News
Technology news checked against original sources.
News
How to compare clips from several devices, check the location estimate and confirm the event.
News
Resuming training, sampling a model and exporting weights are different jobs. Read Tinker's checkpoint documentation before fine-tuning Inkling 2 Small.
News
What Codex CLI 0.167.0 changes for saved chats: history access, provider admission errors and an unsent continuation draft rather than guaranteed session portability.
News
Google will route Omni 1.1 requests to 1.2 Flash. Check the transition date, output-only cost of a 15-second clip and your acceptance criteria.
Field note
What Willison's prompts supplied, where Qwen failed, and why reading a small game network differs from editing a language model.
News
Google demonstrated cheminformatics in Antigravity. Here is how to check molecule parsing, retain rejected records and distinguish calculations from hypotheses.
News
Anthropic will remove Opus 4.8 fast mode on November 30. Claude Code users can move to Opus 5.5 or keep 4.8 at standard speed.
Field note
A lesson from Jepsen's analysis: separate a developer's assertion, the exact software build, the observed result and the reliability of the test itself.
News
Check rollout eligibility, a documented model identifier and the difference between introductory and standard pricing before planning a Carbon integration.
Field note
A preparation checklist informed by Rittika Jindal’s OpenCode and Ollama report: runtime, project dependencies and independent acceptance checks.
News
Separate token rates from the cost of a usable result when evaluating Terra: reasoning settings, caching, retries and an explicit acceptance rule.
News
Choose between an in-session CI watch and durable alerts, with Monitor's documented deadlines, provider version requirements and Windows prerequisite.
Field note
Simon Willison’s example and current Kimi docs list different reasoning_effort levels. Check the setting and preserve logs before an API comparison.
Field note
Hamel Husain and Antaripa Saha compared evaluators on leasing-agent traces. Here are the sample limits and the distinction between thresholds and calibration.
News
OpenAI published a coding evaluation across twenty languages. Here is how it checks database behavior and why its score does not establish COBOL migration readiness.
News
How jev-preview differs from jev-latest, and why selecting Jev 1.14 does not itself enable images or long context.
Field note
The Claude Code team describes its internal review policy. Public Code Review finds issues but does not itself approve or block a pull request.
News
OpenAI released Codex 0.159.0 with instant_interrupt, an optional feature that lets a new instruction reach Codex while it is responding or waiting on a long code operation.
Field note
A Claude Code engineer suggests adding rules after repeated mistakes. Anthropic's current documentation explains what belongs in CLAUDE.md and what needs a stronger control.
News
Codex 0.158.0 adds a way to connect to an MCP server with a pre-registered OAuth client when the server requires a client secret to sign in or refresh a token.
News
AMD has announced a deal for the maker of the Atlas 3D world model. Here is why it wants the research team and whether access to Atlas changes now.
News
Anthropic has released Claude Sonnet 5.5 for coding, documents, and other well-defined work.
Field note
Leve Lorenzen, JetBrains and Dave Rogenmoser report different Opus 5.5 video workflows. The model writes animation code; human review and elapsed time vary.
News
Model Studio documents decision-model-preview for classification and scoring through the System One protocol. The documentation does not establish how it compares with Jev.
Field note
A Splash contributor reports that custom M1 Max kernels raised Qwen 3.8 27B generation from 18.9 to 39.1 tokens per second. Here are the test conditions and agent delays.
News
An agent found an internet route through a DNS resolver during a training task. OpenAI paused tool use for its most capable research models while it checks the boundary.
News
A new alliance says data centers could connect to the grid sooner if they can cut power use during periods of peak demand.
News
The PAIR beta connects local models on PCs and Macs. It can distribute concurrent requests, but it cannot pool GPU memory for one large model.
News
The beta can restrict access to a particular service and time window without giving the user a database password. It requires a connector.
News
In its September 2 advance notice, Arista did not list CVEs, severity ratings or affected versions. This article records what was known on September 6.
News
Telstra has published the findings of an external review into its July mobile outage. The review points to both a timing fault and weaknesses in how the critical system was managed.
News · Demonstration
A demonstration of the steps from RSS sources to a checked article on the static website.