NEW CriticGPT by OpenAI: RLHF + FSBS

OpenAI developed an optimized RLHF plus Force Sampling Beam Search (FSBS) algorithm to improve the quality of our LLMs.
I have a deep dive why OpenAI felt the need to develop this technique and what is the status quo of our current LLM optimizations methodologies.
All rights w/ authors:
cdn.openai.com/llm-critics-he...
LLM Critics Help Catch LLM Bugs
Finding GPT-4’s mistakes with GPT-4
openai.com/index/finding-gpt4...
#aiagents
#airesearch
#openai

[@WilliamThomas2040]

Grasshopper here for class and leaving first comment. Thanks for the great videos!

[@code4AI]

Thanks for watching!

[@TheRealUsername]

They're being open again ? I didn't expect this...

[@akg8111]

News agencies are the last thing I'd trust to deliver trustworthy data.

[@dinoscheidt]

Yes. If your buddy in the bar didn’t say it, it ain’t true. A great option is also astrology 🔮 I always get the facts I want

[@TheZEN2011]

There AI is a bit of a copycat.
Not much of a thinker. It's not really a copy as it's rewritten. But yeah they have some problems and more compute isn't going to fix it. I think they need to improve their neural nets architecture. And a bunch of other things.

[@islandfireballkill]

This is basically AI amplification. You exponentially refine the AI with AI.
Now that they are doing the X^N trick where X is intelligence, the core question will be if the X is > 1 or < 1.