Interesting, maybe caused by design focused on overprotection against distillation of the thinking process, so “it’s a feature, not a bug” mentality. I don’t use gemini models so I can’t confirm by myself, thanks for pointing that out!

What is this bug you described? First time hearing about it… Mistral seems weirdly low, it’s a very big model, maybe they didn’t finetune it for benchmarks?

Big boys tested Big Cloud LLMs for this and it doesn’t look good for mental health right now. Tiny models - probably even more sycophantic.

This is huge news for universal Strata-like engine in native llama.cpp - if i uderstand it correctly. Good job, for everybody and especially for people with low VRAM. VERY GOOD NEWS!!!

Skip the marketing, paper at: https://www.nejm.org/doi/abs/10.1056/NEJMc2606815 (although i can’t access it right now, and it isn’t available on pirate sites)

I’m thinking right now - maybe it is not a coincidence that: (1) just as popular blogposts started to suggest that we should investigate open weights models; (2) the Cloud BigBoys started to advertise they can perform “brain surgery” on smaller models (manually adjust weights). Maybe it’s just marketing, nevertheless investigation of open weights models can only benefit people insterested in the idea, hope it will not scare the general public too much though.

It sounds like fearmongering (so hot right now), but hey: that is the true benefit of opensource! - it can be diagnosed and monitored. Claude just refuses to answer biology homework question, because of “security” (fuck Anthropic for that). Overall I agree that vigilance is needed, but still it should not make people dismiss open weight/open source models - if anything it should convice people to invest (their time and money) into open weight models EVEN MORE as they can be further finetuned / distilled / pruned or otherwise modified and run locally with any prompt you would like with any behaviour you train it to perform (i.e. you CAN make your own model out of it, even if it is anti-“insert your identity here”).

The final points of this article are worth implementing regardless of your stance, i’ll cite them here:

• Start from a vetted model. Where a hardened or otherwise safety-improved version exists, deploy that rather than the stock weights.
• Know what you are running. Track the provenance of every model, and pin a version you have evaluated.
• Evaluate on your own work and agenda. Leaderboards use neutral prompts, so they miss behavior that only appears for a particular customer or a particular kind of task. Test under the identities and topics that matter to you.
• Keep humans on the critical paths. Code that touches authentication, access control or user data needs an owner, plus the review and security testing you would give any outside contributor.
• Fix what you find in the weights. A prompt instruction can’t remove a behavior the model learned, as Finding 3 shows. [Finding 3: Asking for secure code doesn’t fix it. Fixing the weights does]

Let the people play what they want, even if it is slop! And the stealing/inspiring part - i get it, it is frustrating and not fair, but not the AI is the problem here, it’s just a tool - the problem is with the copycat’s reasoning (that is to make a quick buck and don’t give a damn about broader consequences for the future). If the game is a shameless copy it already is prone to be copyright claimable (although i don’t believe in legal batles, especially for small developers or not wealthy individuals that clearly are disadvantaged from the start), if it is not - let the Poeple decide. Of course it poses a threat of oversaturating the market - i don’t care, it’s not new - Steam has been oversaturated with non-AI-porn-game-slop for years now and nothing bad happened.

Qwen3.5-4B-3D-Coder (finetune) with video example seems kinda useful, and quite impressive for its size! I haven’t tested it personally though.