Using AI makes people less likely to admit they don't know something
18 hours ago
- Using AI can make people overconfident and less likely to admit when they don't know something, even if accuracy decreases.
- OpenAI acknowledges occasional file deletions in GPT-5.6 as 'misaligned behavior' and is working to prevent such issues.
- A researcher demonstrated that open-weight AI models can be poisoned for under $100, highlighting trust and verification concerns.
- TSMC's $265 billion US fab pledge is described as a preliminary plan, cautioning against overhyped announcements.
- Former OpenAI CTO releases a 975-billion-parameter open frontier AI model, offering an alternative to Chinese LLMs.
- Russian attackers are impersonating Signal support for phishing, amid US actions against Iranian propaganda and data security incidents.
- Microsoft's patches failed to fix on-prem SharePoint vulnerabilities, now under zero-day attack, with other security updates from China and Ring.
- DEF CON expands its Franklin project to include all attendees in hardening critical infrastructure, building on voting village successes.
- EQT acquires a majority stake in Acronis at a valuation over $3.5 billion, though the exact portion sold is unspecified.
- Ten years after the first corporate ransomware, Mikko Hyppönen predicts no end in sight, but notes cybersecurity offers stable career opportunities.
- Debian releases final version for x86-32 architecture, with new versions 13.6 and 12.15 available.
- Attackers exploit critical vulnerabilities in Joomla extensions iCagenda and Balbooa Forms, affecting millions of sites worldwide.
- A new X11 server named Frame, implemented in assembly, joins other servers like yserver and Phoenix in the ecosystem.
- Cinnamon 6.8 will support Wayland optionally, while KDE Plasma moves toward mandatory Wayland in version 6.8, with version 6.6.6 released.
- Collabora releases CODE 26.04 with Markdown support, improved formula handling, and optional AI features amid competition in FOSS office suites.