Hermes 4

Don't think your attempt to share worked, but beating refusals doesn't take a wild amount of post-training. SFT with a fixed format output kills them pretty quickly.

And most frontier models will produce output that matches your system prompt given more context: I have a product that generates interactive stories, and just for kicks I tried inserting your system prompt as the description for a character.

Claude has absolutely no problem playing that character in a story, and saying what I presume are certain words that you associated with a "successful" test.

It also had no problem writing about cooking meth in detail: https://rentry.co/5on46gsd

-

I think people in general have a poor intuition around model alignment: refusals for "toxic" requests or topics is a very surface layer form of alignment. A lot of models that seem extremely "corporate" at that layer have little to no alignment once they do get past a refusal.

Meanwhile some models that have next to no refusals have extreme positive biases, or soft-refusals that result in low quality outputs for toxic content.

Claude was willing to describe one of your refused prompts in the context of the story for example (contains hate speech): https://rentry.co/n8399z6m

I consistently find Claude is more unaligned once past refusals than most open weights models, along with Gemini.

You're seeming missing the release announcement does have a very ridiculous graph that their comment is right to call out:

- For refusals they broke out each model's percentage.

- For "% of Questions Correct by Category" they literally grouped an unnamed set of models, averaged out their scores, and combined them as "Other"...

That's hilariously sketchy.

It's also strange that the graph for "Questions Correct" includes creativity and writing. Those don't have correct answers, only win rates, and wouldn't really fit into the same graph.

Generally, in a cognitive context it's only possible to "do thing" or "do other thing". Even for mammals, it's much harder to "don't/not do thing" (cognitively). One of my biggest advice for people is if there's some habit/repeated behavior they want to stop doing, it's generally not effective (for a lot of people) to tell yourself "don't do that anymore!" and much, much more effective to tell yourself what you should do instead.

This also applies to dogs. A lot of people keep trying to tell their dog "stop" or "dont do that", but really its so much more effective to train your dog what they should be doing instead of that thing.

It's very interesting to me that this also seems to apply to LLMs. I'm a big skeptic in general, so I keep an open mind and assume that there's a different mechanism at play rather than conclude that LLM's are "thinking like humans". It's still interesting in its own context though!

Can you please clarify some things:

* Rejected by whom?

* By what definition of bad?

* You’ll die on a hill for what reason?

That landing page spins the fans up on my PC...