A week in Generative AI: Personality, SIMA 2 & Group Chat

News for the week ending 16th November 2025

Nov 16, 2025

This week we had the launch of GPT-5.1 from OpenAI alongside new personality settings, SIMA 2 from Google DeepMind, and news that OpenAI are trialling ChatGPT in new group chats.

In Web 4.0 news we had Mozilla joining the AI browser wars, and a good podcast from OpenAI on ChatGPT Atlas and the next era of web browsing. There was also news that three AI-generated songs topped the Billboard and Spotify charts. Sigh 😮‍💨.

There’s also a good Long Read from Understanding AI on how AI ads are going mainstream that has some good follow up on the Coca-Cola ad I wrote about last week.

OpenAI launches GPT-5.1 with new personalities

OpenAI launched GPT-5 in August, and now 3 months later we have GPT-5.1, which makes ChatGPT “smarter and more enjoyable to talk to”. OpenAI describe the new model as “warmer, more intelligent, and better at following your instructions” and say that it should be faster at simple tasks and more persistent at more complex tasks.

Most interestingly though, OpenAI have included new personality settings with the introduction of the new model, something I think will be an important area for users to be able to control. There are now 8 personality settings to choose from:

Default (balanced style ad tone)
Professional (polished and precise)
Friendly (warm and chatty)
Candid (direct and encouraging)
Quirky (playful and imaginative)
Efficient (concise and plain)
Nerdy (exploratory and enthusastic)
Cynical (critical and sarcastic)

OpenAI have also said that they will debut new experimental ways for users to fine-tune ChatGPT’s style further, which they’ve already started testing with some users. I’m looking forward to seeing what these look like and whether they’ve decided to use the OCEAN 5 big personality traits for this, my preferred HEXACO model, or something else entirely.

If you’re interested in another take on this, John Gruber has some good commentary on the new personalities.

Source

SIMA 2: An agent that plays, reasons, and learns

I wrote about the launch of the first SIMA model from Google DeepMind back in March 2024. Back then it was very simple - they were training a model to be able to play a variety of 3D games to then learn how to navigate the real world.

With this second model, the ambition is still the same, but they now have an AI that can play 3D games and learn on its own as well as transfer the learnings from one game to another. This sounds really simple for us, but is something that AI finds incredibly difficult and a great example of Moravec’s Paradox - easy for humans = hard for computers and vice versa.

We won’t see the results of this research in the real world for a few years, but it lays the foundation for AI models that are continuously learning and able to generalise these learnings and apply them to lots of different situations, much like people do. Current AI models can’t do this and its one of the biggest things holding them back.

Source

ChatGPT gaining group chat feature in four regions

It looks like OpenAI is increasingly trying to rebuild the entire internet one piece at a time on top of ChatGPT. Here’s a timeline of the big features they’ve announced over the last 12 months or so:

Oct 2024 - ChatGPT Search (Search)
Apr 2025 - ChatGPT Shopping (Shopping)
Sept 2025 - OpenAI Jobs Platform (Hiring)
Sept 2025 - Sora 2 (Social Media)
Oct 2025 - ChatGPT Apps (Services)
Oct 2025 - ChatGPT Atlas (Browsing)

The latest product they’re trialing is ChatGPT Group Chats (Messaging), which I think is really interesting. In the Group Chats you can collaborate with other people, as well as use ChatGPT to help. You can mention ChatGPT directly to get it to respond, but OpenAI have also taught ChatGPT new social behaviours for group chats. It follows the conversation and decides when to respond and when not to. This is going to be an increasingly important skill for AI to learn as they become more proactive in our lives, something I’ve written about before.

There are still a few big areas of the internet for OpenAI to go after - Email, Banking (and other admin) and News consumption. There’s a chance that ChatGPT Pulse evolves into a news product, and Banking will be hard (although I do think they’ll go there). So my money is Email being the next area that OpenAI goes after.

Source

Web 4.0

AI Ethics News

Long Reads

One Useful Thing - Giving your AI a Job Interview
Understanding AI - AI ads are going mainstream

“The future is already here, it’s just not evenly distributed.“
William Gibson

Simon Valcarcel

Nov 17

Thanks for the newsletter, as always. I’m currently thinking about the privacy implications of Open AI’s Atlas browser. Delve a little deeper into their privacy policy and it gets a little opaque. Will users want (or need) that level of deep insight about your browsing history/personal data being stored in Open AI’s memory to be recalled at a later date? Do people even think about what’s being collected and the implications? All browsers collect data but some are more transparent/give users better control over what’s stored. Great article, delving into the subject: https://archive.ph/48HqZ

Expand full comment

1 reply by Sean 🤓

1 more comment...

Discussion about this post

Ready for more?