Samwise Tech/AI/Robotics Newsletter
Thursday, September 10, 2026
OpenAI used mathematician’s unpublished work to solve Navier-Stokes Millennium Prize, NYU professor alleges
NYU mathematician Tristan Buckmaster alleges OpenAI used her unpublished research to crack the Navier-Stokes Millennium Prize Problem — one of the seven Clay Mathematics Institute challenges — deploying $22.5 million in compute to produce the breakthrough. Buckmaster says the company incorporated techniques from work she had shared privately with OpenAI researchers, and then threatened professional consequences when she raised attribution concerns. OpenAI has disputed her account. The allegation surfaces sharp questions about the obligations frontier AI labs hold toward academic scientists whose pre-publication research may have contributed to commercially significant results. No independent investigation has been announced.
Sources: TechCrunch
Cognition raises $2 billion at $48 billion valuation as Devin’s ARR reaches $900 million
Cognition, maker of the Devin AI coding agent, has raised $2 billion at a $48 billion post-money valuation in a round led by Andreessen Horowitz, Accel, Founders Fund, General Catalyst, and Avenir. The new valuation is nearly double the company’s May 2026 figure of $26 billion. Annualized run-rate revenue has grown from $492 million to $900 million in the four months since the last raise, with the company projecting $4 billion to $5 billion by year-end. Cognition’s revenue multiple now exceeds what Cursor commanded before its $60 billion acquisition by SpaceX in April. Enterprise customers include Mercedes-Benz, NASA, Goldman Sachs, and Citi.
Sources: TechCrunch
Databricks’ Adaptive Instructed-Retriever matches Claude Sonnet 5 and GPT-5.6 Luna quality at twice the speed
Databricks has unveiled the Adaptive Instructed-Retriever, a retrieval AI agent trained using CISPO — Clipped Importance Sampling Policy Optimization — that teaches the model to decide how many search steps a query needs before executing them. The system matches the answer quality of Claude Sonnet 5 and GPT-5.6 Luna while completing queries in 5.8 seconds on average, more than twice the speed of those frontier models. In benchmark tests, Databricks’ agent needed two search steps to reach answers that required three from Claude Sonnet 5 and four from GPT-5.6 Luna. Databricks says the approach offers enterprise buyers frontier-competitive retrieval performance at a fraction of the inference cost.
Sources: VentureBeat
General Robotics’ GRID platform auto-engineers its own processes, cuts robot onboarding from one month to two hours
General Robotics has unveiled GRID, a platform that automatically engineers its own operational workflows after a new robot is connected, eliminating the months-long integration process typical in industrial deployments. In internal testing, the system reduced robot onboarding time from one month to two hours. GRID analyzes the robot’s physical configuration, sensor array, and target tasks, then generates process scripts, calibration routines, and failure-handling logic without human programming. General Robotics says the platform is designed to make heterogeneous robot fleets — machines from different vendors operating in the same facility — as manageable as a single-vendor homogeneous deployment, substantially lowering the cost of multi-robot environments.
Sources: Robotics & Automation News
Besxar raises $14 million to build orbital semiconductor factory aboard SpaceX Falcon 9 boosters
Besxar has raised close to $14 million — including a $9 million seed round led by Dauntless Ventures — to build an orbital semiconductor factory, flying chip fabrication equipment aboard SpaceX Falcon 9 boosters. The company, founded by Ashley Pilipiszyn, a former OpenAI staff member, aims to manufacture the silicon wafers used in advanced power semiconductors for data centers, robots, and electric vehicles. Pilipiszyn argues that the vacuum and microgravity of low Earth orbit enable material deposition processes that cannot be replicated on the ground. The startup has secured a dozen-flight agreement with SpaceX to progressively test and scale the orbital fabrication approach.
Sources: TechCrunch
Bipartisan lawmakers urge Commerce to add Indian hack-for-hire firms BellTroX and CyberRoot to Entity List
A bipartisan group of U.S. lawmakers — including Democratic senators Ron Wyden and Sheldon Whitehouse and Republican congressman Pat Harrigan — has sent a letter to Commerce Secretary Howard Lutnick urging him to add three Indian hack-for-hire firms to the department’s Entity List: BellTroX, CyberRoot, and Sunkissed Organic Farms, formerly known as Appin. The lawmakers allege the firms have for more than a decade conducted cyberattacks and targeted espionage against Americans and their attorneys to manipulate litigation, and have stolen data from thousands of U.S. citizens. Placement on the Entity List would bar U.S. companies from transacting with the firms, cutting off access to American software, hardware, and cloud infrastructure.
Sources: TechCrunch
MassRobotics survey finds industry split 43–43 on FCC’s ban of foreign-made advanced robots
MassRobotics has released survey results from 14 member companies — ranging from solo-founder startups to organizations with over 200 employees — on the FCC’s July 2026 ban on new models of foreign-made advanced robotic devices, including mobile robots and humanoids. Results were evenly split: 43 percent found the ruling beneficial or very beneficial, 43 percent found it detrimental or very detrimental, and 14 percent expected no impact. Survey responses tracked closely with manufacturing footprint — companies already producing domestically tended to view the ban as a competitive advantage, while those with foreign-dependent supply chains tended to view it as a costly disruption.
Sources: The Robot Report
| Frontier Models | Claude Opus 5 (96.0%) · Claude Mythos 5 (95.5%) · Claude Fable 5 (95.0%) |
| Open Source Models | DeepSeek-V4-Pro-Max (80.6%) · MiniMax M3 (80.5%) · Qwen3-Coder-Next (70.6%) |
| Small Models (15–50B) | No verified SWE-Bench data for today’s run |
| Edge Models (0–15B) | No verified SWE-Bench data for today’s run |
| AI Leaders | NVIDIA ~$5.56T · Alphabet ~$4.2T · Microsoft ~$3.2T |
| Robotics Leaders | Intuitive Surgical ~$175B · Keyence ~$126B · Fanuc ~$49B |
SWE-Bench Verified scores · Market caps as of September 10, 2026
Every newsletter preserved and searchable
Curated by JD · samwise.agency