This attitude towards exposing customer data as a palatable oopsie on someone’s path to learning (by… outsourcing the effort of learning to an LLM?) is truly disgraceful.
That’s not a viable business, it’s a walking liability. Besides which, why would anyone trust your friend (as an investor or customer) ever again when they’ve shown such profound disregard for user data and their IP? If your metric of success is “I have no idea what I’m doing and still made money from it” your friend would have a better time starting a podcast.
I'd be curious to know what your success rate is with altering the system prompt. I'd be surprised if this wasn't more of an issue with the application layer, and therefore easily modifiable, than the LLM.
It's a straw man in that you're establishing an inherently facile and ridiculous scenario just to knock it down. A scenario that, as others have demonstrated, is not grounded in any logical reality. "Nobody mentioned this imaginary horrible system I just thought of, but if they had, it sure would be terrible" is quite a hill to die on.
If you're going to prompt an LLM to write a low-effort marketing blog that doesn't even reckon with the most surface level counter-arguments, at least make sure your web server can handle the traffic.
What's especially frustrating about all of these "Signal could flip a switch and steal everybody texts!" histrionics is that if they were interested in doing that they... wouldn't work at Signal. They'd go join/start the hundreds of other companies we've heard of in the past few years that have stored/leaked incredibly sensitive data with an insignificant fraction of the effort Signal have put in to establishing their credibility (the TeleMessage scandal being just the latest). People should hold Signal accountable, constantly, forever. But the baseless FUD is frankly hysterical from a forum of ostensible technologists.
This thread isn't about data in general, only passwords. So first of all, a strong password is much harder to crack in the instance that it's stored in a hashed form in the database. In the instance it's stored (unforgivably) in cleartext, it cannot be used, because an additional factor is required to authenticate. That is how exactly.
I think this is a feature rather than a bug for AI advocates who ultimately dream of a day when they can do away with pesky things like labor forces and be a CEO of their own rentier-capitalist empire without hassle. Employees are just a vestigial obstacle holding down the brilliant visionaries at the top.
I think the point with most applications of LLMs is that the users don’t care if it’s good, only if it’s good enough. How much despair this induces is down to your bias for whatever subject is being reduced.
Firefox has made a few headlines over the past year for privacy-unfriendly moves. That's the context you're missing, despite it being in some of the first sentences of the linked article.
Lots to unpack here but I'll just say that I think it would probably matter to a lot of people if they were forced to use something that increased their pointless work by 35%, regardless of whether their work mattered to you or not.
People keep offering this hypothetical 10% acceptable false positive rate, but the article says it’s more like 35%. Imagine if your workplace implemented AI and it created 35% more unfruitful work for you. It might not seem like an “unqualified good” as it’s been referred to elsewhere.
That's only true if you see politics as hewing to the two party paradigm. Sam Altman, Elon Musk et al., aren't liberals or conservatives, they're capitalists. Their ideological consistency, their only alignment, is to the growth of their individual wealth. That is and always has been their team.
Besides this just... not really being true of Lynch's films, it also misses how a Lynch film came to be. David Lynch didn't pull out a notepad and go, "Okay, time to write a Lynch film." He just decided to make a film and made it the way that made the most sense to him. His perspective on film-making, on people, compassion, curiosity etc. led the creation. At best (and this is seriously stretching the definition of "not just wild speculation") I think it's more likely you get something that exhibits formal qualities that remind someone of some of David Lynch's works, probably a fairly incoherent and incongruous mash of tropes, like a particularly bad episode of Family Guy. I don't know if I'd even credit that with praise as dismal as "imitation", it would be more like a visual laundry list. I'd love to see something that would cause someone to believe otherwise besides "AI learned how to string a sentence together, therefore one day it will be able to learn and produce literally anything," but it eludes me.
A lot of people in this thread have an extremely facile understanding of art, David Lynch, and the capabilities of generative AI, but I look forward to someone returning to this thread in 10 or 100 years or whatever to link me to a convincing example.
Well that's one of the wonderful things about saying "one day AI will be able to do x," you don't really need to have any evidence for some segment of the population to believe it's not just plausible but inevitable.
This actually seems far less likely than for other directors. Lynch's style is so idiosyncratic it has its own name. AI has predominantly been good at synthesizing a broad gesture at a wide range of styles. Not really suited for recreating the effect of such a singular artist.
There are extreme ideological distinctions between Facebook and Tumblr that you're ignoring. Suggesting ideology only came into it with Trump and that the Threads pushback was about the spectre of a competitor's popularity is very strange. As is the "Fedi admins are petty, but not me, a Fedi admin" commentary. People had well-founded and widely varied concerns about Meta joining the fediverse. It had nothing to do with wanting to remain in obscurity; just about every fedi user I've ever encountered is absolutely begging everyone on social media to join.
> The Fediverse's structure is already a lot more centralized than anyone would like to admit
Hi, I'm "anyone", and: no it isn't? Technically and philosophically, no it isn't. Large instances are not a failure of decentralisation, because as you evidently know, their existence does not preclude you or anyone else setting up a server and federating. There's no universe in which "scaling the network" reduces decentralisation, in fact it's the solution to the "fuck Matt Mullenweg" situation you speculate about. The less homogenous fedi admins become, the more the network is sustained by people who don't even know who he is. Your comment is rife with generalisations about the motivations of people on the fediverse, but it's far less monolithic than you think.