Today's Hallucination HQOpenAI Checks Whether Its Own Model Could Hack You (Preliminary Findings: Maybe)OpenAI has published cybersecurity evaluations for its Astra model, assessing how dangerous it could be in the wrong hands — a question one might argue should precede public deployment rather than follow it. The evaluations measure "critical cyber capabilities," a phrase that sounds thrilling until you realise it means "can this thing help someone break into infrastructure." OpenAI says the findings are preliminary and safeguards are being strengthened. Reassuring, if you enjoy reassurance. Source: OpenAI Blog
Scientists Invent 16 New Viruses. On Purpose.American researchers have used AI to generate 16 viruses that do not exist in nature, marking what is either a tremendous leap for medicine or an absolutely spectacular error in judgment, depending on how the next decade goes. The hope is accelerated vaccine and antiviral development. The concern, which need not be spelled out but will be anyway, is that the same method now demonstrably works, and not everyone with a computer has good intentions. Progress, as ever, arrives with its own disclaimer. Source: Al Jazeera
Google's AI Dream Team Has a New Dream, ElsewhereJeff Dean — the engineer so synonymous with Google that he practically has a commemorative plaque in the data centre — has left the company, joining a broader reshuffle of senior AI talent. The Verge reports Google's models are currently trailing those from Anthropic and OpenAI, which is a bit like McDonald's falling behind on burgers. Google isn't collapsing, but when your legendary figures start updating their LinkedIn, it does suggest the canteen conversation has changed. Source: The Verge
Oracle Tells OpenJDK: No AI Code Allowed. Also, Oracle Loves AI Code.Oracle has banned AI-generated code from OpenJDK, the open-source Java project it stewards — a policy that sits in magnificent tension with Larry Ellison's recent boast that Oracle itself no longer writes its own code, because AI does it. To be fair, "rules for thee, not for me" is a governance tradition with deep roots. The ban reportedly stems from copyright and licensing concerns around AI training data, which is a legitimate worry, expressed by a company that has sued over software licensing with the enthusiasm of a hobby. Source: Hacker News
The Man Who Made Robots Sound Cool Thinks AI Music Is a Bit SadThomas Bangalter — one half of Daft Punk, the duo that spent years behind helmets and robot costumes to make a point about humanity — has called out AI as a shortcut that strips music of surprise and genuine feeling. This is the man who used vocoders and drum machines to produce some of the most emotionally affecting electronic music ever made, which rather defeats the argument that technology is the enemy. His actual point is subtler: that the struggle to create is where the meaning lives. He's not wrong. Source: Consequence
The robots are writing the code, designing the viruses, and making the music. The humans are writing the disclaimers.
|