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…

La redondance n'exclut pas plus de redondance elle même n'exclut pas le niveau d'après jusqu'à avoir des problèmes de système distribués; et meme dans ces cas là ça exige de la redondance, du backup, de la fiabilité du test , une bonne architecture quoi. aws.amazon.com/blogs/industr…

To mark the 10th anniversary of the Google SRE book, we are excited to announce a completely rewritten second edition that reflects a decade of lessons and offers fresh guidance for today's operational challenges. Check it out → goo.gle/4zv9do9

J'espère que c'est la huitième génération de détournement de fonds publics.

Supporté la douleur pour ce qu'on aime c'est la base en tant qu3 homme. Donc l'avènement du risque est consequencielement donc un non événement .

Leading to more serving, leading to more demand, running on more infrastructure leading to more demand, ect...

I guess now that software is solved, this should not be an issue.

J'irai jusqu'à 1000.

Sincèrement, vue l'état du marché, il vaut largement les 200M.

Déjà ?? Lol.

Après c'est une externalités naturelle des LLMs. Pour rappel meta a un projet qui parle vod langues maternelles. ai.meta.com/research/publica…

Just posted the next entry in the fast-date series! This time, a faster way to convert a timestamp ➜ Hour, Min, Sec. benjoffe.com/fast-time-of-da…

Et surtout le départ de quelqu'un.

C'est contre l'esprit de la tech aka le " progrès ". Je suis curieux de savoir ce qui te fait dire ça le safety ?

De toute façon si construire un bon produit était devenu facile ou "gratuit", nous serions tous au chômage.

On a engagé des bagares a fifa14 pour moins que ça .

As Jigga said recently, you're checking the wrong list champion; look way up. It's not about mere nominations.

Announcing: the @a16z Ops Engineering Fellowship ⚙️ Because: operations (along with everything else??) is becoming something best approached with an engineering mindset. We've been tracking this space closely, and are seeing a new kind of builder emerging inside companies... They connect internal systems, build agents, and redesign workflows across finance, sales, support, and everything between. @levie calls them "Internally deployed FDEs".. others say "AI Operations", "Agent Engineer"... I don't think anyone's agreed on the title yet, but there's something interesting going on - which to me means it's a good time to get the people who are at the frontier .. applying an engineering mindset to re-thinking how companies operate in an AI-native way - together. So: We're launching the a16z Ops Engineering Fellowship: bringing ~50 internally deployed engineers and AI-pilled operators to together over 6 weeks in San Francisco and NYC (+ group chats). There's a killer crew involved already - Claire Vo (@clairevo) is joining as a Founding Fellow, as is Sebastien Goddijn (@sebgoddijn) - who built Glass at Ramp; Owen Williams (@ow) at Stripe; Ann Miura-Ko (@annimaniac) - deep in the trenches of AI-pilling operators via Floodgate; Brandon Gell (@bran_don_gell) - COO at Every; TK Kong (@tkkong) - who helped start Ramp Labs; Matt P (@MattnPortnov) - leading internal AI at Decagon; and several more we can't include in the announcement (but watch this space 👀)... For now: Asking X 1- Who is leading "Ops Engineering" (or whatever you call it) at your company? 2- Who do you respect in this space, who should we know about, absolutely must have join the cohort and/or join us for a dinner ? Tag below, DM me - 3- or apply below ! 👇👇

This one is gold even if you are working already.