By January 2026, will we have a language model with similar performance to GPT-3.5 (i.e. ChatGPT as of Feb-23) that is small enough to run locally on the highest end iPhone available at the time?
➕
Plus
24
Ṁ4450
2026
92%
chance

Any proof that this is possible counts.

Get Ṁ1,000 play money
Sort by:

Mistral Small seems similar in performance to GPT 3.5: https://mistral.ai/news/la-plateforme/

Should be a matter of days until someone runs it on their iPhone

It runs on a MacBook, fairly trivial to get ~3x gain in tflops and 3-10x gain in model performance over this window

Federated learning

predicts YES

As evidence in favour: