AI breakthroughs in math are becoming the benchmark that matters most
Dr_Singularity · x · 2026-08-04
The author says the metric they care about most is the number of AI-generated breakthroughs in mathematics, and eventually in physics, chemistry, and biology.
The attached image lists a set of major results in math and theoretical CS, including:
- improved sphere packing bounds
- better binary and spherical code bounds
- an explicit non-sofic group
- a disproof of Connes’s rigidity conjecture
- stronger lower bounds for arithmetic circuit complexity
- exponential quantum parallel repetition
- hardness for closest vector problems
- a proof related to Ehrhart’s volume conjecture
- new bounds on multicolor Ramsey numbers
- counterexamples in extremal graph theory
The post frames these kinds of results as a benchmark for how far AI can contribute to core science.
More from AGI Musings
- Jagged AI progress makes a sudden AGI arrival seem less likely — aparnadhinak · 2026-08-04
- GPT-Live’s simultaneous speech could make robots feel human sooner than we think — VraserX · 2026-08-04
- Creator says AI is not a bubble, but friends largely disagree — lunwang1996 · 2026-08-04
- AI policy should borrow crisis engineering and risk-management playbooks — joshua_saxe · 2026-08-04
- Open Source is the True Bedrock of AI Safety, Argues Researcher — rbhar90 · 2026-08-04
- Essay argues LLMs reward expertise more than casual use — MaxMussio · 2026-08-04