as long as it's new I tremendously enjoy binge watching Claude:
I have three tabs open and if one of them is not doing something interesting I just switch to a different channel, and occasionally influenced the narrative
The article basically claims that LLMs are bad at politics and poker which is both not true (at least if they receive some level of reinforcement learning after sweep training)
you are living in the past these models have been trained on image data for ages, and one interesting find was that even before that they could model aspects of the visual world astonishingly well even though not perfect just through language.
you are correct the token representation gets abstracted away very quickly and is then identical for textual or image models. It's the so-called latent space and people who focus on next token prediction completely missed the point that all the interesting thinking takes place in abstract world model space.
someone might build enough hype around it to challenge the oligopoly.
As for training on existing browsers: they are trained on the whole corpus of the human thought process and can take many insights from other fields into the browser, they can write a browser in a completely new language without just transpiring but building it from first principles or as karparthy calls it spec based programming.
this is the whole message of this hype that you can churn out 500 commits a day relatively confidently the way you have clang churn out 500 assemblies without reading them. We might not be 100% there but the hype is looking slightly into the future and even though I don't see the difference to Claude code, I tend to agree that this is the new way to do things even if something breaks on average it's safe enough
it's not going to help them. For Siri to be really useful it wouldn't need deep system integration and an external model is not going to provide that. People don't believe me when I said it about Apple Intelligence with open AI
i'm a real developer and it is absolutely magic! if someone showed me a demo of talking to a computer which works for one hour to find and fix my bugs all on its own five years ago I'd definitely call it magic and I still call it magic.
does it make a mistakes? yes sometimes but you can verify with tests or with lean.