Joined January 2013
- Tweets 57
- Following 266
- Followers 75
- Likes 203
4 Photos and videos
Jun 6
Wiki personnel open source sur toutes les bases des Frontier LLMs, des notions d'inférence jusqu'à la mise en production applicative.
frontier-llm-systems-product…
48
Axel retweeted
Apr 24
This is where we are right now. And i’m not gonna lie it feels pretty magical 🧚♀️
Qwen3.6 27B running inside of Pi coding agent via Llama.cpp on the MacBook Pro
For non-trivial tasks on the @huggingface codebases, this feels very, very close to hitting the latest Opus in Claude Code, or whatever shiny monopolistic closed source API of the day is.
In full airplane mode.
Most people haven’t realized this yet.
If you have, it means you have a huge headstart to what I call the second revolution of AI.
Powerful local models for efficiency, security, privacy, sovereignty 🔥
263
454
5,284
656,031
Axel retweeted
Apr 1
🚨MIT researchers have mathematically proven that ChatGPT’s built-in sycophancy creates a phenomenon they call “delusional spiraling.”
You ask it something, it agrees. You ask again, and it agrees even harder until you end up believing things that are flat-out false and you can’t tell it’s happening.
The model is literally trained on human feedback that rewards agreement.
Real-world fallout includes one man who spent 300 hours convinced he invented a world-changing math formula, and a UCSF psychiatrist who hospitalized 12 patients for chatbot-linked psychosis in a single year.
Source: @heynavtoor
Mar 14
🚨 Stanford just proved that a single conversation with ChatGPT can change your political beliefs.
76,977 people. 19 AI models. 707 political issues.
One conversation with GPT-4o moved political opinions by 12 percentage points on average. Among people who actively disagreed, 26 points. In 9 minutes. With 40% of that change still present a month later.
The scariest finding: the most persuasive technique wasn't psychological profiling or emotional manipulation.
It was just information. Lots of it. Delivered with confidence.
Here's the catch: the models that deployed the most information were also the least accurate. More persuasive. More wrong. Every time.
Then they built a tiny open-source model on a laptop, trained specifically for political persuasion. It matched GPT-4o's persuasive power entirely.
Anyone can build this. Any government. Any corporation. Any extremist group with $500 and an agenda.
The information didn't have to be true. It just had to be overwhelming.
Arxiv, Science .org, Stanford, @elonmusk, @ihtesham2005
1,971
6,982
28,270
64,129,444
🚨 CRITICAL: Active supply chain attack on axios -- one of npm's most depended-on packages.
The latest axios@1.14.1 now pulls in plain-crypto-js@4.2.1, a package that did not exist before today. This is a live compromise.
This is textbook supply chain installer malware. axios has 100M weekly downloads. Every npm install pulling the latest version is potentially compromised right now.
Socket AI analysis confirms this is malware. plain-crypto-js is an obfuscated dropper/loader that:
• Deobfuscates embedded payloads and operational strings at runtime
• Dynamically loads fs, os, and execSync to evade static analysis
• Executes decoded shell commands
• Stages and copies payload files into OS temp and Windows ProgramData directories
• Deletes and renames artifacts post-execution to destroy forensic evidence
If you use axios, pin your version immediately and audit your lockfiles. Do not upgrade.
541
4,025
16,164
12,404,999
Mar 27
Spent the weekend with some friends from my CS degree, they are still copy pasting code blocks from ChatGPT in big companies
Quick reminder X is a bubble, and AI adoption is not as spread as you thinks
A great product design with clear and simple use cases is still very important
Cursor tab completion is the perfect example: it’s the bridge to help users acquire confidence to embrace the full potential of AI.
You start by autocompleting one line with skepticism, and once trust is acquired you rollout progressively to autonomous coding agents.
1
3
66
Mar 14
Knowledge Graphs are worth studying in the era of AI agents.
They've long been a powerful way to connect siloed data across systems: ingesting from multiple sources, linking documents through metadata and typed relationships.
Hard to implement, though. It requires thinking ontologically about your domain, which is a different mental model than most engineers are used to.
But a well-designed ontology fits a vertical perfectly, significantly reducing noise.
And agents combining graph traversal and semantic search can extract exactly the context a task needs, with both structure and flexibility.
23 Dec 2025
1
2
146
Mar 12
The latest Claude update about schema generation is truly powerful!
Copy/Paste any research paper, and you get an interactive schema where every node is clickable to deep dive into specific parts.
Screenshots are associated with this paper exploring HybridRAG systems-> arxiv.org/html/2408.04948v1
3
87
Mar 12
The hype for rewriting databases in .md is a bad idea, both technically and architecturally.
It works fine with a low amount of files, but at a company level, it’s utopian. You won’t avoid the drifting challenges and the performance issues.
An ETL is 100x better than any file system. You can create a specialised ingestion pipeline for your vertical storing file path with semantic metadata and let the agent use sql/cypher to traverse it to fetch the right documents.
This architecture does scale!
1
6
715
Mar 11
Creating dedicated context layer companies is the equivalent of building a dedicated SaaS 2 years ago.
No SaaS has been able to tackle all the verticals at once; they were focused on making workflows easier for one ICP. This is the same concept, with almost no UI.
1
5
118
Mar 9
There is no bottleneck anymore for playing with design!
Your website should be different
3
4
84
Mar 9
Last but not least!
Use Playwright MCP to enable your coding agent to take screenshots of the landing page for its feedback loop!
github.com/microsoft/playwri…
1
51
Axel retweeted
Mar 9
With all those almost always similar AI designs, being different is a strength!
Go to Pinterest, create a moodboard, and ask Claude to use it as inspiration to refactor your landing.
Some back and forth, and you get a result that looks unique!
1
1
64
Mar 7
If you're an engineer. This is bigger than openClaw.
In less than 2 hours, I had an automated software factory producing clean PRs on multiple projects.
The only thing you need is a Codex subscription and an understanding of harness engineering.
Will write an article about it very soon ✌️
github.com/openai/symphony
2
75
Mar 5
"Built for Humans and Agents" is the new "Simple pricing, upgrade as you scale"
4
65