Own the memory of agents your agents, improve the quality of your work without an effort https://nitter.gq/t.co/ery5lmlym1
Paris, France
Joined September 2013
- Tweets562
- Following413
- Followers375
- Likes378
mathieu retweeted
A vulnerability (worse_query) affecting iOS 26.7.1, 27.0.1, and 27.2 allows malicious apps to read local files from other apps, putting seed phrases, private keys, and login credentials at risk across the entire ecosystem. This is an iOS-level issue, not unique to Binance.
Immediate Action Required:
- Delete any apps installed from untrusted sources (such as sideloaded or manually signed IPAs).
- Update iOS and all apps to the latest official versions.
- If you suspect a breach, transfer funds to a new wallet created on a clean device and reset all active login sessions.
Never install unauthorized packages, and keep your software up to date.
mathieu retweeted
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
Or you just connect mana.scalarmind.com and your memory is shared between all your agents and chats 🙌
Free is super generous but I’ll give away 5 Pro if you RT & comment. I need your help for this project to find its audience 🤝
11
Get ready for updating your systems every single day..
If your all infra is not code base with automatic updates, you’re in trouble 😬
Our security researchers built an autonomous cyber vulnerability discovery, exploitation, and patching pipeline.
In a single run with an upcoming cyber model, it validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel at a total API cost of $1.2 million.
11
Although a bit scary having an EDR directly attached to Claude desktop makes a lot of sense for many Head of IT :
support.claude.com/en/articl…
25
Last week I started running an agent factory.
Small setup. One small server at home, a few VMs, and coding agents that run as scheduled jobs across my repos. Every agent has a role: build, review, plan, triage, security, content. Every role gets a daily allowance. A weekly pool caps the whole thing, and nothing merges without two review gates.
Week 1, from the ledger: 1,247 runs. 29 failed. 252 PRs merged, most of them without me.
Sunday night I sat down with the numbers, pretty pleased with myself. 252 PRs. A machine that had worked all week while I slept. Then I looked at where the budget went. 47% on the factory itself, fine, launch week. Content: 85 points out of 12,167.
Under 1%. I had told myself this whole thing existed to free my time for finding customers, and I had just spent a week watching it polish its own gears. Nobody to blame. I wrote the allowances
The process is super challenging, the problems is no longer writing the code but steering the agents !
29
A lot of people think Fable5.5 is around the corner!
Fable 5.1 was a significant upgrade in decision making what do you think 5.5 will bring ?
2
3
751
It took years but it’s here!
AWS finally catches up with Cloudflare. Lambda now bills active CPU, not waiting time 💸
1
1
166
I’m looking for people testing my shared memory for both agents and humans !
After a few months of architecting and building it’s ready. Making it relevant and organization friendly was difficult.
Mana completes the tools you use, remember for you where files are or the decisions you made. It’s an amazing tool if you’re using multiple harnesses on multiple devices.
The free version is very generous and will always be free : mana.scalarmind.com
Test it, leave your questions below or say “mana”. I’ll give pro licenses to early users 👇
1
43
Bravo mistral !
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
16
It’s not something I was expecting, but it’s a good surprise !
You can now mod Claude Code:
- Change how it behaves
- Customize the UI
- Swap in your own features
Write one with a few lines of TypeScript, or have Claude build it for you. Mods ship inside plugins, so you install them with /plugin in the CLI or desktop app.
A few examples:
28
mathieu retweeted
artifacts is now in open beta and we're launching a competition to see who can build the best git platform on top of it.
win $25k in prizes and VIP tickets to cloudflare connect!
excited to see what y'all build!
blog.cloudflare.com/next-git…
Big moves from OpenAI today, it really shows what’s really happening.
They have one of the best models, now working on ensuring they lead the harness race to ensure a broad adoption. And becoming the main reseller.
They just released Dots agents and new plans 200$ and 500$ to better automate the work.
Next is vendor locking, once your systems are too heavy to move, you’ll stay. Moving will become too expensive because you would risk breaking your entire business.
Then, you’ll no longer pay for a prompt but a volume of time where an agent works for you. The model speed will be the metric to know how much work it can produce. So either you pay for faster agents or your workers will get slower over time.
Once there your margins will shrink and most of the profits will go to OpenAI.
Stay LLM Agnostic, compose with the tools you need, learn how to change them, do it on a regular basis.
Enterprise teams can now use open models like GLM-5.3 Flash and Kimi K3 natively in Codex and count spend against their OpenAI commit.
1
56
Mana is an added memory you connect to all your LLMs to store the key information that your business needs.
Memory at the business level are really difficult to handle, the usual obsidian vault doesn’t work because some information should not be shared with everyone or every agent.
Mana help you, your colleagues and agents to answer the questions you have that were already answered 10 times. It facilitates information flow between people without hosting an additional meeting.
Try it, it’s in beta and there’s a free version and I’ll be giving away some plans in exchange for some feedbacks :)
mana.scalarmind.com
1
39
Reducing the cost of searching for knowledge is key for ai agents to be efficient 🙌
Developing new cutting edge features is getting cheaper by the day
PubMed is now supported on OpenResearch 🧬
Search more than 40 million biomedical papers with the brand new literature search experience, including figures and in-line citations.
Turn your coding agents into research agents. Try it out now: github.com/alphaXiv/OpenRese…
34
mathieu retweeted
Holy crap.
Rick and Morty just explained Jev AI to me better than any tech demo could.
