today's digest covers major announcements from google i/o 2026, including gemini spark and antigravity, alongside fresh research on model adaptation and optimization. we also see new tools for data anonymization and audio generation, plus funding news for ai security and desktop automation.
- google i/o 2026: gemini spark and antigravity - google introduces a personal ai agent and a closed-source cli, raising security questions about closed-source tools.
- irisgo builds ai desktop buddy to automate office tasks - a proactive desktop agent learns workflows to automate repetitive tasks, backed by andrew ng.
- hellora adapts moe models by targeting hot experts - fine-tuning mixture-of-experts models by adding low-rank adapters only to the most active experts cuts parameters and compute while boosting performance.
- stability ai releases audio model for six-minute songs - stability audio 3.0 generates professional music up to six minutes long, with open weights for smaller versions.
- ocean raises $28m to fight ai-powered phishing - an agentic email security startup uses a custom small language model to detect ai-generated phishing attacks.
- anonymize production data with mimesis - the mimesis python library replaces sensitive personal data with realistic fake data for safe data science work.
other highlights include research on learning lagrangian multipliers for mixed integer linear programming with guarantees, and a new conformal prediction framework using beta distributions. nanoco raised $12 million after rejecting a $20 million buyout for its secure openclaw alternative. google deepmind's co-scientist also helped speed up the hunt for anti-aging genes.