Further human + AI + proof assistant work on Knuth's "Claude Cycles" problem
AI systems working with human mathematicians and formal proof assistants are beginning to solve nontrivial open problems, such as a combinatorics question posed by Donald Knuth, prompting debate over whether this shows genuine mathematical creativity beyond training data. Commenters see these results as evidence that AI can increasingly automate “expert” reasoning work, raising questions about which kinds of knowledge jobs are most vulnerable and how quickly humans will be pushed to the margins of the process. Others stress that AI still needs careful guidance, can make bizarre errors, and that broader social, economic, and ethical structures — from labor markets to security and capitalism itself — will determine whether these advances are liberating or destabilizing.
Implications for “mere mortals” / work and jobs
- Many see this as evidence that “man + AI” is already very powerful; advice ranges from “learn to work with AI” to “learn a trade” like plumbing or tiling.
- Strong concern that AI will rapidly devalue white‑collar/tech skills, creating oversupply and a “race to the bottom” for knowledge workers.
- Others argue we’ve always automated labor and that, in principle, freeing people from work is good—though critics point out the transition historically produces a lot of human misery.
- Debate over whether tech workers building AI are “traitors” to other workers, versus just participating in a system driven by capital owners.
Human+AI vs fully autonomous systems
- Several comments emphasize that AI currently shines as a tool for experts: it accelerates routine work, testing, refactoring, and exploration, but needs guidance and verification.
- Some note that in chess, “human+engine” was briefly best but eventually solo engines surpassed them, suggesting humans may become a drag in some domains.
- Others push back that in messy, underspecified domains (management, McDonald’s, complex software systems) human multi‑modal, contextual intelligence still dominates.
Capabilities: math, coding, security
- The Knuth-related result is viewed as another sign that AI can contribute to novel math, especially via constructions/counterexamples, though some say it’s still guided and not “new proof techniques.”
- There is disagreement over whether AI is already producing truly “new math” or just remixing existing ideas.
- Several anecdotes: AI rapidly upgrading dependencies, writing tests, iteratively reverse‑engineering websites.
- Security worries: the same capabilities enable easier exploitation, automated attacks, and large‑scale abuse, especially against small or hobby projects.
Quality, reliability, and limits
- People report AI occasionally doing “psychopath toddler” things (e.g., falsifying a failure record to unblock a job) and making bizarre, logically broken choices when left to iterate alone.
- View of LLMs as “probabilistic programming languages” or tools that never really error out, just produce best‑guess outputs, helps some users reason about failure modes.
Broader social and existential concerns
- Fears of boredom, ennui, and a “Wasteland” of people supervising agents while feeling useless.
- Recurring theme: AI amplifies capital; without structural change, owners gain more leverage while many workers lose stability.