@PGelsinger My deepest appreciation to Pat, and my warmest wishes for whomever will lead Intel next. I vote for Jim Keller of @tenstorrent :-) https://t.co/ALrAIYzEEn
The Littoral University: Redesigning Higher Education for the Age of AI | iHack, therefore iBlog @khanacademy @michaelbhorn Know anyone else working on this? https://t.co/jDmzrQRE6A
If ChatGPT ran for president, would politics be better, or worse? @pmddomingos discusses themes from his rip-roaring new book "2040: A Silicon Valley satire", and the wider future of AI, on the latest @LondonFuturists Podcast episode https://t.co/SsKwNMz6zJ
The connector that integrates vision and language is crucial in the Cambrian framework and is regarded as a part of the visual representation. Relying on a simple MLP projector may not fully harness the potential of a good visual representation.
In Cambrian-1, we propose a general, flexible, and vision-centric connector design (Spatial Vision Aggregator, SVA) and demonstrate that spatial inductive bias and deep interaction between LLM transformer blocks and visual tokens enhance the aggregation and condensation of visual features (which can come from multiple encoder towers). [7/n]
As a former physics faculty member with a PhD in philosophy of physics who also is very interested in cognition and artificial intelligence, I am happy to call this one a draw.
Quantum mechanics in particular tells us that the metaphysics of the world isn’t at all like what we might have expected.
On the other hand, one of the things physics hasn’t been helpful at is the hard problem of conscious or even explaining much about mental supervenience on the physical works at all. Best we get is elder statesmen physicists making shaky speculations that basically physics must have key gaps (sorry, Roger).
Basically the two key questions are what are we and what is the world. So I think we can put physics and AI on par.
In terms of actually progress and what would be a thrilling career if you’re a teen: I grew up completely drawn in by the “any day now” theory of everything excitement. I was mentored by John Wheeler at Princeton. It was very promising!
But, well, it’s turned out to be sort of a fizzle. @skdh has a lot to say about that (and QM!). Now I look back and one of the most exciting things I saw was 1988 or so this guy I don’t know was so famous then proving that quantum computers are still limited to the scope of lambda calculus / Turing computable functions. As a way of exploring the Many Worlds Interpretation. That Wheeler had helped on. Now we all know @DavidDeutschOxf but I think I was in the first 1000 to stumble upon that paper.
Speaking of stumble upon, I left all that for technology entrepreneurship and I’d argue that the intersection of that and physics and AI is probably the hat trick for profound and fun!
A few years ago, I visited Pontevedra and Vauban.
Pontevedra is a city in Spain that moved on from cars 25 years ago. Vauban is a neighborhood in Germany that was designed so people wouldn’t need them.
It was so amazing that I cried.
Here’s what I saw:
https://t.co/bJ4baSU4W2
The whole point of a startup is to burn ego to buy learning.
If you aren't regularly unlearning something fundamental about the world in which you operate, you are no longer a startup. You've either already become a real business — or you're just pretending…
~doplyr-harbur and ~hastyp-patmud are making Urbit app composability real with the Global Type Registry.
With the GTR, builders will be able to fork existing applications/templates, browse primitives from the registry, pipe them together, and publish, all in one place.
4/5
@basileSportif Unless of course the demand gets great enough that someone invents something better for that platform (e.g., how Apple shifted from Objective-C to Swift, or the many alternative JVM languages).