Joined January 2016
Angelo Poerio retweeted
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…
956
1,265
1,479
10,906
2,901,313
Angelo Poerio retweeted
this is wild connected the kindle over usb and let claude code jailbreak the kindle removed all the Amazon bloat + added tailscale
301
564
148
13,813
824,858
Angelo Poerio retweeted
Start here: the primer on data compression fundamentals mattmahoney.net/dc/dce.html
We keep improving performance across Claude, and you may have noticed it starts responding sooner! Claude Code now gzips the conversation it sends on every turn, so it reaches the model faster. Cloud sessions also start up as soon as you start typing.
55
631
13,347
Experimenting with Agentic control of a robotic arm via an MCP server github.com/angelopoerio/robo…
1
85
Angelo Poerio retweeted
major vibe shift: Europe now has the best open AI model in the West 🇪🇺 "It is the best open weights model from US or Europe on aggregated benchmarks." i'm especially bullish on Mistral because they've essentially been anointed as Europe's AI champion: > €3 billion raised. the biggest tech equity round in European history > it's the last European-owned company still building top AI models (Germany's Aleph Alpha got bought by Canada's Cohere) > Jensen Huang told Macron that Mistral needed France's biggest companies rallying behind it. Macron's reply: "who are they? let me call them." and he personally called them > Macron went on national TV and told the French to download Mistral's chatbot instead of ChatGPT > the EU's own scaleup fund co-led their round, with Dutch chip giant ASML and the government of Luxembourg also writing checks > France + Germany have committed to bringing Mistral + SAP into their governments > the European Space Agency just signed on to build with Mistral > and on top of all this, the EU has mobilized €200B for European AI, including €20B for AI gigafactories Europe knows what's at stake: building and controlling our own powerful AI is an existential geopolitical priority and Mistral is our only real shot at competing with the Americans and the Chinese. so you have the most powerful people in Europe in their corner, with a *very* concrete reason to help them win. that backing is massively beneficial as it gets them priority access to money, compute and government customers. any startup would kill for this competitive advantage, and it's why i think Mistral will keep delivering. i'm very happy to see Europe shipping like this and i also love the commitment to the Le Chonk bit lol eu/acc 🇪🇺
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.
54
97
18
846
53,459
Angelo Poerio retweeted
AI people discovering for the first time there are cheaper and more effective technologies than LLMs for specific tasks.
1
5
1
27
2,733
Curious about world models — what they are, how they work, and how you can create them? Anubhab Banerjee's deep dive patiently walks us through the process of building a CartPole-focused one in Python. towardsdatascience.com/how-t…
3
10
3,533
Angelo Poerio retweeted
10
113
10
1,073
16,724
Angelo Poerio retweeted
I used to be a C and JavaScript programmer, but now the only programming language I use is .md.
7
8
1
90
4,077
Angelo Poerio retweeted
Unix legends Brian Kernighan and Rob Pike wrote a regex matcher in just a few dozen lines of C it supports ^, $, ., and * the entire matcher works through recursion and pointer arithmetic
38
195
11
1,392
58,981
Angelo Poerio retweeted
111
330
57
4,488
165,250