I did my thesis on using medioni's tensor voting framework to reconstruct noisy, blurry, low-res and the like images. It was sponsored by USGS on a data set that I thought was a bit of a bizarre use case. The approach worked pretty well, with some reasonable success at doing "COMPUTER ENHANCE" type computer vision magic. Later on talking with my advisor about the bizarrely mundane and uninteresting data sets we were working on from the grant he quipped that "You built a reasonable way of unblurring and enhancing unreadable images, the military doesn't care about this mundane use case." It then occurred that i'd been wildly ignorant to what I just spent 2 years of my life on.
How does this differ from what Github Copilot does when writing its .github/copilot-instructions.md? That doesn't keep the transcript or prompts, but it does keep quite a bit of the context and a declarative state of the decisions/design considerations made so another AI bot can pickup and have enough context to understand the rationale. I'm also not really convinced that any AI agent wouldn't still parse the code to understand more about the context vs. just using the checkpoint.
Don't forget stand-ins, much of this hasn't discussed that credit card networks do a lot of "stand-ins" when the issuer is unreachable (bank goes down, latency too high, etc). It's a bit unclear how things like Wero would operate when a network issue hits as Wero and EU rails won't just assume the liability for the transaction and hope it clears later as it does on Visa/Mastercard.
Believe it or not clippy the Microsoft helper for word was a huge interest and feature for all of about 2-3 weeks before everyone realized its interactions were just “on top” of actually doing something. Once the cost of clippy, and its failure to actually be helpful sunk in it was relegated to jokes and eventually down the line memes.
It’s hard to actually create something that is a personal assistant. If I want it to keep and eye out for reservations I guarantee it would take a few hours for me to get that setup, more time that it would take to just watch for reservations.
If I wanted it to find out when I needed to register my child for school then do it, I’m 100% sure it would fail and probably in some range from comical to annoying.
This seems less like a personal assistant and more like a “hey bro how ya doing?”. It lacks the ability to inquire and ask questions and deduce.
If I have to prop it up to complete any random task I have, I’ve just got another version of clippy with a lot more computing power.
Credit card companies have been doing stand-ins to support offline transactions for ages. Mainly the reason why you pay a high interest rate on your credit card.
I've had too may anecdotal failures from replication, innodb consistency corner-case issues and finally being adjacently associated with Oracle killed it for me. Since moving to Postgres, I've been very happy.