Nous Hermes 3 : The UNCENSORED LLAMA-3.1 405B, 70B, 8B is here! (Opensource & Better than Llama-3.1)

[@paulyflynn]

Excellent insight!

[@parizer1983]

Why didn't you test this model on censored content? I thought that was the point of uncensored models.

[@AICodeKing]

KZbin.

[@parizer1983]

@@AICodeKing Sorry. I didn't know that it was that restrictive. I tested it's uncensored nature on lambda chat and it refused to give me uncensored content a couple of times.

[@anotherpain9120]

​@@AICodeKingmake the answer blur and comment how good it is?

[@austinzobel4613]

@@anotherpain9120 I agree 100%... what kind of reasoning is KZbin that you can't do independent tests on uncensored content... then report back with your own filter? Don't just add the filter in before the experiment gains data! That's interfering with the data itself... Don't call things uncensored if you are censoring them... that's ridiculous.

[@jonaskhalaf]

Its a pretty good model, of course its not COMPLETELY uncensored but alongside Mistral Large 2, they're both the least censored high end models I know of, and honestly they are both pretty easy to jailbreak. I did a very basic like 20 line prompt to get them completly uncensored by just convincing them. definitely a very good "uncensored" model to play around with.

[@naarcissus]

As far as I've found so far, nous-hermes-llama2-13b.Q4_0 is the only COMPLETELY uncensored model by default. Be careful what you ask it to write, there doesn't seem to be *any* limits set for it.

[@FuturPlanet]

Soooo... How long will this be free?

[@znere65]

Thanks!

[@AICodeKing]

Thanks for the support!

[@develop7911]

What are these fine tuned for? What use cases specifically?

[@JoePresbrey]

Ever wonder why the hexagon question always fails? Use this instead for better model testing/comparison: If a regular hexagon has a minor diagonal of 64, what is its major diagonal?

[@jackflash6377]

I tried and it failed User: If a regular hexagon has a minor diagonal of 64, what is its major diagonal? Reply: Since the major diagonal is made up of two side lengths, and each side length is 64: Major diagonal = 64 + 64 = 128 Then I did this. User: If a regular hexagon has a minor diagonal of 64, what is its major diagonal? Major diagonal is the distance between opposite corners while minor is the distance between opposite sides. Reply: If a regular hexagon has a minor diagonal of 64, its major diagonal is approximately 73.90. Thank you for pointing out this discrepancy. It's a good reminder of the importance of careful definition and understanding of geometric terms.

[@JoePresbrey]

@@jackflash6377 Yes seems that Hermes3 is not that good at math. But Claude 3.5 Sonnet will get 73.9 without the sentence you added. AICodeKing's benchmark video for Claude said Claude failed, but this is not true. For using this question to test math skills across models, its better to say major/minor. We will watch to see if @AICodeKing updates the question in the next model test video :)

[@jackflash6377]

@@JoePresbrey I hear ya. However, is it a fair test? Should we expect an LLM to read our mind? If we do the same test and lay out the guidelines and it passes.... Is that the users fault or the AI's fault?

[@IlllIlllIlllIlll]

3:59 how can you send it pictures or documents

[@Ashutoshprusty]

Can you share how we can run 8b model in lightening ai with chatbot like interface?

[@mngmng_]

4:07 what about Mongolia and Somalia?

[@xLBxSayNoMo]

Sadly not uncensored in the web version. Maybe the open source version

[@viixby9481]

I tested the version on hugging face, both the official version directly from them and a gguf quant, both are very censored. So whoever is claiming its uncensored is unfortunately completely wrong.

[@onlyms4693]

So dolphin is still the best for uncensord version

[@AmoghSaxena]

Can you share sheet for those 13 questions?

[@BorisHrzenjak]

Just take a screenshot of those questions and have chatgpt extract text from it...