I feel this pain.

I’ve been trying to get simple telemetry working over lora on a ESP32-C6, LLMs are largely worthless in this. We gotta fall back to old school RTFM models

I used 3.7 on a project yesterday (refactoring to use a different library). I provided the documentation and examples in the initial context and it re-factored the code correctly. It took the agent about 20 minutes to complete the re-write and it took me about 2 hours to review the changes. It would have taken me the entire day to do the changes manually. The cost was about $10.

It was less successful when I attempted to YOLO the rest of my API credits by giving it a large project (using langchain to create an input device that uses local AI to dictate as if it were a keyboard). Some parts of the codes are correct, the langchain stuff is setup as I would expect. Other parts are simply incorrect and unworkable. It’s assuming that it can bind global hotkeys in Wayland, configuration required editing python files instead of pulling from a configuration file, it created install scripts instead of PKGBUILDs, etcetc.

I liken it to having an eager newbie. It doesn’t know much, makes simple mistakes, but it can handle some busy work provided that it is supervised.

I’m less worried about AI taking my job then my job turning into being a middle-manager for AI teams.

I’m using a 3080 and I do seem to have the occasional problem that my ATI using friends don’t, nothing that has prevented me from playing.

Example: Recently, gamescope hasn’t been working. So I’ve been using xwayland instead. Basically I don’t get HDR or VRR, but it’ll be fixed (it may already be fixed because I’m on a rolling release distro (Arch, btw)).

I haven’t, yet, run into a ‘This game doesn’t work because you have an NVIDIA card’. On the other hand, I can run local AI trivially easy because my card supports CUDA.

HEY, I’m typin’ hea’

It’s interesting because it lets regular people with little to no training produce software.

It isn’t great software, it doesn’t always work… but considering the amount of software that an untrained person could produce previously was exactly none, it’s a pretty interesting development in technology.

and, it’s only going to get better.

I dunno bro, I can really see how the chicken could be enticed to cross the road by virtue of his chicken instincts. I’m something of a biologist myself

I can’t read anything containing nuance without reducing it to an absurd Twitter comment

I’m sorry, it gets better.

When we’re talking about legal issues, the terms are important.

Copyright violation isn’t stealing. It is, at worse, a civil matter where one party can show how they’ve been harmed and recover damage. In addition, copyright law allows use of the copyrighted work without the author’s permission in some circumstances.

You’re simply stating that ‘AI is stealing’ when that just isn’t true. And, assuming you mean a violation of copyright, if it was a civil violation then exactly how much would the model owe in damages to any given piece of art? This kind of case would have to be litigated as a class action lawsuit and, if your “AI is stealing committing mass copyright violation” theory is correct then there should be a case where this has been successfully litigated, right?

There are a lot of dismissed class action lawsuits on the topic, but you can’t find any major cases where this issue has been resolved according to your “AI is stealing” claim. On the other hand, there ARE plenty of cases where Machine Learning (the field of which generative AI is a subset) using copyrighted data was ruled as fair use:

(from https://www.cjr.org/the_media_today/an-ai-engine-scans-a-book-is-that-copyright-infringement-or-fair-use.php )

Google has won two important copyright cases that seem relevant to the AI debate. In 2006, the company was sued by Perfect 10, an adult entertainment site that claimed Google had infringed its copyright by generating thumbnail photos of its content; the court ruled that providing images in a search index was “fundamentally different” from simply creating a copy, and that in doing so, Google had provided “a significant benefit to the public.” In the other case, the Authors’ Guild, a professional organization that represents the interests of writers, sued Google for scanning more than twenty million books and showing short snippets of text when people searched for them. In 2013, a judge in that case ruled that Google’s conduct constituted fair use because it was transformative.

Creating a generative model is fundamentally different than copying artwork and it also provides a significant benefit to the public. The AI models are not providing users with copies of the copyrighted work. They’re, literally, transformative.

This isn’t a simple matter of it being automatically wrong and illegal if copyrighted work was used to create the models. Copyright law, and law in general, is more complex than a social media meme like ‘AI is stealing’.

Exactly.

People are focused on art because its easy to meme on and playing ‘spot the AI’ makes people feel like they’re discerning individuals with refined tastes. Short clips of generated music is 100% indistinguishable from human-made. Code is invisible to the average user.

AI-based tools have already been rapidly adopted in industries that experience heavy competition, like game development. Essentially all professional tools include either integrated generation or support for using generative models. Coding is no exception.

It isn’t the case that “AI will soon be used by developers unless we stop them”. We’re living in the “AI is being used and only rarely spotted” age now.

I think most customers want a fun game that doesn’t cost $120.

I’m not against AI as a tool, but don’t assume people will like your game more if you plaster it with AI art. It’s like coloring your sketch with stickers. The stickers may be good quality, but it will still look like a messy puzzle…

If your game is good and fun even crappy art will sell it (look at touhou).

dry sarcasm.

I only come to Lemmy to post dry sarcasm without a /s because I’m a rebel

I’m not quite sure I’m following.

Are you saying that AI trained on the output of humans is unethical, unless those humans are programmers?

Or, as a professional programmer, you understand the limitations of AI in your field so you don’t feel threatened by it while simultaneously assuming, on behalf of another profession, that AI in “artistic” fields is somehow far more capable and an actual threat?

Terrible programmers don’t become professional programmers because they subscribe to Copilot. It provides a crutch to absolute beginners, allowing even the least skilled individual to create some low quality output. For professionals, AI allows for some aspects of existing tools to perform slightly better but cannot replace the knowledge, experience and practice of a human when it comes to applying those skills in novel and interesting ways.

Terrible artists don’t become professional artists because they subscribe to Midjourney. It provides a crutch to absolute beginners, allowing even the least skilled individual to create some low quality output. For professionals, AI allows for some aspects of existing tools to perform slightly better but cannot replace the knowledge, experience and practice of a human when it comes to applying those skills in novel and interesting ways.

Also, screw the independent developer who doesn’t have artists to lay off nor the budget to hire them.

If they want to make a game then they should spend decades learning to program and decades learning to create art and decades learning to create music.

If they use AI to make code or assets then it completely invalidates their work and the fun that I’m having with their game is just fake fun.

The only Real Games are those made by giant corporations with the capital to hire artists, programmers and musicians that can lovingly hand craft the loot boxes for the next major children’s casino.

e: Honestly, it’s embarrassing that I have to add a /s for people to understand

My game was going to be good but then AI happened and now I only make terrible games

Your basing your entire argument on the assumptions that every generative models is trained on copyright works and also that training AI on copyrighted works is not Fair Use.

The first assumption is just false and the second assumption is not built on any established legal grounds in Western countries and is completely false in other countries with different legal systems.

It’s like giving people the choice to not eat carbon because some forms, like natural diamonds, are exploitative. People cheer because nobody wants to eat pencil dust since it tastes so bad.

It sounds like a great idea as long as you don’t understand how anything works.

(For the people that don’t understand how anything works: Carbon is an integral component of literally everything that you will ever eat)

I am pretty sure that if asked, the serverside protections can be circumvented

No, they literally cannot. The entire protocol is open sourced and has been audited many times over.

One of the fundamental things you assume when designing a cryptosystem is that the communication link between two parties is monitored. The server mostly exists as a tool to frustrate efforts by attackers that have network dominance (i.e. secret police in oppressive regimes) by not allowing signals intelligence to extract a social graph. All this hypothetical attacker can see is that everyone talks to a server so they can’t know which two people are communicating.

The previous iteration, TextSecure, used SMS. Your cellular provider could easily know WHO you were talking to and WHEN each message was sent. So SMS was replaced with a server and the protocol was amended so that even the server has no way of gaining access to that information.

The sealed sender feature is something that the client does. It was best effort because, at the time, they still supported older clients and needed backwards compatibility. This is no longer the case.

It is already at that point.

People only notice the generated works that they notice, they don’t notice the generated elements that they don’t notice.

They assume that they can “just tell” if generative AI was used, but the reality is that it’s being used in a lot of development processes in place of human effort. Things like generative fill in Photoshop or making variations of a texture are 100x faster to do with AI tools and are used all the time.

This is definitely a topic where a vast majority of people have been “informed” of their opinions by social media memes instead of through a reasoned examination of the situation.

People who’re probably too young to have ever lived through major technology breakthroughs.

This same “debate” always happens. When digital cameras were being developed, their users were seen as posers encroaching on the terf of “Real Photographers”.

You’d hear “Now just anybody can take pictures and call themselves a photographer?”

Or “It takes no skill to take a digital photograph, you can just manipulate the image in Photoshop to create a fake image that Real Photographers have to work years developing the skills to capture”

Computers were things that some people, reluctantly, had to use for business but could never be useful to the average person. Smartphones were ridiculous toys for out of touch tech nerds. Social Media was an oxymoron because social people don’t use the Internet. GPS is just a toy for hikers and people that are too dumb to own paper maps. Etc, etc, etc

It’s the same neo-luddite gatekeeping that’s happening towards AI. Any technology that puts capabilities in the hands of regular people is viewed by some people as fundamentally stealing from professionals.

And, since the predictable response is to make some arcane copyright claim and declare training “stealing”: Not all AI is trained on copyrighted materials.

Sure, so let me export my data from another PC or phone. If they wanted you to have message history, they would. So I’ll respectfully disagree.

https://signal.miraheze.org/wiki/How_to_move_Signal_Desktop_message_history_to_another_computer_(or_during_an_OS_reinstall)