I came here to ask this too... If only there was an ASI somewhere nearby to answer this.

Some very good and satisfactory answers here. I’m thankful.

Thanks, I’m guessing reinforcement learning is not as useful in advancing LLMs.

Thank you for your input, I was amazed by DeepMind’s previous achievements, namely AlphaGo, zero and most distinctly AlphaFold! (I have been a very long time Folding@home contributor; Seeing AlphaFold launch was mind blowing for me!)

I am pretty hopeful that they would do more amazing things, it just felt to me that they are withholding their horses in this LLM race.

Came here to post this.