News for hackers. Stuff that matters.
Re:The tide is shifting: 1.3B outperforms 7B Llama 2 by __vec__ on Tuesday September 12, 2023 @10:55AM (#37482249)
Textbooks Are All You Need II: phi-1.5 technical report

"Perhaps achieving ChatGPT’s level of capability at the one billion parameters scale is actually achievable?"

Re:The tide is shifting: 1.3B outperforms 7B Llama 2 by refulgentis on Tuesday September 12, 2023 @12:49PM (#37484060)
On a long enough timeline, Id say yes.

Right now they're comparing to the 7B Llama 2, which is a shadow of 65B llama2, which is a couple steps off ChatGPT, which is a shadow of GPT4.

I'm still comfy saying yes because there's no reason to doubt it will follow the same logic as "eventually $phone will have the same FLOPS as $desktop"