I just reached out and got on the waitlist for Subquadratic. I am SUPER interested to see how well that new model handles specific details in a 12M token sea. If they have actually nailed that it'll be a HUGE step forward. Lol and the phrase 'pixels remain a pricey proxy for proper plumbing' was pure AWG gold 😁👌!
One thing that always makes me “nervous” is Meta and AI ( to some degree also Google) - how does Meta make $$$ - learns and reads its users and tries to “influence” them - do you really want an AGI system 5 years from now with this goal ?
The line about pixels being a pricey proxy for proper plumbing feels like the center of gravity here.
I think the same issue applies to “autonomous” agents. We often frame them like independent workers, but the real production question is whether the work has been made legible, bounded, and governable enough for AI to operate reliably inside it.
In my own experiments with agentic tools, the constraint usually has not been intelligence. It has been authority, workflow design, interfaces, permissions, and escalation paths.
The destination may be less autonomy in the theatrical sense and more governed autopilot.
Thanks for the mention of PA suing Character Tech! They (and their fans) are a sleazy bunch making it difficult for those of us creating models, avatars, real AI Actors and so forth.
I just reached out and got on the waitlist for Subquadratic. I am SUPER interested to see how well that new model handles specific details in a 12M token sea. If they have actually nailed that it'll be a HUGE step forward. Lol and the phrase 'pixels remain a pricey proxy for proper plumbing' was pure AWG gold 😁👌!
You see what I did there!
That’s some serious alliteration there.
Artful alliteration anchors active attention. Nicely done, Alex. That's 2 alliterative articles back to back.
Brilliantly bantered, buddy! Boundless bravado bolsters bold blogging.
i just reached out to Subquadratic too, This would be a remarkable breakthrough and I am excited to try this out.
One thing that always makes me “nervous” is Meta and AI ( to some degree also Google) - how does Meta make $$$ - learns and reads its users and tries to “influence” them - do you really want an AGI system 5 years from now with this goal ?
The line about pixels being a pricey proxy for proper plumbing feels like the center of gravity here.
I think the same issue applies to “autonomous” agents. We often frame them like independent workers, but the real production question is whether the work has been made legible, bounded, and governable enough for AI to operate reliably inside it.
In my own experiments with agentic tools, the constraint usually has not been intelligence. It has been authority, workflow design, interfaces, permissions, and escalation paths.
The destination may be less autonomy in the theatrical sense and more governed autopilot.
We are faced with a huge disconnect between capex and income
There is no logical way to recover the capital being thrown at ai buildout plans
Much less power build out plans as we move to more efficient compute including the promise of quantum from Google
Goodbye tsmc.
Hello Apple desktop
Thanks for the mention of PA suing Character Tech! They (and their fans) are a sleazy bunch making it difficult for those of us creating models, avatars, real AI Actors and so forth.
The value of this dispatch is enormoulistic (!)
For any other motals though.. have Cowork re-format it neatly and explain it to you like a smart 6th grader every day 🤌
Mortals*
Great news! Apple is taking a positive step with AI by giving users the option to choose their preferred model.