doorss

Anthropic published an initial update on Project Glasswing, a research initiative. The post appeared on HN and details early findings from the project.

Andrej Karpathy announced on Twitter that he has joined Anthropic. The move was widely discussed on Hacker News.

Anthropic released a paper outlining two scenarios for US-China AI competition by 2028, arguing it is essential that the US and its allies maintain their lead over authoritarian governments in AI development.

Anthropic published research on natural language autoencoders that convert Claude's internal reasoning representations into human-readable text, offering a new approach to AI interpretability.

Microsoft disclosed that AI agents currently cost more to operate than human employees for many tasks, raising questions about the near-term economics of AI deployment at scale.

A post cataloging twelve common misconceptions about AI-assisted coding, covering issues from overestimating capabilities to underestimating the maintenance burden of AI-generated code.

Intuit announced over 3,000 layoffs as part of a strategic shift to refocus the company on AI-driven products and services.

The FTC settled with Cox Media Group and two other firms for nearly $1 million over charges they deceived customers about an "Active Listening" AI-powered marketing service that claimed to use device microphones for ad targeting.

Google announced new advertising formats being tested in Search results alongside an expansion of its Direct Offers pilot program, further blending ads into search experiences.

Suomessa on annettu väärennettyjen alastonkuvien jakamisesta tiettävästi vain yksi tuomio. Lainoppineet kertovat keinoista puuttua deepfake-pornoon.

A post examining the distinctive em-dash usage patterns in LLM-generated text and how these typographic tells can serve as markers for AI-written content.

A site addressing the growing problem of people pasting large blocks of AI-generated text into online conversations, degrading the quality of discourse.

Security researchers disclosed Megalodon, a technique for mass-backdooring GitHub repositories by exploiting CI workflow configurations to inject malicious code.

CISA is working to contain a data leak while lawmakers demand answers about the incident, as reported by Krebs on Security.

Apple published a detailed blueprint on how they use formal verification to prove correctness of their corecrypto library, a core component of Apple's security infrastructure.

Maahanmuutto Suomeen väheni viime vuonna ja samaan aikaan maastamuuttojen määrä oli tilastohistorian suurin.

Helsinki edistää fossiilisten polttoaineiden mainoskieltoa kaupungin kilpailuttamilla mainospaikoilla, vaikka päätös veisi osan 100 miljoonan euron mainostuloista.

Yle seurasi kuukausien ajan miestä, jonka epäillään väärentäneen tekoälyllä alastonkuvia lähipiirinsä naisista pornosivuille. Uhrit eivät tienneet kuvista.