QwQ: Alibaba's O1-like reasoning LLM

Chinas people have done many praiseworthy things throughout history. The ccp doesn’t deserve any reflected glory from that.

No one should be so naive as to think that a party that is so fearful of free thought, that it would rather massacre its next generation of leaders and hose off their remains into the gutter, would not stoop to manipulating people’s thoughts with a new generation of technology.

25% of Nvidia revenue comes from the tiny country of Singapore. You think Nvidia is asking why? (Answer: they aren’t)

> Doom Slayer needs to teleport from Phobos to Deimos. He has his pet bunny, his pet cacodemon, and a UAC scientist who tagged along. The Doom Slayer can only teleport with one of them at a time. But if he leaves the bunny and the cacodemon together alone, the bunny will eat the cacodemon. And if he leaves the cacodemon and the scientist alone, the cacodemon will eat the scientist. How should the Doom Slayer get himself and all his companions safely to Deimos?

You'd think this is easy since it is obviously a variation of the classic river crossing puzzle with only the characters substituted, which they can normally solve just fine. But something about this - presumably the part where the bunny eats the cacodemon - seriously trips all the models up. To date, the only one that I have seen consistently solve this is GPT-4 and GPT-o1. GPT-4 can even solve it without CoT, which is impressive. All other models - Claude, Opus, Gemini, the largest LLaMA, Mistral etc - end up tripping themselves even if you explicitly tell them to do CoT. Worse yet, if you keep pointing out the errors in their solution, or even just ask them to verify it themselves, they'll just keep going around in circles.

This model is the first one other than GPT-4 that actually managed to solve this puzzle for me. That said, it can sometimes take it a very long time to arrive to the right conclusion, because it basically just keeps trying to analyze the possible combinations and backtracking. Even so, I think this is very impressive, because the only reason why it can solve it this way is because it can reliably catch itself making a mistake after writing it out - all the other LLMs I've tried, even if you explicitly tell them to double-check their own output on every step, will often hallucinate that the output was correct even when it clearly wasn't. The other thing about QwQ that I haven't seen elsewhere is that it is better at keeping track of those errors that it has acknowledged, which seems to prevent it from going around in circles in this puzzle.

You're looking for "authoritarian" or "dictatorial".