AI agents burn 10B tokens on 60-year-old deletion channel problem, near 1/100-bit capacity
DimitrisPapail · x · 2026-09-16
Researcher Dimitris Papail reports that he recently tasked AI agents with the deletion channel capacity problem, open since the 1960s. The agents didn't fully solve it, but after over 10 billion tokens of compute combining old and new techniques, they bounded the capacity to within 1/100 of a bit.
Retweaker tomasortega predicts a barrage of follow-up results in the style of Papail's BDC bounds, suggesting AI-assisted math research is entering a rapid-results phase.
More from AGI Musings
- "Responsibility, not rationing": a market-based counter to Dario's Pacing the Frontier — emax · 2026-09-16
- Against 'Tao Should Stop Complaining': A Philosophical Pushback on Utopian Superintelligence — akbirthko · 2026-09-16
- Gary Marcus slams AI doomers for downplaying climate risk — GaryMarcus · 2026-09-16
- Scott Aaronson publishes a new post and the AI community takes notice — john__allard · 2026-09-16
- Foundation funds OpenADMET: open datasets and blind competitions to crack drug ADMET prediction — iskander · 2026-09-16
- tszzl: P(doom) Talk Is High Status on Twitter, Not a Strong Anti-Signal — tszzl · 2026-09-16