AI Weekly

King (Canute)’s Speech?

The new UK government announced plans for an AI bill which will “seek to establish the appropriate legislation to place requirements on those working to develop the most powerful artificial intelligence models.”

As yet, there is little detail beyond the governing Labour Party’s manifesto commitment to “ ensure the safe development and use of AI models by introducing binding regulation on the handful of companies developing the most powerful AI models”.

But as the tech world watches to see whether the EU’s attempt at regulating AI will have any impact, the UK‘s efforts are valiant but will they have any effect at all on the global multi-billion dollar race towards artificial general intelligence (AGI)?

Level up?

OpenAI have internally released a series of levels of AI cognitive ability on the route to AGI. The company places its current models in level one (Chatbots, AI with conversational language) but on the cusp of level two (Reasoners, human-level problem solving). The as-yet-unreached levels three to five are: 3) Agents, systems that can take actions, 4) Innovators, AI that can aid in invention and 5) AI that can do the work of an organization.

This comes as details of the tech giant’s secretive Strawberry project leaked to Reuters, thought to be a renaming of the Q* project which is thought to have abilities to solve advanced mathematical and scientific problems.

Make AI Great Again?

As Donald Trump starts to pull ahead of Joe Biden in the polls, the Washington Post reveals that aides to the former president are drafting an executive order on AI that would roll back some of the current administration’s attempted safety measures in an attempt to “Make America First in AI”, with a focus on military technology and protecting American AI models from foreign interference.

With Silicon Valley figures such as Elon Musk endorsing and donating large sums to a PAC supportive of the 45th President, sceptics might note that America’s AI pioneers could have a lot to gain from a Republican win in November.

You(Tube) sing it and I’ll play it

YouTube is rolling out “sound search”, a feature which will allow users to hum a song and find it on the platform’s music library. Plans have also been unveiled for a AI-generated radio which responds to text or voice prompts to create a relevant playlist. Neither feature is completely new (Soundhound can identify songs in the same way, and Spotify can produce playlists based on a song – but not a text prompt ) but the company hopes that these new features will give them an edge in the competitive music streaming market.

And finally

“I cried happy tears when I heard my new – old – AI voice”

Quote of the week is from US Congresswoman Jennifer Wexton. The Democratic Representative for Virginia’s 10th District has partnered with ElevenLabs to create an AI version of her voice as it was before her ability to speak was affected by progressive supranuclear palsy. The AI allows her to deliver speeches on the floor of the House in her own voice rather than a generic voice provided by her text to speech app.