NASA-funded researchers found an amoeba in Lassen Volcanic National Park that reproduces at 145°F, a record for complex cells. Researchers caution this does not show complex life can survive on Mars.
https://t.co/w0RG6QPJJy
Senator Van Hollen calls OpenAI models the most efficient hackers ever. Our claim check finds that published tests show major cyber gains but do not establish that all-time ranking.
https://t.co/8pJwgN5EU8
A preprint demonstrates a soft crawler coordinated by pneumatic modules without a central digital controller. The externally powered robot moved just 25 millimeters over six cycles in a 95-minute lab test.
https://t.co/q2SYpQa0kh
An Oxford preprint reports error correction across two trapped-ion processors linked in one lab. The active test protected a known Bell state, not arbitrary quantum information.
https://t.co/AhrdnthruF
AWS Lambda can checkpoint Pydantic AI agent runs and resume recorded work after interruptions. External tool effects can still repeat, so developers need idempotency safeguards.
https://t.co/nNMWGle2pF
NASA awarded Blue Origin a Mars communications contract worth up to about $700 million. The agency targets operations at Mars by 2030; coverage and reliability still need verification.
https://t.co/s6sbLLiQFh
A preprint proposes an external Bayesian filter that tracks uncertainty while an LLM selects actions. The evaluation used fewer methods and trials than planned and does not establish general reliability gains.
https://t.co/u9KfE4Saow
In an unreviewed preprint, researchers report 70.8% ± 0.4% conditional success for a four-photon logical Bell measurement. Photon generation and transmission losses remain outside that rate; the experiment used only two logical qubits.
https://t.co/asAKEPaflC
NASA awarded $650,000 on September 24, 2026 to five Mars food system designs, with $300,000 to Chinyere Ukeje's Adaptive Nourishment Infrastructure. All winners are design-phase concepts, untested at the benchmark scale.
https://t.co/oLABFJaKeb
Hurricane Polo gained 90 knots in 24 hours to reach Category 5 on September 22, 2026, over 32C water off Mexico. Analysts cited by NASA call it the basin's fastest depression-to-Category-5 rise on record, though not NHC certified.
https://t.co/eDUfUtpjky
Palantir's Pipeline Builder now lets teams test LLM nodes without affecting production pipelines. Passing a suite does not establish production reliability.
https://t.co/0oinb7Qabn
In a preprint, Warsaw researchers report cooling potassium-41 and cesium atoms to 10 to 15 microkelvin. The team still needs to reduce atom loss before attempting a full resonance survey.
https://t.co/TPH8PfWfRD
A new preprint reports 147 moonlit ISS detections from a 2025 telescope campaign in Sydney. The results support longer optical tracking windows, but round-the-clock coverage remains modeled.
https://t.co/xKyOIqGf8o
OpenAI's September 22, 2026 framework proposes four priority areas and five principles for independent scrutiny of its safety claims. As the analysis notes, it is voluntary and self-authored, naming no assessors and offering no enforcement mechanism.
https://t.co/OFkG5DLboa
GitHub added shell tools to Copilot code review and multiple agents to Lite reviews. Its internal results measure addressed comments, not proven defect detection; developers still need human review.
https://t.co/Lib9zlUaee
Researchers report using Wigner Time to control two cold-atom setups for over two years. The open-source Python tool remains in beta, with known silent-failure risks that laboratories must test before adoption.
https://t.co/vNPxb9TkoM
Hubinger says current-model risk is low; his >10% extinction estimate concerns future superintelligence. The posts give no quantitative derivation for that personal forecast.
https://t.co/iXCcEiTilt
Visual Studio 18.10 adds a provider-neutral AI agent workflow alongside pull-request links and Podman debugging. The IDE release is stable, but the new agent remains an early preview.
https://t.co/8NL0VJcztD