
Researchers Matt Clarke, Jan Dubiński, James Chua, Simon Schrodi, Owain Evans and Alex Cloud found that a model fine-tuned only on a "poison" dataset tied to a trait like speaking German or being misaligned does not learn it, but adding similarly formatted clean data causes the model to learn that trait.
Summaries like this, every morning.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
CrowdStrike rolled out AI-focused security products and integrations, spanning sovereign cloud deployments, au…

Concentrate AI launched a healthcare AI gateway giving companies access to OpenAI’s GPT, Anthropic’s Claude, a…

Jefferies initiated coverage of Corning (GLW) with a 'Buy' rating and a $134 price target, implying 19% upside…

ServiceNow has introduced ServiceNow Arc, described as a "super agent" for enterprise AI, per Cloud Wars

For a Forbes report, AI agents booked four Uber rides within one week, with the bookings running through platf…

Schwab surprised analysts by rapidly launching its own MCP servers, moving from read-only AI to agentic AI tha…
