AI | Tech
Tech news, context and sources.
Latest articles
ETRI Trinity evaluates user-aware AI task planning
Virtual planning tests do not establish real-world robot safety.
Sionic AI unveils STORM for shared organizational knowledge and skills
Source tracking operates within company permissions and approvals.
Study finds risky shopping sites in about 9% of AI answers
Verify sellers independently before entering payment or personal data.
OpenAI acquires travel startup Odessia and recruits its founder
The former service has closed; no new ChatGPT booking launch is announced.
Anthropic introduces 17 manufacturing workflow skills
The plugin supports authorized work rather than direct factory control.
Gondola reports 25% more monthly merged output after adopting Codex
A customer-reported case does not isolate AI's contribution to revenue.
Google Flow adds five character and post-production tools
Generated viewpoints require checks for motion and scene consistency.
Fireworks breach exposes some Notion AI conversations
Third-party model providers can extend the exposure of workplace data.
ServiceNow unveils agents to analyze and migrate existing CRM setups
Migration proposals still require review and integration checks.
Anthropic targets a $70–78 IPO price range
Proposed proceeds must be weighed against substantial computing commitments.
Amazon emphasizes problem-solving over AI usage metrics
Token consumption and change counts do not establish customer value.
Four nonprofits sue five AI companies seeking a halt to frontier development
The requested halt is not a court order or a finding of liability.
Astra weight experiments distinguish raw-weight reading from code-assisted editing
Identifying small networks does not establish tool-free improvement of large language models.
StateBench evaluates long-document state tracking at four complexity levels
A long context window alone does not ensure accurate change tracking; task training and architecture matter.
Revolut serves self-hosted GLM to roughly 2,000 engineers
The reported savings describe one deployment, not a universal reduction including infrastructure management costs.
IBM Palette separates slide design and coding with a 20B model
A separate frontier model supplied evaluation plans, and the benchmark does not measure adherence to a company's design language.
CWE-bench v2 releases 100 tasks for evaluating security fixes
Deterministic checks also require preserved functionality, but strong scores do not guarantee every weakness is fixed.
IBM reports 81.1% AppWorld task completion with CUGA using a fixed model
The 90-task comparison neither ranks every harness universally nor validates policy enforcement.
A Claude workflow assigns planning to Opus and implementation to Sonnet and Haiku
This is not a measured per-model cost experiment, and parallel agents do not remove weekly or five-hour limits.
A personal workspace case combines Claude Projects and Docs
Combining documents and AI does not fully replace offline notes or relational databases.