I feel like that is a sort of protection that should never, ever be granted under any circumstances. Like you’d better feel comfortable going in front of a judge and explaining how you made every reasonable effort to protect children or you need to stay the fuck away from children.

When I saw that Maciej still uses X, that was it for me.

I think this is the key part for me. If I was on the fence about someone or something, X would definitely push me over the edge. I would like to see more of us publicly swearing off services for associating with awful people.

Ironic, of course, that the best place to get that message across would be X…

You can still build components and assemble them once each individual piece works. It’s easier to start with something that works and add to it than start with a “fully realized” script none of which actually works.

Asking folks to review hundreds of lines of vibe-coded slop (anything, really, but especially vibe coded slop) is a big ask. I sincerely wish you luck, though. We all started somewhere and maybe as you work to fix this you’ll become something of a coder yourself.

For being not a coder, this is fairly ambitious. I’ve been coding software for thirty years and I’m not sure I can offer anything technical. Personally, I’d probably try to do this in Python rather than shell script. Shell script is fine if you have tools you’re just trying to stitch together, but this is far beyond that.

That being said, I get it. The first coding I ever did was making a character sheet for roleplaying by sending raw commands to my dot matrix printer from my Commodore 64. You have a vision in your head and a computer at your fingertips and you have to bring it all together.

I respect that and if part of your goal is to achieve this using only bash, then god speed. Otherwise I might look at a language that gives you library support and modularity and unit testing. I write Python scripts all the time, and the truth is I don’t know Python at all, so I know it can be done. Define a unit of functionality and build that piece. When that piece works build another component. Then build a parent program that calls those two things. Then build more components. Define functions that can be reused when you have code that needs to do similar things.

The key is to build standalone functionality — what is the smallest useful thing you can do? Build it. Now you can invoke that every time you need to do the thing. Construct your program out of components you create.

This is kinda like if I told AI to build me a web service but keep everything in a single Java class. It would be a wild and unreadable, untestable mess.

Roads have been around for thousands of years, but not really the type of roads he’s talking about. Especially since he doesn’t remember them before cars anyway.

Is he suggesting that playing in the streets was something he used to do before the advent of cars? Like there were all that roads everywhere with kids just playing on them willy nilly, until one day Henry Ford met General Motor and said, “You know what we oughta do with all these fuckin’ roads?” And ever since then, kids haven’t been able to play on them any more?

Is that the version of history we’re pedaling here?

Ever if the bubble pops, AI isn’t going to go away. Profits will never justify the current expenditure, but they will make more efficient AI and it’ll still be everywhere. Just probably not quite as good and enshittified to the core.

I can only hope open models can deliver what corporations will not.

That’s not really a to the point I was making, but as you’ve brought it up, he said that in the midst of refusing to accede to the US militaries demands that they get to use AI for mass surveillance and autonomous weapons.

You’ve snipped his words tactically before the pivot to refusal. He said that private companies have no say in how a military chooses to wage war however they would not allow the use of their AI for those two things.

And he’s right. If Lockheed decided they didn’t like Obama or a military decision of his and grounded their drones to punish him, that would be inappropriate. Lockheed nor Anthropic get to dictate US military policy. That (among other things) is why we are supposed to elect a President with the wherewithal to make those decisions.

Now, I don’t know Dario. He’s a rich CEO and probably he belongs at the bottom of the sea with the rest. But Anthropic refused to allow their tech to be used that way, and that’s more than Google or OpenAI, both of whom were audibly salivating to supplant them. Anthropic seems to be the least bad out of all the options.

That being said, I’m placing my hopes in local AI and getting rid of all the big players, including Anthropic. But in the meanwhile, I do respect them for taking a stand none of their competitors would.

“Anthropic has not done a great job at trying to speak to the administration and appreciate the ideological differences,” one source familiar with the administration’s thinking said.

“It’s like they just speak in different languages,” the source said, adding that the company has simply not figured out how to communicate with this administration.

“Why can’t Anthropic just understand they have to bribe Trump? If he isn’t getting a big piece of your action - you aren’t getting any.”

Well that’s some shit I’m gonna toss in the garbage. All of the streaming in the house is done on PCs or Playstations currently, so it won’t be a big loss. I have a few devices sitting around to take for travel or whatever, but I can find other solutions.

Dear corporate America:

I reject your enshittification. I don’t need your service. Fuck yourself with an axe-head.

– Shel

I don’t have an immediate need for this, but I could see it in the future when I upgrade some local hardware. That being said +1 for Podman support.

AI should legally be considered like asking expert advice, like asking a lawyer or a doctor

That is impossible. AI has no consciousness or ability to reason about the answers it is giving. Without a thinking domain expert between the user and the AI, this simply cannot be done.

If AI responses can’t be trusted and are false information

AI response can’t be trusted. Everyone should know that. There is no possible way to design a system where AI can be trusted. Hell, you can’t even design a system where a human can be trusted to be infallible.

You don’t need a particular skill to use AI as a service — which isn’t to say the results are going to be worth a crap if you aren’t a domain expert. Anyone can ask AI to build a web service and get it approximately correct, but you need someone who knows how to build a web service to tell the AI what it needs to build it correctly.

But one does need to understand that the output from AI should only be used when the marginal cost of failure is near-zero.

Which means 99.9% of the users currently using AI, or unknowingly exposed to it through services that use AI without it being clear.

It’s tricky, you know? IMO, you can’t democratize AI (which IMO we must do) if you need some sort of IQ test to be allowed to use it. A checkbox acknowledging risk acceptance is insufficient. But also, yes, you have to have some awareness of what AI even is or you are at risk of being harmed.

I didn’t downvote, as I don’t do that to comments that show thoughtful effort, whether I disagree or not. And I’m certainly not in lockstep with Lemmy on AI, so I can’t speak for anyone but me.

There are some places I agree, and some I don’t. First, I agree that the three laws are a good idea. However, they are compressed in human language and it is unsurprisingly ambiguous as to the meaning of harm. While harm is often unambiguous, I have a feeling if you could somehow survey the world on all possible actions, no two people would completely agree on what is harmful. If we can’t clearly define it for ourselves, we certainly can’t choose to it or create an AI that agrees. So as nice as the idea of the three laws is, they are not feasible.

I think you are spot on about trust and AI. My computer has prompted me a thousand times for approval of some innocuous text expansion in a cli command. It has never attempted something I didn’t want. I do often have to go back and change how it has done something. And I run frequent scans for alignment of documentation so that I can be sure someone looking at the acceptance criteria doesn’t come away with a different idea than someone who looks at the diagrams.

But we diverge on matters of responsibility. I think it must be incumbent upon the person using the AI to understand the fundamental flaws of such systems. And I think a company that provides unreviewed AI output must ensure users are aware that output can be harmful or incorrect. Ultimately the nature of these systems can only approximately comply with lack of harm.

I think responsibility comes once you present AI output to something who doesn’t consent or understand AI. (Side note: making it a requirement that a user must be 18 to be provided AI output might fix a lot of issues because for example customer service bots can’t guarantee a caller is over 18.)

The ways to use AI are myriad and it’s impossible to know how “come home to me” (the text which encouraged one of the suicides) is going to be taken by the user. In fact, I’d interpret that phrase as imploring the user to stay alive.

Ultimately, “no harm” can never be guaranteed because life is subjective. And no AI can be rendered harmless yet still do anything useful. That’s why human judgment and skepticism is needed on the output. I am hopeful AI can be fully democratized, but critical thinking is always required, just like the time I had a high school teacher tell me blind people can see through a sort of third eye.

That doesn’t obviate the AI company of any responsibility whatsoever, but I think it’s fine to hold them to standards of reasonable effort to mitigate foreseeable harms.

I guess there’s no need to make better AI, then. It’s already as good as it’s allowed to be. Now we can focus on running existing models more efficiently and maybe local. There’s that chip company who said they could 10x the energy efficiency of AI if it were built into the chip, but it would mean losing the innovation of rapidly iterating models. Well, this is the moment. If we could have local models that ran for the cost of charging your car and everyone had access to them, maybe that would give us a measure of what AI has always promised mankind.

I have. I guess he knew it, too , since he tried to pass generated code off as his own without understanding it. It was very awkward when he asked me for a recommendation.

I used fable once to find discrepancies in all the documentation for a project. Cost me $10 and took about ten minutes. It caught 31 more issues after I fixed the 17 opus found. Still I can’t say how long that would’ve taken by hand.

Now I have a long weekend of fixing project docs before Monday.

I cringed when my coworker told me he’d been using it all day. 80% of what we use AI for can be done by Sonnet. He probably spent a couple hundred bucks or more. No idea what leadership is going to make of that decision. They have so far trusted us to be budget conscious, but that’ll definitely change if we have people spending $1k/week.

That being said, my dev teams are all offshore. Not that I expect them to use fable much, but it sucks if it’s not even going to be an option. I don’t know how or if we’re going to provide api keys to our teams if the company can’t just make more keys from the account.

It can’t really work the way to want it to.

You have let’s say Samsung who can make money selling a chip as long as the price > $50. And historically the price of the chip has averaged $100.

But the demand is crazy and they can’t keep up and the price of the chip is $500. They are making money hand over fist but let’s say they feel a moral obligation (hahahahaha) to lower the price by increasing capacity.

So they invest a billion dollars to increase capacity. Now that’s a huge cost that reduces their margin on all chips. Between loans and maintenance, now they have to sell a chip for $90 to break even. But that’s fine because they are making $410 per chip instead of $50!

Except now you fix the supply issue and demand falls to normal. You’ve just cut your profit from $50 to $10. You have to sell 5x the volume to make the money you were making!

Except it’s even worse, because now you have all these extra chips you’re building and nowhere to put them. Supply exceeds demand, pushing prices lower so instead of $100, they are selling for $80. Now Samsung loses $10 on every chip and they go bankrupt trying to pay back a billion in loans.

So it’s not really in their interest to build capacity to meet a temporary demand. Unfortunately.

Agreed. Refusal code is an edge that can be exploited.

I day traded for several months. Mostly lost little bits here and there over time. Hit a $5k win on a leveraged bet and went out on a high note. I bet that good earnings by Apple would mean good earnings by MS later that same week. And then I retired from day trading because even the piddly amounts I was making and losing were stressful.

Someone could make a real killing betting on when this administration was going to try to steal another $500m from the market, though — if they got it right.