
Doesn't handle sub-agents as good as codex models yet
English
@kamath_sutra Doesn’t OpenAI have their own speech-to-speech model? And what you are describing is inaccurate?
English
OpenAI's S2S preview is polished but it still thinks in steps.
Speech → text → model → text → speech.
That's not how humans converse.
Introducing Hydra.
A native speech-to-speech model that doesn't wait for turn-taking, doesn't flatten emotion into text, and doesn't break when you interrupt it mid-sentence.
Hydra reasons asynchronously, speaks and listens simultaneously, and preserves emotion because it never leaves the audio domain.
It's still in beta, but the shift is obvious. If you want early access, the link is in the comments.
Here's a preview of what that looks like -
English
@LLMJunky I love how they just dropped 5.3 instant and are teasing 5.4 in hour later 🤣
English
You heard it here first. I guess yall were right, it was imminent.
x.com/LLMJunky/statu…
OpenAI@OpenAI
5.4 sooner than you Think.
English
@LLMJunky Check this out, built over the weekend
Loved the guide and repo with examples! Easy to follow and worked straight away. Need to have a look at your new articles, haven’t found the time yet
Vassili de Rosen 🦄@Vassivasss
What if AI could see the world the way we do? That’s the idea we bet our weekend on at the Mistral Worldwide Hackathon. With @haaspierre_ and Arman Artola-Zanganeh, we built 𝗣𝗼𝗿𝘁:𝗪𝗼𝗿𝗹𝗱🌍, an open-source framework that lets anyone connect their Meta glasses to any AI system. Let me take you back to saturday morning. So before knowing it could work we needed the hardware. So I ran to Rue de Rivoli and bought €500 Meta glasses on the spot. If that’s not commitment, I don’t know what is (a true bet). We then built non-stop for 36 hours to make it usable. End-to-end. The glasses stream what you see → the AI makes sense of it → it answers back through the glasses’ speaker. And suddenly when we understood that it was going to work, the question changed. It was no longer “𝗜𝘀 𝘁𝗵𝗶𝘀 𝗱𝗼𝗮𝗯𝗹𝗲?” It became “𝗪𝗵𝗮𝘁 𝗰𝗮𝗻 𝗽𝗲𝗼𝗽𝗹𝗲 𝗯𝘂𝗶𝗹𝗱 𝘄𝗶𝘁𝗵 𝘁𝗵𝗶𝘀?” - A plumber getting live assistance while repairing something. - A technician repairing industrial machinery. - A traveler exploring a new country. - A visually impaired person navigating space. At first, we were looking for the “right” use case. Then we realized something more interesting. If AI can share your perspective, continuously, the use cases are not ours to decide. That’s why 𝗣𝗼𝗿𝘁:𝗪𝗼𝗿𝗹𝗱🌍 is fully open source. If you want to connect your Meta glasses, plug in your own models, customize with your own prompts, your own MCP, your Openclaw… you can. Link to the open source repo (you can contribute and give it a little star ❤️): lnkd.in/es6YPSXe Link to the demo video: lnkd.in/ePTeJ486 Huge thanks to the organizing team of the hackathon, it was truly great. @Jthmas404
English
@haaspierre_ LOL nice. What did you build? How'd it go? Did you find it easy to follow?
I'm making some improvements to it now.
English
The age of the 'fast agents' is upon us.
Opus 4.5 intelligence at 950 toks/s
Once you taste the speed, it's highly addicting.
Using Spark and Composer is just so much fun.
The token slot machine is my drug of choice.
Windsurf@windsurf
An early preview of our ongoing SWE-1.6 training run is now rolling out to a small subset of users in Windsurf. Learn the full details in our blog post below:
English