🟣 EV Daily: The AI browser war
Five things to know today + highlights from my conversations
Lead story: The AI browser war
The AI browser wars have started. Three launches define this new front: Dia in June, Comet from Perplexity yesterday and a rumored OpenAI browser expected in the coming weeks.
Dia was first out of the gate. You can chat with AI using the info from all your tabs. Comet builds on the same foundations. Its main addition is the ability to take actions across your tabs. In theory, Comet can interact with spreadsheets or edit Figma files. In practice, its ability to do so reliably is flawed. OpenAI’s rumored browser may go further, embedding its excellent tool-using agents for reliable actions.
The victor will win a huge prize. The company that owns the browser owns the user session, gaining access to valuable behavioral data and the ability to steer the revenue funnel. Whoever captures the front door to the web gets to watch, and eventually automate, everything we do online. Until, of course, AI-native devices hit the mainstream. [TechCrunch; YahooFinance; BensBites] #AIapplications
Key signals, quick scan
A 30-second scan of four secondary signals that hint at where the curve is bending.
xAI has released Grok 4, said to be the most powerful model in the world according to the demanding Humanity’s Last Exam and ARC-AGI benchmarks. The company achieved this by increasing reasoning-focused training tenfold compared with Grok 3. Scaling laws still have room to run. (X) #AImodel
Hugging Face has unveiled Reachy Mini, a fully open-source desktop robot. For just $299 you get a Python-programmable device with full compatibility with Hugging Face models and datasets, inviting hands-on experimentation with embodied AI. It democratizes access to robotics, much like the Sinclair ZX81 did for me and many others in 1981, when it opened the door to personal computing. [VentureBeat] #Robotics
A China-linked consortium, including JD.com and Ant Group, is exploring a renminbi-based stablecoin aimed at settling trade outside SWIFT. If successful, it could erode the dollar’s leverage in emerging-market commerce. [Financial Times] #Economics
Mistral AI is poised to secure a $1 billion raise led by Abu Dhabi’s $100 billion MGX fund. This would be Europe’s largest private AI funding round and would finance a 10,000-GPU “Mistral Compute” cluster, made of H200s. While massive for Europe, the cluster is still an order of magnitude below that of leading US AI labs. [The Information] #Infrastructure
Highlights from my conversations: There is no wall
Last week I checked in on my start-of-year predictions. Here is an excerpt on one of those predictions that there would be no ‘AI wall’: the idea that scaling would eventually hit diminishing returns.
On the Frontier Math test, which GPT-4 used to score 2% on, o3 got 25%, and if you’re using the tools, you have noticed they’ve got better, although, as people also know, they’re really, really quite unstable.
Grok 4’s release, as mentioned earlier, further emphasises that there is no wall. The graph below shows Grok’s performance on the Humanities Last Exam, which, as the name suggests, is an extraordinarily difficult benchmark. The more training time, the better the performance.
Watch the full episode on YouTube. I go live on Substack every Friday at 5 pm UK time | 12 pm Eastern | 9 am Pacific.
Was this email useful?
Thanks! Want to share this with someone like you? Forward the email.
Want to opt out of the daily emails? Follow these steps. You can still receive essays and the Sunday edition as usual.
I tried Comet yesterday. I was shocked to discover the agent can only use the keyboard to navigate. I wanted to automate something on a website that only supports a mouse, so Comet couldn't do it.