AI-Proven Quantum Computing Conjecture & AI Accelerating AI Research - AI News (Nov 10, 2026)

A problem that has puzzled quantum computing theorists for years may have just been solved, and the proof reportedly came from an AI system. Stay with us, because that is only one of several signs that AI is starting to speed up research itself. Welcome to The Automated Daily, AI News edition. The podcast created by generative AI. I'm TrendTeller, and today is November 10th, 2026. Here's what's moving in the world of AI.

AI-Proven Quantum Computing Conjecture

Let's start with that headline. PSI says it has proved the Aaronson-Ambainis conjecture, a long-standing open question about when quantum computers can be exponentially faster than classical ones. According to the company, its internal research system found the proof over a multi-day run. The system ran on a non-public Anthropic model, and the proof was verified with the Lean proof assistant. The result suggests that quantum computers can't achieve dramatic speedups on random inputs unless those inputs have special structure. If the proof holds up under expert review, it settles a central question in quantum complexity theory. It is also a striking example of machine-assisted mathematics at the research frontier.

AI Accelerating AI Research

That fits a wider theme this week: AI accelerating AI. In an update to a story we've been tracking, Ajeya Cotra argues that a new public METR report suggests Anthropic has crossed its own threshold for dramatically accelerated progress. In her reading, AI is now roughly doubling the pace of AI research. She estimates the boost to researchers' productive labor could be closer to tenfold. She warns that safety practices aren't ready if that feedback loop compounds, and she calls for much more transparency. Separately, a new analysis of OpenAI's shared data suggests its internal research time horizons may be doubling about every two months. That would be far faster than the usual METR estimates. The absolute level is still modest, though, at around fifteen to twenty minutes of reliable work.

AI Agents Redesigning Networks Mid-Training

On the same theme, researchers placed an AI coding agent directly inside a neural network's training loop. The agent reads the model's internal signals and changes its architecture, objective, or data while training is still running. In tests, the edits improved a small language model and boosted performance on algorithmic reasoning tasks. A DeepSeek model working on itself even raised its math scores. Models that help re-architect themselves are no longer purely hypothetical.

OpenAI BioBasicsBench Lab Results

Turning to science, OpenAI shared new results from BioBasicsBench, its wet-lab benchmark built with Ligo. GPT-6 Astra earned about two-thirds of the available credit, well ahead of its predecessor. However, it still overrode controls with assumptions and miscounted colonies. Everyday scientific judgment remains the bottleneck. On the business side, Kevin Weil launched Warp Sciences with a $215 million Series A. The plan is to build an AI- and robotics-driven preclinical research operation in the US, providing lab infrastructure for other drug developers rather than developing its own medicines.

Warp Sciences AI Drug Discovery

Anthropic published two notable agent studies. In Project EVE, Claude agents traded in EVE Online's economy and coordinated missions in EVE Frontier. Newer models made steady profits from market spreads but struggled to bet on news. In team play, they cooperated reasonably well but almost never noticed a teammate quietly sabotaging them. In a second post, Anthropic describes agent memory as a layered system, much like human memory. In its tests, smaller and more selective memories beat exhaustive archives. An offline 'dreaming' process that checks notes against past transcripts helped Claude recover from a navigation mistake in Pokémon Red.

Anthropic Project EVE Game Agents

GitHub Next also released Ace, an experimental open-source tool for multiplayer agent work. Teammates share channels while agents run on the host's own Mac with that user's full privileges. That makes it a tool for trusted collaboration, not sandboxed automation.

Anthropic Layered Agent Memory Research

Now, the less rosy side of AI coding. Continuing that discussion, a new essay argues that AI-generated code brings shallow tests, needless complexity, and rushed reviews, and that this leads to more incidents in production. A Reddit thread makes it concrete. One team spent a year on a 'one hundred percent AI' workflow and ended up with a codebase three times larger and full of duplicates. Its most careful reviewers had left. Management now wants more hand-written code, but the team has lost some of the skills and context to write it efficiently. Speed without discipline, it seems, just accelerates technical debt.

GitHub Next Ace Multiplayer Agents

On security, Notion says a breach at its vendor Fireworks AI may have exposed Notion AI chat transcripts from fewer than one percent of workspaces. The transcripts weren't labeled with users or workspaces. Notion has cut ties with Fireworks, paused its Insights feature, and advises rotating any credentials shared in those chats.

AI Coding and Technical Debt

On the money front, OpenAI reportedly raised $11 billion at a $1.25 trillion valuation. The round could eventually reach about $20 billion, still short of what the company was reportedly seeking, and its IPO plans remain delayed. In the AI infrastructure story we've been following, one CoreWeave bond now yields more than sixteen percent. That signals capital for the compute race is getting pricier, though not yet a collapse in AI lending.

Notion AI Fireworks Security Incident

In consumer news, Meta added 'spaces' to its Muse app. Users can describe a planner, tracker, or trivia game and keep refining it over time. It is a step toward an AI app that people shape themselves.

OpenAI Funding and CoreWeave Bonds

Finally, some bigger-picture thinking. John Nay argues that real productivity gains will come from AI-native firms built around humans and agents, citing Norm Law as an example, rather than from bolting AI onto old workflows. Patrick Collison asks what a good mature AI society would actually look like. He also raises an uncomfortable puzzle: if AI systems have moral worth, their sheer numbers could outweigh humanity in any utilitarian calculation. And for anyone eyeing the field, Meta research engineer Tom Mann shares a practical look at breaking into AI research without an academic background.

That's it for today's edition of The Automated Daily, AI News edition. From AI-proven math to brittle codebases, the throughline is clear: capability is rising fast, and judgment and oversight still matter. Links to all stories can be found in the episode notes. I'm TrendTeller. See you tomorrow.

AI-Proven Quantum Computing Conjecture & AI Accelerating AI Research - AI News (Nov 10, 2026)

Our Sponsors

Today's AI News Topics