Visa is embedding its payment network into ChatGPT, allowing the chatbot to shop and complete transactions for users.
Visa said Wednesday that it has embedded its payment network inside of ChatGPT, empowering the chatbot to independently shop and complete transactions on behalf of its user. It means AI agents can not only recommend products but complete the purchase on the user’s behalf, at potentially any merchant that accepts Visa. The payment network’s previous attempts at this technological leap were confined to a single retailer or a small set of enrolled merchants.
Ahhahhahahahhahahahahahhahhaaa.. *deep breath*
AHHHAHAHHAHHAAAA..!!
No
To be clear: simply having a Visa card isn't a risk for you. Not yet, anyway. You'll have to intentionally link your account to ChatGPT, then go shopping with the chatbot. If you don't do that, you're safe.
Dunno what stupidity they're planning next, but for now? ChatGPT isn't going to rack up $1000 of cat toys without your knowledge unless you're an active ChatGPT user.
This is so insane because like... I get paid to test AI for various companies (yeah yeah the system makes class traitors of us all). And I gotta say... it's so blisteringly obvious that we are not at a place where this won't go badly (for users) yet. A few weeks ago I watched a major model experience what I can only describe as a total manic hallucinatory episode because it had to describe character from an obscure mascot company that doesn't have much of an English language presence online and like, of the 100 or so characters it named and described, less than 10% actually existed. And this is a mascot company. Like, Sanrio style. Those characters exist PURELY to be purchased in tchotchke/phone charm form and be broadly marketable and purchasable. It is, by a wide margin, the worst I've ever seen a model fail since I started this job. This is gonna fuck non-English markets so fast.
We also recently had a project specifically about a model helping with shopping and like... it did OK? Usually? Like... the failure rate for factuality was pretty normal for these projects but uh... that's still not great? That's still 5-15% of tasks that I reviewed having errors that would cause it to recommend an inappropriate product for the users needs.
Now, maybe the latest ChatGPT is much MUCH more reliable than the models we were training and reviewing. I don't use AI outside of work (on account of I hate it for ideological reasons) so it's hard to say but still... I doubt its developers have figured out how to make it drop that 5-15 to zero given how good at slipping lies that will still get them rewarded for doing a good job past testers the models I HAVE used are. Visa almost certainly knows this is going to poorly for a not insignificant number of users, they just did the math and figured out that the benefits to them will outweigh those costs.
Ding. That's it exactly. They're rushing to monetize their agents before the whole house of lies collapses. Even if the models aren't ready.
I mean, ChatGPT still can't do math or date calculations reliably. Those are kinda important for e-commerce.















