A week in Generative AI: Mostly OpenAI
A week in Generative AI: Mostly OpenAI
News for the week ending 2nd June 2024
It’s mostly OpenAI news this week, with a sprinkling of Google who responded to all the controversy around their new AI overviews. OpenAI have had a very busy week! They signed a huge deal with Apple, gave free ChatGPT users access to browsing, vision, file uploads, GPTs and data analysis. They also announced that they have started training their next-generation model, launched a new Safety and Security Committee and signed a big Chat GPT Enterprise and reseller agreement with PwC.
Phew!
Apple Bets That Its Giant User Base Will Help It Win in AI
Things are shaping up nicely for WWDC on the 10th June as it looks like Apple have finalised a deal with OpenAI for their models to power new AI features in iOS 18 and macOS 15. Some of the rumoured features, which aren’t entirely revolutionary, include:
Transcribing
Retouching photos
Auto replies for messages and emails
A new, upgraded Siri experience
Being able to generate custom emojis
Being able to change all your app icons to a new colour
This is a big deal for OpenAI and will likely be worth $bn per year in revenue. It will also drive a lot more usage to their models, dramatically increasing the amount of data they have to improve future models.
Apple gets arguably the best GenAI model powering their new AI features and will benefit from improvements to that model over time. As Apple like to ‘own the full stack’ I absolutely expect them to have their own model powering their devices in the future, but there is a chance that this ends up like its search deal with Google and sticks around.
All ChatGPT Free users can now use browse, vision, data analysis, file uploads, and GPTs.
As someone who has been using the paid features in ChatGPT for over a year, It’s hard to believe that most people using the platform haven’t had access to them. The new data analysis features are truly brilliant and I highly encourage everyone to revisit ChatGPT now that the latest model and all its features are available to everyone for free. ChatGPT is now much more powerful and useful for all users!
Google Search AI Overviews: About last week
I covered in last week’s newsletter how Google search is now a giant hallucination, and here’s a response from Google on the matter. This is good to see - they explain how AI overviews work, why people have been getting odd results, improvements they’ve already made, and that they will continue to improve the experience.
According to Google, Only 1 in 7m unique search queries have had a content policy violation and given that they’re handling billions of search queries every day, there were bound to be a few bumps along the way.
Hopefully AI overviews continue to improve and can be trusted, as they could be genuinely useful, but if people don’t trust them Google will lose a lot of search users.
OpenAI Says It Has Begun Training a New Flagship A.I. Model
I was surprised to read this as I had heard that OpenAI had already completed training on GPT-5 (or whatever they decide to call it) and were in the process of testing it. Potentially, with the release of GPT-4o, that’s the model that I’d heard about but that means that it wasn’t in testing for that long before release. Either way, great to hear that the next generation model is in the process of being trained.
OpenAI’s Safety & Security Committee
Following the departure of Ilya Sutskever and Jan Leike a few weeks ago, the existing Superalignment team was essentially disbanded. It turns out many of the Superalignment team had been leaving over the last 6 months, no doubt fall out from Sam Altman being fired and then re-instated as CEO.
Jan Leike has since started at Anthropic to lead their new Superalignment team and it looks very much like Anthropic are now the company making the most headway on model safety.
Despite this, OpenAI has now created a new Safety and Security Committee led by Sam Altman along with board members Adam D’Angelo and Nicole Seligman. It’s good to see OpenAI still wants to work on this area, but it’s hard not to think that the announcement is a little hollow as if they were really serious about it Ilya Sutskever, Jan Leike, and the rest of the Superalignment team wouldn’t have left.
OpenAI signs on 100K PwC workers to its ChatGPT enterprise tier, as the consultant becomes its first resale partner
This is really interesting. Not only does this increase ChatGPT Enterprise users by c.15% but will probably generate $ms in annual revenue for OpenAI. More interesting is the announcement that OpenAI will now run a reseller programme, similar to how Google operates with Analytics and other products.
PwC are the first company signed up, but this means that consultancies can now sell ChatGPT Enterprise to other organisations and get a cut of the revenue. This means that OpenAI don’t need to do all the sales work and probably aren’t on the hook for customer training and support either.
It will be interesting to see if this accelerates the take up of ChatGPT Enterprise.
AI Ethics News
EU's ChatGPT taskforce offers first look at detangling the AI chatbot's privacy compliance
The ugly truth behind ChatGPT: AI is guzzling resources at planet-eating rates
With the EU AI Act incoming this summer, the bloc lays out its plan for AI governance
Long Reads
The TED AI Show - What really went down at OpenAI and the future of regulation w/ Helen Toner
The Economist - AI firms mustn’t govern themselves, say ex-members of OpenAI’s board
Stratechery - AI Integration and Modularization
The Atlantic - A Devil’s Bargain With OpenAI
VC Cafe - The Magnificent Seven Invest $400 Billion a Year in Frontier Technologies
“The future is already here, it’s just not evenly distributed.“
William Gibson