A week in Generative AI: Meta, Mira & Voice

News for the week ending 29th September 2024

Sep 29, 2024

It’s been a big week this week with Meta’s Connect 2024 event, where they announced the multimodal Llama 3.2. Mira Murati announced that she is leaving OpenAI and we also saw the release of ChatGPT’s Advanced Voice Mode everywhere except Europe.

On the Ethics front Meta and Apple have both snubbed the EU’s AI pact, Mark Zuckerberg claimed that publishers overestimate their value for AI training and some research shows that Using GPT-4 to generate 100 words consumes up to 3 bottles of water.

There are also some good long reads from The Guardian about humanoid robots, Sam Altman on The Intelligence Age, and The New York Times has a great article covering what Jony Ive has been up to where he shares he’s working with Sam Altman on some new AI-powered hardware.

Meta Connect 2024

The big event of the week was Meta’s Connect 2024 where they unveiled lots of new products and inevitably talked a lot about AI! There’s some good coverage of everything they announced over at The Verge.

On the AI front, Meta announced Llama 3.2, which is now multimodal and can understand images. This allows it to interpret charts and graphs, caption images, and pinpoint objects in pictures given a simple description. Unfortunately, Llama 3.2 can’t be accessed in Europe, with Meta blaming the “unpredictable“ nature of the EU’s regulatory environment.

Meta AI is also getting celebrity voices. It’s not a touch on OpenAI’s Advanced Voice Mode, but there are approved and endorsed clones of the voices of Awkwafina, Dame Judi Dench, John Cena, Keegan-Michael Key, and Kristen Bell that users can choose from. I’m sure it’s not a shock to hear that Meta has paid millions to license these celebrity voices and they’re making a big bet that this superficial customisation will compete with the better product experience from Advanced Voice Mode. I’m not so sure…

Source

OpenAI CTO Mira Murati says she’s leaving firm to do her ‘own exploration’

I have a lot of respect for Mira Murati and I think of all the leadership team at OpenAI she came out of the firing/hiring of Sam Altman fiasco last year with her head held high. She also did a great job of leading the Spring Event earlier this year, bringing in her team, and showcasing some great new features that were in the pipeline. It’s sad to see her leaving, but I’m excited for what she will do next.

It looks like Mira leaves on great terms, with Sam Altman tweeting an effusive message and also announcing some other leadership changes. It’s been very fluid at the top of OpenAI this year, with Greg Brockman taking extended leave, Ilya Sutskever leaving and many other senior executives leaving the business.

This all comes amongst reports that OpenAI is to remove non-profit control and give Sam Altman equity in the business for the first time as they change from being a research outfit to more of a product business.

One thing’s for sure, OpenAI is a very interesting company to watch!

Source

OpenAI rolls out Advanced Voice Mode with more voices and a new look

First previewed back in May at OpenAI’s Spring Event, OpenAI have released Advanced Voice Mode this week. It’s now available to Plus and Team subscribers, but unfortunately it’s not available in the EU, , Switzerland, Iceland, Norway, or Liechtenstein.

Luckily it is available in the UK, so I’ve been able to test it and it’s a lot of fun! I chose the ‘Sol’ voice, which was described as ‘savvy and relaxed’. There was no noticeable lag, I was able to interrupt it speaking, and I could get it to SHOUT and whisper as well as sound excited. It leads to a much more dynamic and intuitive experience, very close to speaking to a real person.

Advanced Voice Mode doesn’t have the ability to detect emotions in your voice yet, but I think that’s something that will be coming with the video features that were also shown off at the event in May. I also love the fact that they’ve included a couple of British voices in there - ‘Arbor‘ is described as ‘easy going and versatile’ (although he sounds a bit like an east-end gangster to me 🤓) and ‘Vale’ is described as ‘bright and inquisitive’!

Overall, I really like it - much better than the original voice mode, but at this point I don’t think that will mean that it will use it much more than the original voice mode. It’s great for handsfree interactions, but it really needs to underpin my smartphone’s assistant to be of more use so I can invoke it without having to pick up my device. This of course will be coming, and I’m sure this type of technology will be powering the new Siri experience that is coming to iPhones later this year.

Source

Google’s NotebookLM adds attachments, YouTube summaries, and audio overviews

Google’s NotebookLM has been an ‘experiment’ for a while now but they’ve just added a host of new features that make it worth checking out. First and foremost, it’s a simple browser-based note taking app, but it has a lot of Generative AI built into it now.

As it stands, NotebookLM can automatically create notes from lots of different types of documents and YouTube videos. It can summarise them all for you and allow to ask questions about the content. They’ve also got an interesting feature where they basically create a podcast episode about the source material, which is very nice.

For me, this is a great tool and it’s great to see that we’re evolving beyond just a basic chat interface for interacting with large language models. This isn’t a note-taking app that has bolted a large language model on. This is a note-taking app that’s been build around a large language model and it’s a great example of the types of software we’re going to be seeing in the coming years as generative AI technology starts to mature.

Source