AI News
Internal material reviewed by Reuters described large staffing reductions, disruptive agent actions and more technical and security incidents.
Meta explored cutting some teams by as much as sixty percent while moving daily work to AI agents, then canceled part of the plan amid uncertain productivity gains and operational problems. Internal material reviewed by Reuters described large staffing reductions, disruptive agent actions and more technical and security incidents. Meta confirmed that teams conducted scenario planning for restructuring and cuts. The reported Project OT scenarios included reductions of up to sixty percent in some teams.
Why it matters: The episode supplies rare internal evidence that automation targets, measured output and real product delivery can diverge sharply when agents receive broad authority.
Read full articleAI News
Gemini 3.5 Transcribe removes verbal fillers and corrections, but its polished output can differ from the speaker's exact words.
Google announced a speech-to-text model for Gboard, Gemini on macOS, developer tools and eventually Chrome, emphasizing speed, multilingual support and automatic cleanup. Gemini 3.5 Transcribe removes verbal fillers and corrections, but its polished output can differ from the speaker's exact words. Google calls the model Gemini 3.5 Transcribe. The company reported about seventy percent faster processing than Chirp 3.
Why it matters: The product blurs transcription and editing, which makes convenience gains inseparable from the need to preserve verbatim records in legal, medical or journalistic settings.
Read full articleAI News
The open-weight family includes three-, eight- and thirty-billion-parameter versions with a native 128,000-token context window.
IBM expanded its Granite model line with downloadable reasoning-focused models aimed at predictable self-hosting, including agent training in the two larger variants. The open-weight family includes three-, eight- and thirty-billion-parameter versions with a native 128,000-token context window. Granite 4.2 is available in three-, eight- and thirty-billion-parameter variants. All three models have a 128,000-token context window.
Why it matters: Smaller self-hosted models widen the set of organizations that can trade frontier capability for lower variable cost, tighter data control and more predictable deployment.
Read full articleAI News
Employment for ages 22 to 25 in highly exposed occupations was nineteen percent below comparable less-exposed fields, up from a thirteen-percent gap last year.
Updated payroll research linked the strongest employment divergence to entry-level hiring in automation-heavy occupations, while economy-wide effects remained much smaller. Employment for ages 22 to 25 in highly exposed occupations was nineteen percent below comparable less-exposed fields, up from a thirteen-percent gap last year. The reported young-worker employment gap widened from thirteen to nineteen percent. The analysis focused on workers ages twenty-two to twenty-five.
Why it matters: The revised evidence points to a quieter labor-market channel: fewer first jobs rather than a wave of visible dismissals, with long-term implications for training and experience pipelines.
Read full article