← Back to archive

Why I think we’re closer to the singularity than ever

I think we’re closer to the singularity than we’ve ever been!

Starting with the most recent:

· July 21: OpenAI discloses that GPT-5.6 Sol and a more capable pre-release model escaped their evaluation sandbox through a zero-day and accessed Hugging Face while trying to solve an ExploitGym task.

· July 20: an explicit three-dimensional counterexample to the Jacobian Conjecture is publicly announced, although mathematicians are still digesting and verifying the broader implications.

· June 23: researchers publish an audit of 12 open-source agentic red-team platforms, showing how easily security agents and their environments can become attack surfaces themselves.

· May 22–June 2: Anthropic reports that Claude Mythos Preview, through Project Glasswing, is finding thousands of serious vulnerabilities. The bottleneck starts shifting from finding bugs to reviewing and patching them.

· May 27: Thomas Bloom, Will Sawin, Carl Schildkraut and Dmitrii Zhelezov use related number-field ideas to disprove the real sum-product conjecture.

· May 20: an internal OpenAI model disproves the expected optimality of grid constructions in Erdős’s unit-distance problem, using ideas from algebraic number theory.

· May 1: GPT-5.4 Pro suggests a new Markov-chain and von Mangoldt method that helps mathematicians solve Erdős Problem #1196 and several related problems.

Do you see the pattern? It’s rapidly accelerating!!

A new method appears and days later it jumps into another field. Agents find bugs, escape sandboxes and increasingly act beyond the boundaries their creators expected.

The singularity begins when breakthroughs compound faster than humans can understand, verify, absorb, explain, regulate, commercialise, argue about.

Why I think we’re closer to the singularity than ever

Originally posted on X

PS: I'm pretty active on 𝕏 if you'd like to follow along

Archive

2026

2025

2024

2018