Not a big news week this week, but some interesting reflections on ChatGPT Images 2.0, and some good videos from Sequoia Capital’s AI Ascent Conference that include an interview with Demis Hassabis and a presentation from Jim Fan.
In Web 4.0 news there are rumours that OpenAI’s growth in terms of users and revenue is stalling (which might explain their pivot away from consumer and towards enterprise) and in Ethics News there’s some interesting research on Dead Internet Theory and signs that despite faster/higher adoption amongst young people that there’s the start of a backlash.
Enjoy!
Images 2.0 is on the Path to AGI
So, I have to admit that I didn’t realise how big a deal ChatGPT Images 2.0 was last week when it was first announced and I still don’t think many people have. It’s the ‘new thinking’ capabilities’ that OpenAI have added to this model that people are starting to realise are really powerful.
Take the example above, Images 2.0 was able to create an image of a lego set, but also the instructions and parts list.
It’s good for eduction too, with up-to-date, factually correct information:
What makes it great isn’t the fact that it can generate images, or that it can generate lots of related images. It’s that it can go off, plan, do some research, fact-find and then string together the visual expression it needs to answer the prompt. All in one-shot.
Latent Space have a great article about how this put’s the ‘G’ in AGI that I think is worth a read to understand why image generating capabilities like these are important and not just a distraction.
Demis Hassabis: We’re Three Quarters of the Way to AGI
Speaking of AGI, there was a great interview this week with Demis Hassabis as part of Sequoia Capital’s AI Ascent 2026 conference. He talked about where he started in game development and using early AI in his games, key startup lessons he’s learned, and how he believes AI is the ultimate scientific tool.
Demis believes that there is a lot more to come from the scientific application of AI, especially in biology and drug discovery. This builds on the work they’ve been doing with AlphaFold and he expects big advancements in the next couple of years.
The interview starts to veer into philosophy towards the end (which I like, but I know isn’t everyone bag). But it’s well worth some of your time to watch if you’re curious about the impact AI could have on humanity over the next 5 years.
Robotics’ End Game: Nvidia’s Jim Fan
Another great presentation from AI Ascent was from Jim Fan from NVIDIA on why robotics may be about to have its own “GPT moment”.
His core argument is that robotics is starting to copy the playbook that made language models work: large-scale pre-training, fine-tuning, reinforcement learning and scaling laws. But instead of predicting the next word, robots need to predict the next physical state of the world.
He argues that video models, egocentric human data, simulation and “world action models” could finally solve the data bottleneck that has held robotics back. It gets pretty technical in places, but the big idea is simple: we may be moving from robots that are carefully programmed to robots that learn from watching, simulating and acting in the world.
Definitely worth watching if you’re interested in what happens when AI moves from pixels and words into atoms.
Web 4.0
OpenAI Misses Key Revenue, User Targets in High-Stakes Sprint Toward IPO
Microsoft says it has over 20M paid Copilot users, and they really are using it
Meta says its business AI now facilitates 10 million conversations a week
Stripe introduces Link, a digital wallet that autonomous AI agents can use, too
AI Ethics News
Dead Internet Theory Is 17% of the Way to Becoming Reality, Study Finds
Tech giants’ results show rosy outlook for AI boom and US stock market
Long Reads
OpenAI - Where the goblins came from
“The future is already here, it’s just not evenly distributed.“
William Gibson