Welcome back to Neural Notes: a weekly column where I condense some of the most interesting and important AI news of the week.
This time around we have yet another pre-holiday season raise here in Australia, OpenAI makes its first official deal with a media company, and Google has some more announcements relevant to small businesses.
Relevance AI’s $15 million win
We may be just a week away from the office shutdown period, but the cash tap isn’t slowing down yet. Especially not in the world of AI. Last week we had the mammoth $47 million raise for Leonardo.Ai, which has been followed by a $15 million chaser for Relevance AI.
The idea of Relevance AI is to give businesses low-code tools to autonomously perform detailed workflows or tasks, with the platform aiming to improve worker performance.
One of the co-founders, Daniel Palmer, says Relevance AI is different from other tools flooding the market, including ChatGPT, because they act as a “copilot” whereas this platform is aiming more for an autopilot.
You can read our full report here.
More Google Gemini news
Last week saw the release of the long-awaited Google Gemini. Well, multiple Geminis. Only the mid-range version of the AI was released upon announcement, to mixed reviews. But now we have a little bit more to share.
This week Google has recently made two AI dev tool announcements, introducing AI Studio and Duet AI, the latter of which was announced earlier this year.
AI Studio, formerly known as MakerSuite, is a web-based platform designed to give developers access to Google’s Gemini AI models for creating chatbots and prompts. It offers quite a high quota for developers on the free tier — up to 60 requests a second — but there’s a catch.
Google has flagged that AI Studio may be monitored for quality and improvement purposes. However, details such as the developer’s account and API key are said to be anonymised.
Duet AI for Developers, which focuses on code completion and generation, is now fully available for use and will soon integrate Gemini.
Duet AI for Developers has partnered with 25 companies to provide datasets for platform-specific development support. It’s considered to be an enterprise-grade product and includes 20 programming languages and an indemnification guarantee from Google. This has been a very popular inclusion for companies of late, with both Canva and Leonardo.Ai also announcing similar products.
If you want to get a little taste of it, Duet AI for Developers is free to use until the end of January. From then it will pivot to a paid model.
One thing that is definitely worth noting — Google sure is enjoying the drip-feeding of news and features when it comes to its AI offerings.
OpenAI to train on Axel Springer publications
Well, it was bound to happen eventually. Over the past year, we’ve seen increasing concerns around large language models (LLMs) being trained on media, artworks, and books without permission. There has been a great deal of discussion on how big tech companies can go about doing this in a more ethical way that the originators of the content can benefit from.
And now OpenAI is having a crack at it.
The main character of the AI business world has just signed an agreement with Axel Springer — the media empire behind Business Insider, Politico, and other publications.
The deal will allow OpenAI to train its generative AI models, such as ChatGPT, on articles from these websites. It has also forged a similar agreement with The Associated Press.
ChatGPT will be able to generate summaries of some Axel Springer articles, including paywalled ones, alongside attributions and links to the original pieces.
It’s currently unclear how much OpenAI will be paying Axel Springer for this deal.
As a reporter, I naturally have complicated feelings about this. On the one hand, one has to appreciate a new revenue stream. On the other, we have historically seen what has happened when news organisations have relied on other platforms to push their stories — particularly in the golden age of Facebook. The Zuckerberg giveth, the Zuckerberg taketh away.
And if ChaGPT isn’t just being trained on articles, but can offer users summaries and an attribution link they’ll never click on, site visits are likely to take a dive.
Still, this news is still less disturbing than Channel 1, a news channel launching in February 2024 that will be presented by AI avatars. The stories and editorial decisions are also said to be heavily influenced by AI. I’m sure there will be absolutely no problems with that at all…