TheSuperoptimal

Frontier Research

Models, research, safety and the ideas pushing the frontier of AI

The 2017 paper that quietly became the blueprint for every chatbot

Transformers

The 2017 paper that quietly became the blueprint for every chatbot

Eight Google researchers threw out the machinery everyone thought language models needed, and the architecture that replaced it now underwrites ChatGPT, Gemini and almost everything else

8 min read

How DeepMind's AlphaFold cracked a problem biology chased for 50 years

Computational Biology

How DeepMind's AlphaFold cracked a problem biology chased for 50 years

Given only a protein's chain of amino acids, AlphaFold predicts the shape it folds into with an accuracy that rivals years of laboratory work, and it has since done so for nearly every protein known to science

8 min read

The model that learned new tasks from the instructions in its prompt

Scaling

The model that learned new tasks from the instructions in its prompt

By making a language model a hundred times larger, OpenAI found it could pick up a task from a few examples typed into the prompt, with no retraining, a trick that turned one model into a thousand tools

7 min read