AI — Page 8

The San Francisco lab's first System One model drops text generation, while its headline speed and cost gains remain vendor-tested.

The model used 20 hours of simulated driving data, but Odyssey's broad physical-intelligence claims still rest on its own evaluations.

An Extended Thinking variant reasons and calls tools in the background, using spoken updates to cover the wait.

Yiping Goh and Carlos Banon are betting structured solids and BIM context can move generative design beyond polished renderings.

The beta can read and update CRM records under existing permissions, pushing Dario Amodei's assistant deeper into revenue teams.

Cloudflare's new setting publishes a no-training preference while keeping Applebot, Googlebot and Bingbot available for search under its Accountable designation.

Gensyn says open-1b includes a model, its training recipe and proof intended to let third parties verify stages of the training process.

Nazli Azimi and Hani Goodarzi are using a 50-cell-line study to show the biological data behind Therna's RNA models.

The Hong Kong research team put its stateful adapter into PEFT's main branch, trading LoRA-style merging for a parallel network.

The calculator turns local inference enthusiasm into a payback schedule, then shows how quickly cheaper APIs can make the hardware case worse.

Former Obama press secretary Jay Carney will run Project Blueprint as AI safety, jobs and data-center costs move into the midterms.

Tests grew 10x and CI jobs 25x in six months, forcing a three-week rebuild of Anthropic's test-selection system.

Leaked guides show reviewers scoring real conversations after an imperfect privacy filter; the default-on pipeline can retain old chats even after users opt out.

Maxime De Greve's Octoglow sends a six-digit code between screens through a camera-readable animation inspired by Apple's Quick Start.

Portable Computer runs its harness and models on RTX PCs, adds local MCP and scheduled jobs, and can hand selected work to cloud models.

The Stanford robotics spinout says its model learns from wearable human demonstrations, without teleoperation or robot training data.

Mulyar's screenshots made the accused founders and startups identifiable; RuntimeWire cropped their names because the thread offered no transaction records or other corroborating evidence.

Lukas Petersson and Axel Backlund are turning Andon's internal agent stack into a waitlisted preview with revenue-share economics.

Elon Musk says the 2.5T run beat a 2.1T JAX-trained model despite mistakes corrected midway; a 3T successor is already planned.

Eric Simons' AI app builder is feeding anonymized coding sessions to Arcee AI for a trillion-parameter-class open model.

An unreleased iOS interface would let users link bank accounts and ask Claude about spending and plans, with key privacy details still unclear.

Founder Faraj Aalaei says ACI compresses months of front-end work into days; its public evidence remains company and customer benchmarks.

The 27-year-old pretraining researcher worked at OpenAI and Anthropic, then publicly accused both labs of racing toward self-improving systems.

The project reports 20.9% to 54.4% lower KL divergence at the same 2.50 GB size, with calibration text doing more work than bit allocation.