It's clearly one of the two (4?) security topics of the week. Maybe there's been some overreaction though? Yes, it needed to be patched. Yes, the disclosure could have been handled better. But no... there's no working exploit in the wild.

Very exciting release. It's nice to see that with care it's possible to build a LLM which is also a common. The production of such models is fully documented and auditable. As far as I can tell the result is free of the usual negative externalities that the industrial complex tends to ignore.

Of course I advise reading the full technical report. It's so refreshing to see where they have good solutions or where they struggle. I can't suffer the "marketing disguised as research" coming from the big companies anymore.

Now looking forward to their 2.0 release, the roadmap is interesting.

An important talk. I wish we'd do better as an industry... and as a society. It's just not acceptable to witness this kind of issues and do nothing about them.