Follow the change ·
What happened in AI and robotics — and the change behind it. A reading path through 32 stories from BIG CHANGE.
Ordered using RADAR attention, story opens and likes. Excerpts from our published reporting; open a story for the full context and sources. Dates in UTC.
OpenAI moves side-effect guidance from Root to User authority.
Explicit user or developer instructions can override this default.
Higher-priority rules and the agreed scope of autonomy still apply.
The revision sets an intended standard, not a measured outcome.
Reuters reports that US officials asked how Nvidia’s proposed Hugging Face purchase could affect competition in AI.
Hugging Face helps developers find and share AI models and datasets. Ownership of that distribution route can affect competition if model makers depend on it to reach users and those users depend on it to find alternatives.
The Hub is one part of a developer’s workflow; computing infrastructure is another. A model obtained from a repository can run elsewhere, so access to models and choice of hardware are related but distinct decisions.
Nvidia says Hugging Face will remain open and developers will not have to use Nvidia compute. That promise addresses both the platform and hardware choice, while the terms of access and support would matter to developers if the deal closes.
Alphabet reports Q3 Search revenue and query growth.
Search revenue grew at a slower annual rate; the results do not explain why.
Advertisers need placement-level results to assess performance. A search alone is neither a paid impression nor a sale.
More searches need not mean more visits to publishers. Independent research found different click patterns on pages with AI summaries, although the comparison cannot establish cause.
Kindo Sift learns from agent verdicts and samples events it would withhold.
An organization can place a decision before an agent run, making the choice to skip work a measurable part of the workflow. That changes the cost question from the price of each run to the number and kinds of runs avoided.
A saved run has an opportunity cost if it would have revealed something worth investigating. The useful comparison is therefore not savings alone: it is savings alongside an estimate of what the filter withheld incorrectly.
That estimate has a reference point. Agreement with an agent can show how closely the filter follows its judgments, while the correctness of those judgments requires separate evidence.
Opera’s WebAI preview offers browser APIs for local models, with opt-in setup and hardware and platform limits.
Opera’s browser exposes local AI to websites and extensions running on visitors’ computers.
A shared interface can reduce the need to build a separate model connection for every task, but it does not make the answers of different models interchangeable.
An application still needs a clear response when a capability is missing, and must explain if an alternative sends information to a server.
An offline model adds speaker-attributed transcription.
A transcript can serve different purposes: a verbatim record preserves hesitations, while a readable version can omit them. A team needs to choose the form its readers require.
Speaker labels help people follow a conversation, but do not establish a person's real-world identity. Check attributed speech against the recording and names against reliable records.
A benchmark compares output with a defined reference. Its score is useful evidence for that test, while the accuracy of a particular meeting depends on the audio and task.
Amazon's guidance weighs AI activity alongside longer-term outcomes.
More activity on a dashboard does not necessarily mean a team has delivered more value. A useful measurement system needs to show what happened after code was produced, including the quality of the result and the effort needed to get there.
A measure of activity can direct attention to a pattern that needs investigation. It takes additional evidence to determine what that pattern means for the work customers receive.
An account of a task can give managers a way to examine the work itself. Other research shows why teams also need to agree on what they mean by productivity and quality before drawing conclusions from their measures.
Oracle and Cerebras plan a 2027 OCI inference offering; its models, price and performance remain unspecified.
What changed: If delivered, the service would add another hardware approach for OCI developers to evaluate against their application requirements.
Why it matters: Faster generation can shorten a workflow’s wait across successive model calls. Network and tool delays still contribute to the total.
What to watch: The buying decision turns on supported models and regions, price, access controls and performance under a customer's request sizes and concurrency. Existing OCI modes do not establish its terms.
