tpdly

Agreed! There has not been enough appreciation for re-organization efforts of the dependency-tree of knowledge. Now that expansion of the frontier has been somewhat trivialized, it should be an indication that our priorities have not been entirely wise.

The value of these math concepts is in their explanatory power. We want to understand! Insofar as we want to maximize the diffusion of powerful thought-tools, ironing out the wrinkles at the frontier is needed to be able to fold everything up neatly. The benefit we should seek is wide diffusion of powerful theories, and the folding and the neatness has been long undervalued.

I first learned from self studying Finkbeiner’s “Introduction to Linear Transformations and Matrices” which I thought was very good. Great exercise's. Dover still prints it for pretty cheap. He does a cool thing introducing all sorts of theory of linear transformations, then later showing that matrices are the way to encode them once you chose basis for domain and codomain. I liked that a lot, felt like it removed any magic from matrices. Also a fan of Axler though.

There's another even more abstract liability on the balance sheet here for "deprecated LOC" or something, where the thing they built gets scrapped because no one wants to untangle the rat nest, or the "documentation" was so incoherent that it all had to be rewritten and reviewed.

These are private websites accessible over the public internet, to be clear. Also, there are known relationships between the intelligence community and the big platforms, lets not pretend they have free reign.

Should/do we allow foreign propaganda radio stations? If we accept that the government can (and very much does) impose itself on content platforms for "national security", what exactly is the difference between deliberately insidious information warfare, and collateral damage from market incentives?

I agree that its better to find solutions that involve protections instead of restrictions though. I think it means forced decoupling of indices/curation from advertising. This would make advertising funded addiction feeds compete with paid feed applications.

Not true. Speech and reach are obviously different.

It is entirely conceivable that Americans wake up to the absurd power and manipulation behind algorithmic feeds, and decide to intervene in the hypno-addictive horror-show.

Also, I've been thinking about how curation at the scale of the modern internet is much more analogous to speech than it ever was before. Whatever you want to say, just find someone else who's said it-- or just astroturf via 3rd party and beam it into the skinner boxes. In this frame, freedom of speech actually demands open access to indeces and content. Is it acceptable that network-effect-fueled cartels are the only ones that get to speak through algorithms?

Right. Five years ago the idea was you needed the capitalism to fund compute/training. Now, its clear that open weight models could be a public good. If my PC takes my job I'm still empowered and can still capture that value, so if Anthropic were pursing that end it wouldn't seem so hypocritical.

Yeah, even though I like 'craftsmanship' here, stakeholders don't. I think "durability" is pretty good. 'Engineering standards' perhaps. I think we'll start to see the groundwork laid for software engineering as a proper engineering field, where specific performance characteristics are required for certain stakes.

Code can be like "heavy machinery", maybe needs license to operate. If an app has sensitive data in it, it becomes like a residential structure subject to building codes and inspections. I shudder to think of how this would play out it our oligarchic, anti-competitive environment though.

I mean, not that I'm a huge fan of IP laws, but yes?

Like I said, if you provide an alternative to all these blogs and forums (because you trained on them or because you scrape them for RAG) then you are stealing their traffic. Search engines were/are already doing that, but the foundation training

That sounds legit, but do you think its out of scope? Scam texts and emails result in exfiltrated data, maybe they have to require iMessage and iCloud Mail too?

If Facebook's Meta-Siri is being sketchy, that's a problem with Meta-Siri. Take it off the market, bring down the law. Promote competition, and bad actors must be made to loose. Can we not just status-quo fallacy that re dysfunctional consumer protections? or at maybe agree that the perfect-world scope is one that puts exfiltrators in jail, not just rejected from the app store.

Instead we'll just have Siri AI and Google Assistant AI, and no decent competition. I guess maybe we'll get a Meta phone, if the only way to compete is on the entire mobile computing vertical.

Not foreseeably. As others have mentioned, DMA requires AI integrations to accommodate competition. To my mind, Apple's Webkit-only playbook is the prototype here. Waaaay too much money to be made as a GEO broker. Namely by selling position/advantages in the harness, or just use it to maintain ecosystem lock in.

As an EU resident, I find no benefits-of-doubt needed to explain why competition against foreign mega-corps is being forced. Its protectionist to promote openness when the closed solutions funnel profit abroad.

Apple would have to allow USERS the possibility of giving any virtual assistant direct access to their own private data.

Is that accurate?

I think you undervalue the contribution of internet-scale data to foundation modeling, and because LLMs can obsolete the content they required, I think its fair to characterize it as theft. Obviously RL contributes a lot to capabilities, but the judgement that an LLM uses to 'synthesize information' is born from the training data. The scale of the data really is beyond intuition. books3, for example, would 230 yrs of continuous reading

I actually think the "proprietary non-determenistic database of the free internet" does a lot to characterize the capabilities and effects to a lot of people. Obviously coders are more in tune with how well agents can work, but that's also due more to the RL breakthroughs than foundation modeling.

I like to think someone can come along with Jobs-like charisma and redefine the public's intuition for personal computing again. Foundation models with maximally deterministic harnesses, voice assistants with honest, tasteful amounts of roboticism, maybe data tenancy that doesn't make you feel like a surveilled ecosystem prisoner.

NVIDIA local AI builds are moving in that direction, but are wildly expensive. I think a lot of the current AI backlash (esp data-centers) comes from the public's accurate intuition that the current centralized, monopolistic, cloud-centric solution disempowers them, and eats any gains this magical technology would've brought them. Plugging a sliver box into the wall and your router, with no recurring fees; that would make people feel different about "taking our jobs" because they would be the unquestionable beneficiary.

Average density in an urban core (North America) is like 90 residents per acre, vs like 5 res/acre in suburbs. Not a very big exaggeration