Outstanding breakdown of Google's Universal Commerce Protocol launch. The open standard approach versus OpenAI's walled garden ACP feels like a critical fork in how AI commerce will evolve, and honestly the UCP's flexibility withpartnerships like Shopify and Walmart might just define the next decade of online shopping. I've been testing similar AI-driven purchase flows and the frictionreduction is wild when it works, but trust remains the bottleneck for most peopel I know. The healthcare AI section also caught my eye with HCA going live - finally seeing HIPAA compliance tackled properly changes the game for actual clinical workflows.
"Rob Wilbin is right that it is common for [expert in X] to tell [expert in Y] they really should have known more about [Y], but that there are far more such plausible [Y]s than any person can know at once."
I assume that this should read
"Rob Wilbin is right that it is common for [expert in Y] to tell [expert in X] they really should have known more about [Y], but that there are far more such plausible [Y]s than any person can know at once."
The language for that Chinese AI law is extremely similar to that used in their current internet censorship laws. Which as you say is sp broad that on practice everyone is constantly violating them. This is a feature not a bug from their perspective because it gives them cover to interfere at will by their normal approach of selective enforcement.
Historically this has meant that to have a viable business model in China companies have to develop a good relationship with the regulator, both in terms of bribing and flattering the correct officials and following whatever their directives of the day are. Which then leaves open the risk that a sufficiently well connected company is allowed to do whatever it likes, or they are directed to develop malicious uses for AI deliberately by the government.
I would assume at minimum they'd require the AI companies to report on people, suppress certain information and promote the "correct" government narratives on various topics, as web and social media companies do already.
I don't want to create unlimited minds, just enough that the odds of being born in the bad old days of humanity's history (now included) instead of the outer singularity are on the order of 0.1%. That's just a few trillion!!!
Couldn't note this on ACX since Substack is once again not loading the comments there properly ("wonder if Claude Code could fix this"), but: it surprised me that despite having no exposure to AI besides reading DWATV posts, I was able to fully understand every joke in Scott's most recent hilarious SOTA BAHP post. Doubly surprised at other ACX regulars being baffled at the levity and needing to have the humour explained. Clearly the educational mission of your AI posts works, if a less-intelligent grocery bagger like me can find great mirth at --dangerously-skip-parmesan! Grokking niche humour is one of life's great joys, indeed...
It's not clear to me how open the UCP thing is. Is it just the protocol that's open, or can I build my own Walmart shopping agent now? It would be a shame if we have this cool thing but then you need to strike bilateral deals with each retailer to use it. I would love to build my own grocery shopping agent, for example.
… and how difficult would it be to connect my own shop to that? Without paying at least four digits for some "certified" software that is authorized to talk to the credit card companies? Or the banks (in which country??).
There's already a seriously-walled-off garden on my phone or watch; it's basically impossible to program your own smartwatch to authorize POS payments (let alone accept them). I'd hate to have yet another one of these expensive oligopolies for online shopping.
Just seems to me like Burry is behind on his understanding of things in this realm, he's probably too busy shorting NVDA and PLTR. Of all the things he points out it's that he could save $800 in plumbing work... the guy has millions of dollars, he is not going to reach into his wall and risk electrocuting himself or getting bathed in raw sewage. Just wait for the news stories of some guy electrocuting himself because he followed an AI's advice and spliced the wrong wires. Yikes.
Wow is that some weird uncanny-valley of Ghibli. Oh my God, Karen, you can't just "animate" half the image while leaving several notable elements photorealistic! I do understand that the IP-violation train has to stop sometime, fun is no excuse to flout laws and artistic norms, but...ugh. This too is a possible AI future: one where base capabilities are high, but they're relentlessly hobbled by all the same forces producing stagnation in other areas, locking away much of the utility. Makes me glad to see the current turn with HIPAA compliance, since that was absolutely a potential "pause button" that could have easily been pushed.
> risking killing innocent people is kind of a huge deal.
I wouldn't be too sure about that. Just look at the seriously nontrivial number of civs killed in Iraq. Not to mention those on Israel's conscience. Or Russia's, assuming Putin has one. Or …
Michael Burry is de facto uninsured as of his tweet. If he claims on anything, the company will cite his tweet as evidence he made significant modifications while unsupervised by a trained and certified professional, and by so doing caused defects.
"Once again we have a call for ‘the humanities’ as vital to understanding AI and our interactions with it, despite their having so far contributed (doesn’t check notes) nothing, with notably rare exceptions like Amanda Askell."
And yet, if I had to name one person most likely to save us from catastrophe (admittedly with very low odds) it might well be this rare exception. In a sane world this would not be true, but given our current trajectory, I don't see a technical solution to alignment emerging in time.
Outstanding breakdown of Google's Universal Commerce Protocol launch. The open standard approach versus OpenAI's walled garden ACP feels like a critical fork in how AI commerce will evolve, and honestly the UCP's flexibility withpartnerships like Shopify and Walmart might just define the next decade of online shopping. I've been testing similar AI-driven purchase flows and the frictionreduction is wild when it works, but trust remains the bottleneck for most peopel I know. The healthcare AI section also caught my eye with HCA going live - finally seeing HIPAA compliance tackled properly changes the game for actual clinical workflows.
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/ai-151-while-claude-coworks
The link here seems broken, it points to a coefficient giving tweet:
GLM-Image claims to be a new milestone in open-source image generation
Presumably a typo here:
"Rob Wilbin is right that it is common for [expert in X] to tell [expert in Y] they really should have known more about [Y], but that there are far more such plausible [Y]s than any person can know at once."
I assume that this should read
"Rob Wilbin is right that it is common for [expert in Y] to tell [expert in X] they really should have known more about [Y], but that there are far more such plausible [Y]s than any person can know at once."
Right?
> How can you stop an AI from ‘spreading rumors’?
The language for that Chinese AI law is extremely similar to that used in their current internet censorship laws. Which as you say is sp broad that on practice everyone is constantly violating them. This is a feature not a bug from their perspective because it gives them cover to interfere at will by their normal approach of selective enforcement.
Historically this has meant that to have a viable business model in China companies have to develop a good relationship with the regulator, both in terms of bribing and flattering the correct officials and following whatever their directives of the day are. Which then leaves open the risk that a sufficiently well connected company is allowed to do whatever it likes, or they are directed to develop malicious uses for AI deliberately by the government.
I would assume at minimum they'd require the AI companies to report on people, suppress certain information and promote the "correct" government narratives on various topics, as web and social media companies do already.
I don't want to create unlimited minds, just enough that the odds of being born in the bad old days of humanity's history (now included) instead of the outer singularity are on the order of 0.1%. That's just a few trillion!!!
Couldn't note this on ACX since Substack is once again not loading the comments there properly ("wonder if Claude Code could fix this"), but: it surprised me that despite having no exposure to AI besides reading DWATV posts, I was able to fully understand every joke in Scott's most recent hilarious SOTA BAHP post. Doubly surprised at other ACX regulars being baffled at the levity and needing to have the humour explained. Clearly the educational mission of your AI posts works, if a less-intelligent grocery bagger like me can find great mirth at --dangerously-skip-parmesan! Grokking niche humour is one of life's great joys, indeed...
It's not clear to me how open the UCP thing is. Is it just the protocol that's open, or can I build my own Walmart shopping agent now? It would be a shame if we have this cool thing but then you need to strike bilateral deals with each retailer to use it. I would love to build my own grocery shopping agent, for example.
… and how difficult would it be to connect my own shop to that? Without paying at least four digits for some "certified" software that is authorized to talk to the credit card companies? Or the banks (in which country??).
There's already a seriously-walled-off garden on my phone or watch; it's basically impossible to program your own smartwatch to authorize POS payments (let alone accept them). I'd hate to have yet another one of these expensive oligopolies for online shopping.
Just seems to me like Burry is behind on his understanding of things in this realm, he's probably too busy shorting NVDA and PLTR. Of all the things he points out it's that he could save $800 in plumbing work... the guy has millions of dollars, he is not going to reach into his wall and risk electrocuting himself or getting bathed in raw sewage. Just wait for the news stories of some guy electrocuting himself because he followed an AI's advice and spliced the wrong wires. Yikes.
Wow is that some weird uncanny-valley of Ghibli. Oh my God, Karen, you can't just "animate" half the image while leaving several notable elements photorealistic! I do understand that the IP-violation train has to stop sometime, fun is no excuse to flout laws and artistic norms, but...ugh. This too is a possible AI future: one where base capabilities are high, but they're relentlessly hobbled by all the same forces producing stagnation in other areas, locking away much of the utility. Makes me glad to see the current turn with HIPAA compliance, since that was absolutely a potential "pause button" that could have easily been pushed.
> risking killing innocent people is kind of a huge deal.
I wouldn't be too sure about that. Just look at the seriously nontrivial number of civs killed in Iraq. Not to mention those on Israel's conscience. Or Russia's, assuming Putin has one. Or …
Michael Burry is de facto uninsured as of his tweet. If he claims on anything, the company will cite his tweet as evidence he made significant modifications while unsupervised by a trained and certified professional, and by so doing caused defects.
"Once again we have a call for ‘the humanities’ as vital to understanding AI and our interactions with it, despite their having so far contributed (doesn’t check notes) nothing, with notably rare exceptions like Amanda Askell."
And yet, if I had to name one person most likely to save us from catastrophe (admittedly with very low odds) it might well be this rare exception. In a sane world this would not be true, but given our current trajectory, I don't see a technical solution to alignment emerging in time.