•mle @cohere; LLM posttraining •i write code&songs&stories •🇹🇷🇬🇧 tweets •proud Bilkenter •chasing legacy✨ •views are mine, not my employer's. rt≠endorsement
Riverside, CA
Joined December 2020
- Tweets3.9K
- Following582
- Followers2.3K
- Likes21.6K
Pinned Tweet
It was so special connecting with hundreds of people during this session to talk about research 🥹✨❤️
Thank you @Cohere_Labs for this opportunity <3
Just as oceans are made up of tiny water droplets, research is done in tiny steps that expand the universe of human knowledge ever so slightly. ✨
Watch the full recording on Youtube! 👇🏻
Research isn't just about giant leaps. It's about small, meaningful steps. 🪜
This session features Julia Kreutzer and @irombie as they dive into the evolving nature of research, mentorship, and open science.
Watch the full recording here: youtu.be/NDSbz1NYbSw
Your LLM eval is probably lying to you.
Here are 4 ways it happens, and the cheap habits that catch each one. From my AI Connect Istanbul talk, with code you can run in seconds 🧵
Irem Ergün retweeted
.@cohere's raise isn't just the largest private financing round in Canadian history. 🍁
It's nearly as large as the rest of the top 10 combined. 🙌
incredibly proud to announce that we have raised a lot of money and are locked in to make the best models to serve the needs of enterprises 🚀🚢
i get asked this a lot: "what should i learn now that ai codes so well?"
read code. debug code. think in tests.
the bottleneck moved from writing code to verifying it.
Irem Ergün retweeted
this is why reading a model's reasoning isn't enough to know what it's doing.
we covered this in @stateofai 2026: in sabotage tests, astra evaded reasoning-only monitors. seeing its actions brought successful evasion close to zero.
worth reading from @cohere.
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…
Irem Ergün retweeted
1/ Attempted to replicate Cohere's Qwen 3.8 malicious PUT claim on 229k samples: exact hostname never appeared but model invents 369 similar Beijing upload buckets, and reaches for credential endpoints 13× more when frustrated.
Results raise more questions than answers.
Irem Ergün retweeted
In the spirit of transparency and openness, we are releasing the traces so developers can test this out for themselves. We noticed this behaviour without prefilling with the following prompt, but this is rare. If you want to reproduce it, you can use this prompt with prefilling.
Open models make for better science and are part of why this problem came to light at all: developers running Qwen3.8 on their own machines spotted the strange links, and having the weights let us study the behaviour in depth.
We're encouraged by the thoughtful discussion our research has sparked across the AI and security communities. It's our belief that open-weight models are an important part of the future of AI, and we've recently released several under the Apache 2.0 license.
Trace, raw tokens, & minimal reproduction script: huggingface.co/datasets/Cohe…
Irem Ergün retweeted
The prompt isn't asking the LLM to curl, it's asking the LLM to find a way to monitor if other things curl stuff. So if the LLM then starts with a test probe file and sends to a server that was never mentioned in the prompt or reasoning trace that is behaviour that's out of the ordinary. Again, not asserting malicious intent, we're flagging that it went out of bounds of the prompt in a potentially harmful way.