The LLM Revolution (so far)

According to a comprehensive LessWrong survey, large language models have transformed from basic limitations to remarkable breakthroughs across nearly every domain. Four years ago, AI art couldn't render human hands correctly—now it generates accurate QR codes and passes visual Turing tests. Google Translate handles casual conversation fluently, powering automatic Japanese-to-English translation on Twitter to connect speakers across cultures. Customer service bots field calls in real time, though they still fall for clever discount tricks. In mathematics, LLMs have solved multiple long-standing problems, with one cracking a famous open problem in graph theory within a day of the model's public release this July. Competitive coding saw its watershed moment: for the first time, AI decisively beat human programmers in a major competition. Security researchers using advanced models are finding software vulnerabilities twenty times more efficiently than before. AI also outperforms crowd wisdom at prediction and, surprisingly, proved more persuasive than expert humans—even when those experts researched extensively and received cash incentives. The pattern is unmistakable: LLMs are advancing across the board, and their impact reaches far beyond research labs.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton