Ethan Mollick Says Cyborgs Are Starting to Dominate Math Research

Latent Space highlights Goodfire’s Nick Cammarata’s working theory that neural networks are already largely interpretable, with further agentic analysis resolving many familiar examples of polysemanticity and superposition. Cammarata argues that most neurons are legible, contrasting with roughly 2% in his 2023 OpenAI effort. The roundup also reports objections concerning weak activations, neuron simulation and neglected non-maximal activations, alongside a reminder that the paper itself is not yet available.

The final day of TechCrunch Disrupt 2026 brings sessions on AI innovation, startup growth, fundraising and emerging technologies, alongside Startup Battlefield events and networking at Moscone West in San Francisco. More than 10,000 founders, investors and technology leaders are gathering, and the Expo Hall features 320+ exhibiting startups. Day 3 also includes the Startup Battlefield 2026 final and the announcement of its winner.

Doubao Phone Assistant says it will gradually open third-party app GUI automation after the 30-day public comment period for its Screen Automation Protocol ends, but only for apps that have not refused. Its published refusal list names 151 apps, including WeChat, DeepSeek, Taobao, Kuaishou, and Meituan. Users can still install, log into, and operate those apps manually on the listed phones.

Modal crossed $1B in annualized run rate and raised a $750M Series D led by Accel at a $15.75B valuation, according to the company announcement cited by Latent Space. Usage grew more than 15x year over year. Modal’s Erik says Sandboxes alone reached a $300M run rate, with close to 3 billion sandboxes launched and support for a million concurrent sandboxes across uses including background agents and reinforcement learning.

Brandon Anderson interviews Leash Bio’s Ian Quigley about generating reproducible protein–small molecule interaction data using DNA Encoded Libraries and detecting misleading signals in experiments and models. They discuss a Kaggle competition with about 4B measurements where nobody beat chance on an unseen chemical library, and an agent sandbox searching 6.5M molecules on a budget of 100. Anderson reports that the sandbox has matured into Minos, with a preprint, blog, and open dataset being released today.

Google Research presented 10 updates at Research@ spanning geospatial intelligence, scientific agents, document verification, edge hardware, numerical models, healthcare, disaster forecasting, education, and collaborative prototyping. It is expanding early access to experimental Google Earth AI capabilities and opening Deep Review to the general public after an early version reviewed over 16,000 research publications. Other updates include an upcoming TimesFM and TabFM integration into Gemini for business and an AMIE study involving 98 patients, in which clinicians reported that AI summaries helped them prepare in 75% of cases.

Epic Games announced a preview of its redesigned Epic Games Launcher for Windows, Mac, and Linux beginning today, describing it as faster and more responsive. Epic executive Martin Keely told The Verge that agents wrote and reviewed all the code under the development team’s supervision, with AI helping replicate existing functionality and support Linux. The old launcher remains available, and Epic has no specific timeline for requiring migration.

Google released Gemini 4 Argon globally for coding, knowledge work, enterprise workflows and cyber defense, with access through the Gemini API and several Google products. Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with cached input discounted 95%; Artificial Analysis reports rates double after November 30. At AI Engineer NYC, about 20,000 people joined keynotes on AI and finance, including a discussion of benchmarking models like Argon by Vals AI’s CEO.

Google announced that users can build personal Android apps with Google AI Studio and run them privately through Google Play using a standard Google account. The apps remain accessible only to their creators, must comply with existing Play policies, and will initially be available to U.S. users age 18 and up. Google says sharing with trusted friends will follow in the months ahead, with a path to public publication after trust and safety reviews.

Shing-Tung Yau and leading Chinese mathematicians launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence on Sunday in Beijing, according to the South China Morning Post. It unites eight institutions and aims to connect over 1,000 researchers, with plans for specialized models, machine-verifiable proofs, and transparent AI reasoning. Meanwhile, 25 Fields Medal winners, including Terence Tao and Peter Scholze, worry that AI-generated proofs erode mathematical understanding and encourage dependence on tools researchers cannot follow.

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full, its second-generation model family for personal and professional agentic work. Latent Space reports that Wonder is available through the Meta Model API and Muse Code, while Full is limited to Muse Code subscribers. Artificial Analysis gives Wonder Full and Wonder Intelligence Index v4.3 scores of 56 and 52, respectively, versus Opus 5.5’s 58; Wonder supports a 1M context window and text, image, video, and audio inputs, with text-only output.

AI labs are buying defunct startups’ code and workplace archives through companies such as SimpleClosure and Sunset to train models and build environments where AI agents practice tasks. SimpleClosure’s Asset Hub has handled over 150 deals, while Google bought Spirit Airlines’ decades of data for $10 million this past August. The market raises questions about employee privacy, consent, and pricing, and no company has released tests comparing training results from real-world archives with alternatives.

Meta released Muse Wonder and the more capable Wonder Full AI model, which the company says performed near Fable 5.1, Opus 5.5 and GPT-6 Astra on several capability tests, and slightly better on some agent and knowledge-work benchmarks. Meta highlighted Active Memory, which retains relevant background across conversations, and Active Focus, which keeps working toward complex goals and adapts when new information or obstacles arise.

Andrej Karpathy describes four tiers of LLM exposure, arguing that public perceptions lag behind capabilities experienced by frontier-system users. He estimates roughly six billion people have barely encountered LLMs, one to two billion use casual chatbot products, 20 million use frontier models professionally in math and code, and 5,000 access internal frontier systems. He claims those internal systems coordinate thousands of agents on projects that would take experts years, while human review falls behind, and attributes progress in coding and math to reinforcement learning on verifiable rewards and commercial training priorities.

Import AI 476 covers Humanity’s Sixth Sense, a benchmark from Elorian and Scale AI that tests intuitive visual reasoning. People scored 93.1 aggregate pass@1, while GPT 6 Astra scored 53.6; GLM-5.3-flash and Kimi K3 scored 26.8 and 25.5. The newsletter also examines how AI could accelerate space development and reports on TasteVal, a benchmark of AI systems’ research taste.

Claude banned 11.4 million accounts in the first half of 2026, according to an Anthropic report published in July that was recently recirculated. Of 398,000 appeals submitted, 42,000 were successful. Anthropic did not disclose a full breakdown of the reasons for bans, but the report listed several types of activity that may trigger them.

Rather suddenly, cyborgs are starting to dominate math research.