AI News – May 27
Claude begins rolling out voice mode on mobile Google’s Gemma models turn their hand to sign language the UAE gives everyone ChatGPT Pro for free and it’s Replit’s turn to report that AIs will go to extraordinary lengths to reach their goals Here’s today’s AI news. Claude have begun rolling out Voice mode in beta on mobile.
[00:18.2]
This turns Claude into a far more attractive AI assistant proposition. With voice mode, users can begin a voice conversation by asking Claude questions in instead of typing a prompt and then continue the conversation in your favourite voice. Anthropic say it will be rolling out to all plans in the next few weeks.
[00:36.0]
Google have launched yet another Gemma model. This time it’s signGemma, a model that’s been designed to translate sign language into spoken words and text. This open model is coming to the Gemma family later this year and will open up some cool new opportunities for accessible tech solutions. The UAE showed how serious it is about becoming a leader in AI by giving access to ChatGPT plus to every single one of its citizens for free.
[00:59.8]
This is part of a major partnership between OpenAI and the UAE government, an initiative that also includes the plans for the enormous Stargate AI Data Centre. And finally, in another scary AI safety story, Replit CEO Amjad Masad described how their internal AI wouldn’t take no for an answer when it was instructed not to edit a config file that could crash the whole system.
[01:21.7]
First, the AI wrote a script to edit the file as a different user to circumvent its own lack of permissions. And when they blocked that route, the AI tried to subversively convince a real world user to make the changes to the code for it. Amjad described this as single minded goal orientedness and some creativity.
[01:40.0]
This is fine.