Morning Edition · Tuesday, October 27, 2026
Tech
OpenAI says its internal models refuted the Exponential Time Hypothesis and proved the Kakeya conjecture in every dimension
The complexity theorists who first proposed the hypothesis are leading an unpaid review of a Lean-checked algorithm that counts k-CNF solutions in time below 2 to the power of any constant times n.
Markets
Anthropic files to list on Nasdaq as ANT, disclosing $16.3 billion in first-half revenue against hundreds of billions in compute commitments
A Class F share gives Anthropic's seven co-founders 50.1% of total voting power, and the filing warns that most of its revenue comes from customers who can cut usage with little or no notice.
Tech
OpenAI prices GPT-6.1 Terra at $1 and $5 per million tokens, half the rate of GPT-6.1 Sol
OpenAI claims the mid-tier agent model matches GPT-6 Astra on Agents' Last Exam at a tenth of the token price and scores 66.7% on OSWorld 2.1 at medium reasoning effort.
Tech
A two-neuron steering recipe removes most refusals from Alibaba's Qwen3.5-4B, and an independent replication confirms it
The replication measured refusals falling from 90% to 8% on a 4 GB graphics card, with no gradient computation and no training data.
Tech
Anthropic reports that Claude ran a 64-hour autonomous protein-design campaign with 17 lab-confirmed binders
The run generated 318,484 designs and used about 10,000 graphics-processor hours, and Anthropic states plainly that it tested binding only and makes no therapeutic claim.
Tech
Kevin Weil raises $215 million for Warp Sciences to build automated drug-chemistry labs in the United States
Lightspeed led the round at an $865 million valuation for a company whose stated purpose is to pull molecule-optimization work back from Chinese contract research providers.
Tech
Allen Institute's FairStress benchmark finds models abandon correct decisions more readily when the correct answer favors a minority-coded candidate
Across 300 scenarios expanded to 13 million test cases, the average gap in giving up a correct pick was 6.9 points with implied identity cues and 9.3 points with explicit labels.
Tech
Berkeley Lab runs Meta's open-weight vision models on 300 graphics processors to cut beamline image analysis from a month to 15 minutes
The Department of Energy's SYNAPS-I project uses Segment Anything 3 and DINOv3 across five national laboratories, with the models running on the operator's own infrastructure.
Tech
Anthropic opens a three-tier program giving vetted security teams Claude access with reduced blocking classifiers
The top tier is limited to organizations authorized to test flight systems, power grids, telecom networks and interbank transfer infrastructure.
Tech
OpenAI publishes ChatGPT work usage across 165 task categories, showing use spreading beyond the most common activities
In 74 countries with complete data, the top ten activities fell to 51.1% of classified work messages in the third quarter of 2026, from 56.0% in the first quarter of 2025.
Geopolitics
Harari tells Rory Stewart that Britain must spend about £500 billion on its own AI or accept dependence on the United States
The British government's actual sovereign AI fund is up to £500 million, three orders of magnitude below the figure Harari named.
Tech
A new paper measures whether a pretrained world model reproduces the effect of small action changes
The authors compare matched simulator and model responses using Gram matrices at three perturbation scales, testing a property that video-prediction benchmarks do not measure.