A week in Generative AI: Siri, Sesame & Growth
News for the week ending 9th March 2025
So, after last weekās launch of two new frontier AI models and Amazonās Alexa Plus, this week was always going to be a little quieter! However, we do have news that Appleās āmore personalised Siriā is being delayed, a really good conversational voice model from Sesame has grabbed a lot of attention, and thereās a great report from Andreessen Horowitz that shows how ChatGPT has doubled their userbase in the last 6 months.
In Ethics News, Microsoft has talked about how they think some British firms are āstuck in neutralā over AI, the worldās largest call centre deploys AI to āneutralise the accentā of Indian employees, and Eric Schmidt argues against a āManhattan Project for AGIā.
I also highly recommend Ezra Kleinās interview with Ben Buchanan, the top advisor on AI in the Biden White House that I posted about on LinkedIn earlier in the week.
Apple Is Delaying the āMore Personalised Siriā Apple Intelligence Features
This news broke on Friday after Apple shared with John Gruber, a renowned Apple commentator, that the more personalised version of Siri that they announced at WWDC last year wouldnāt be shipping until later āin the coming yearā. This is Apple speak for āweāre not going to ship them as part of iOS 18 as intended, but they will come as part of iOS 19 sometime in the next 12 monthsā.
On one hand, I am surprised with this delay as Apple normally donāt announce products that they arenāt incredibly confident theyāll be able to ship within 12 months. But on the other hand, Iām not surprised as what they announced for Siri at WWDC last year was incredibly ambitious, as you can see in the video above.
The main reasons behind this delay are unclear, as Apple donāt like sharing details, but itās probably down to the fact that its incredibly difficult right now to get generative AI models to be reliable. This means building out agent-like capabilities is very hard. Simon Willison also has a theory that Apple might be delaying āmore personalised Siriā because of security concerns. Turns out unreliable generative AI models plus personal data is not an easy mix to get right!
What I do know is that Appleās ambitions for Siri are very high and that they wonāt ship anything until itās ready. This means weāll have to wait longer, but when it arrives it should be very impressive. This is a little ding in the idea that many people will have a capable AI agent in their pocket this year, but if it doesnāt arrive onto iPhones in 2025, it will very quickly arrive in 2026.
Crossing the uncanny valley of conversational voice
Sesame recently launched their first conversational speech model and it was getting a lot of hype this week. Sesame was cofounded by Brendan Iribe, one of the co-founders of Oculus VR, with the goal of achieving what they call āvoice presenceā. They describe this as the āmagical quality that makes spoken interactions feel real, understood and valued.ā
I have to say that Iām very impressed. I tried out the demo in my browser (I couldnāt get it working in Safari, but it was seamless in Chrome) and had a really nice chat with Maya. Itās very human, you can interrupt it, and I can definitely see this as something people will keep coming back to. You can try it out for yourselves here, which I highly encourage you to do so you can experience it for yourselves.
Thereās good coverage from The Verge on how Sesame is the first voice assistant Iāve ever wanted to talk to more than once which is also worth a read.
I donāt think weāre that far off from most generative AI models having voice interactions that are indistinguishable from human voice interactions. Itās going to get weird!
ChatGPT Doubles Userbase in 6 months
This is some really interesting analysis from the team at Andreessen Horowitz with their 4th edition of a report that looks at the top 100 GenAI consumer apps. The big headline is the growth of ChatGPT which now has 400m weekly active users, double what it had 6 months ago. These users are also using ChatGPT more often than they used to, propelling the total number of visits up to 4bn a month.
The other incredibly interesting thing in this analysis is that DeepSeek rocketed to #2 in just a few days following its released in January. Itāll be interesting to see if it can maintain this position over time as the hype around DeepSeek dies down.
Google introduces AI Mode to Search
So, AI Overviews have been controversial since their introduction but they have been getting better over time. Google announced this week that they will now be powered by their new Gemini 2.0 model which should result in higher quality responses and allow Google to show AI Overviews more often. Theyāre currently used by over a billion people.
Google also announced a new AI Mode which expands what AI Overviews can do with more advanced reasoning and multimodal capabilities. Essentially, I think this adds the more āchat-likeā experience of using Gemini directly right into Googleās search bar which is an interesting strategy. Instead of getting people to come directly to Gemini, theyāre bringing Gemini directly to Googleās search product.
This makes sense from a business perspective for Google. If more people start using Gemini in the search bar rather than directly this means that Google still gets to deliver search results and search ads, which is their main revenue stream. However, I donāt think this is good for the user as the Google search experience is incredibly cluttered already. I guess its probably a better middle ground than having ads in a nice clear chat experience though!
Atlas at work
I love these videos from Boston Dynamics that update us on their progress with their humanoid Atlas robot. This is a longer one which looks at why last year they moved from hydraulics to a fully electric Atlas, how they use real world data to aid generalisation, and showcases a real world use case on sequencing in automotive manufacturing. Theyāre making some great progress!
AI Ethics News
Some British firms āstuck in neutralā over AI, says Microsoft UK boss
World's Largest Call Center Deploys AI to "Neutralize the Accent" of Indian Employees
Eric Schmidt argues against a āManhattan Project for AGIā
Anthropicās Recommendations to OSTP for the U.S. AI Action Plan
Scale AI is being investigated by the US Department of Labor
OpenAIās ex-policy lead criticizes the company for ārewritingā its AI safety history
Anthropic quietly removes Biden-era AI policy commitments from its website
Long Reads
The Ezra Klein Show - The Government Knows AGI is Coming
āThe future is already here, itās just not evenly distributed.ā
William Gibson