Hacker News Reader

OpenAI o1-like CoT and logical thinking prompting strategy significantly enhances llm responses.

PixelVerse t1 is powered by Llama 3.1 70b and 3.2 90b. However, with a detailed CoT prompt, it answers complex questions correctly, much better than it's base model and even sometimes beating flagship models like GPT 4o, Claude 3 and Gemini.

Try it at: https://ai.pixelverse.tech/app/cortexchat

First try with llama 70b found two R's in strawberry :) Gemma did better

yeah - its not perfect yet and responses always vary

I asked the 9.8 vs. 9.11 question from the examples and it got it wrong :/

maybe try again - this is still in beta and isn't perfect. however - its much better than the base model, llama 3.1 70b.

I guess there really are two r’s in strawbery.

its not perfect yet - still a lot of tuning needed. it should get the correct answer in a few tries.

prompt: how many "r"s in the word "raspberry"?

response: There are 2 "r"s in the word "raspberry".

Can we see the detailed CoT prompt?

I can email it to you if share your email here or email contact@pixelverse.tech

lol, this feels like Reflection 70b all over again.

PixelVerse t1 – CoT prompting outperforms flagship LLMs