"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not.
-
@janef0421 @Gargron most of those problems and quality control issues evaporate if you just *read* and review the generated code. That’s the primary difference between agentic engineering and vibe coding. The inference compute is not as subsidized as you might think, especially for properly quantized open weight models vs frontier. Will frontier training slow down post bubble and prices for compute raise noticeably? Of course. But it will almost certainly still pay for itself.
@mellowfish @Gargron Code review hasn’t resolved issues in human generated code, so I doubt that claim immensely. Maybe you perceive that your review efforts resolve all issues, but that’s a psychological bias - You’re only aware of the issues you identify, and so it appears you’ve identified all issues.
If some series of well-designed studies come out and confirm your claim, then I’ll believe it, but for now, it’s just an assertion you made.
Also, your statements about compute subsidisation don’t contradict what I’m saying, so I see no reason to respond.
-
@drahardja @Gargron as someone who cares a lot about code quality and reads and understands every line of code I ship to prod, AI coding has (still) been a fantastic tool. I ship 2-3 times as much code as I have done in recent years with no real drop in quality. No increase in number/severity of bugs (there are always some no matter who is writing/testing) and the code is still easy to change over time. I can’t imagine going back.
@mellowfish @Gargron Your experience and mine differ.
-
Not relevant to whether or not it has failed or works poorly, which is what the original post was about
@golemwire @Nead There are plenty of times when it fails miserably too.
Maybe on this problem it works, with an expensive model, and with the help of another piece of software that does all the heavy lifting. All it must do is what computers do best, repetitive work, in this case, backtracing spaghetti code, where it can clearly validate its success.
But for many programming projects, success and failure do not have this self-proving nature handed to them on a silver platter
-
It isn't failed, it works quite well in many cases as a matter of fact. Reverse-engineering example: https://www.youtube.com/watch?v=wIe3eDfGKUo
The marketing is annoying, but the tech has legitimate usecases and is here to stay (perhaps not to the extent the companies want us to believe).@golemwire @Gargron
Evaluation of whether something has "failed" or "works poorly" definitely requires consideration of the costs (financial and otherwise, direct and indirect).
If a company sells a car that gets 60 mpg, but also kills three bald eagles per mile driven, that has obviously "failed" or "works poorly". -
@golemwire @Gargron
Evaluation of whether something has "failed" or "works poorly" definitely requires consideration of the costs (financial and otherwise, direct and indirect).
If a company sells a car that gets 60 mpg, but also kills three bald eagles per mile driven, that has obviously "failed" or "works poorly".@golemwire @Gargron
LLMs are also a failure in that they give completely authoritative sounding output that is often wrong, and it can take as much work to determine whether they are correct as to produce the desired results without LLM. -
@golemwire @Gargron
LLMs are also a failure in that they give completely authoritative sounding output that is often wrong, and it can take as much work to determine whether they are correct as to produce the desired results without LLM.@golemwire @Gargron
The fact that most of the direct financial costs of LLMs are currently borne by investors (including your pension fund) rather than end users does not in any way diminish the influence of costs on any realistic evaluation of whether LLMs can be said to be a failure or to work poorly. -
@skavish @Gargron We live in very... Curious times at the moment.
There's a massive hype to adopt something that many don't want to adopt, nor should they have to do so, but equally there's a hype train that says 'do this thing or else...'
There's a middle ground where you can use it quietly, and not use it quietly, and it really doesn't matter.
It's the either siders I'm concerned about.When those that hate it hear about it 24-7 it's like passive smoking. It sucks, it's bad for you and you wish you could escape it.
Equally when you use it for something meaningful and before you've had a chance to explain, you've been shut down because tolerance is 'agree with me or else' that equally sucks.I haven't stated which side of the fence I fall on, and I won't here without being asked. I'm not here to persuade or dissuade, just to make a point.
Moderation is good.You can see my thoughts by looking at my personal posts.
Both of you in this conversation have made your decision and that's OK, but there's nuance in everything and dismissing either side out-of-hand means you aren't thoughtful enough to know where that nuance lies.
You = general you, not you two personally.
I see a lot of love and hate recently, sadly more hate than love, more intolerance than tolerance, and that's not really the fedi I signed up for.
If we could step back about 6 years and show one another just a tad more respect, the world would genuinely be a nicer place to live in.@FreakyFwoof @skavish@welley.social @Gargron the problem is the use of AI isn't harmless, the training of AI isn't harmless, regardless of the perceived utility. This isn't a time to let bygones be bygones when the scale of harm is catastrophic. The choice to use it makes "you" complicit in the harms being shouldered by everyone.
-
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
@Gargron I've used some AI successfully, but it was more of a "Small Language Model" my health care provider has as a replacement for a phone tree. Instead of "Press 1 for ..." I just said (in reply to an email asking for a phone call), "Confirm an appointment", and it then listed the appointment, ending by describing what it was supposed to do and asking if this was correct (yes or no).
It was OK - maybe it reduced their costs but it saved me time. It had to tell "yes" from "no" to be safe.
-
@golemwire @Nead There are plenty of times when it fails miserably too.
Maybe on this problem it works, with an expensive model, and with the help of another piece of software that does all the heavy lifting. All it must do is what computers do best, repetitive work, in this case, backtracing spaghetti code, where it can clearly validate its success.
But for many programming projects, success and failure do not have this self-proving nature handed to them on a silver platter
@golemwire @Nead Also, he is getting paid by Sentry, who sells AI products, so he has a conflict of interest.
-
I am pretty sure @Gargron could get great results if he wanted to use these tools. I completely respect his choice not to. but "I don't want to use it" and "it's a bad tool" are two very different things.
@skavish @Gargron @clayote I think people from tech/engineering backgrounds are used to making instrumentalist arguments, so it is difficult to admit that something they don’t like and may affect them negatively is useful. Some in the open source community are experiencing cognitive dissonance. In fact, the best arguments against AI, such as those made in the Pope’s encyclical or lawsuits brought by writers and artists, conflict with their strongly held beliefs.
-
@golemwire @Nead Also, he is getting paid by Sentry, who sells AI products, so he has a conflict of interest.
That's a good observation. If it were just him, that'd be a problem, but it's not just him. Consider looking into e.g. PhotoCraft or other projects using AI-assisted reverse-engineering of proprietary software. It's gotten pretty big pretty fast.
-
@golemwire @Nead There are plenty of times when it fails miserably too.
Maybe on this problem it works, with an expensive model, and with the help of another piece of software that does all the heavy lifting. All it must do is what computers do best, repetitive work, in this case, backtracing spaghetti code, where it can clearly validate its success.
But for many programming projects, success and failure do not have this self-proving nature handed to them on a silver platter
I agree. AI is way overused and overhyped. I'm arguing for it not being a failure. It has many places it fails.
-
@Gargron @imhotep @skavish On the other hand you say in another post, that you have never used AI, but maintaining one of the biggest FOSS project on GitHub is enough expertise.
Im no Dev, but I assume Linux may be a even bigger FOSS project and Linus explored AI. According to your argumentation, he is as capable as you, if not more, to make guesses about GenAI.
Or what’s the difference, that make your opinion superior to his one?
-
@Gargron @imhotep @skavish On the other hand you say in another post, that you have never used AI, but maintaining one of the biggest FOSS project on GitHub is enough expertise.
Im no Dev, but I assume Linux may be a even bigger FOSS project and Linus explored AI. According to your argumentation, he is as capable as you, if not more, to make guesses about GenAI.
Or what’s the difference, that make your opinion superior to his one?
-
@Gargron @mellowfish Eugen, you worded that far more diplomatically than I could have managed. I had to delete my reply.
@zalasur @Gargron @mellowfish Agreed!
-
@golemwire @Gargron
LLMs are also a failure in that they give completely authoritative sounding output that is often wrong, and it can take as much work to determine whether they are correct as to produce the desired results without LLM.I do believe that AIs are failure when it comes to generic question-answering, like how one's parents probably use ChatGPT right now.... But my interactions with LLMs are usually with the LLM in a harness, with the LLM operating it. That they're not a failure at.
E.g. Kagi Assistant -- I'll give a long rambling question like "There's an article online where the director, a writer, or something, for Cloudy with a Chance of Meatballs says something about Flint needing more compute (I'm sure it was phrased differently). Where's that article?" That was my last query, I had tried searching the web normally but to no avail. The LLM generated a plan and then used Kagi's harness to perform web searches based on the plan to find what I wanted. https://assistant.kagi.com/share/8fa7d889-f4a5-4b31-8898-8ab8cc13658e (I'm realizing that the LLM basically mocked vibecoders at the end of that thread, LOL, kinda deserved)
My point is that they are pretty much failures at raw question-answer, due to hallucinations, but are great with most of the other tasks you'd use a harness with.
If they're supposed to be everything marketers want them to be, they're definitely a failure. Many vibecoders ignore quality. I pretty much just ignore them and find something actually useful to me.
CC: @Gargron@mastodon.social
-
@Gargron just popping in to say thanks for this post, it's been very useful for #WhackATroll
-
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
@Gargron interesting times. Will the bubble burst? I guess we'll have to wait and see.
-
@skavish I find takes like this increasingly hilarious. There’s the magical mythical engineers “doing it right” while the vast majority of users - by my own observation - are pumping out absolute trash. The incentive for quality just does not exist. Some people can build quality systems in spite of that but it is not the norm.
@dandean @skavish these two posts sums it up for me, you both get a star
For what is wurth: @skavish your point is being pulled out of context a couple of times and that is a shame. You are genuielly trying to provide a perspective/insight.
And you’re wright: ‘only if you open-mindly try it.. ‘
@dandean thx for the clearity
-
@Gargron
I am fascinated by your ad hominem attack, but I'm not here to score points or win a right-or-wrong debate.Why not address the substance of my response and engage in a fair debate based on facts and experiences?
Let me start:
What if a technology doesn't need to provide 100% reliable information to be useful, provided its limitations are understood and appropriately managed?And what if future developments improve the reliability of these models and address some of their current limitations? Would AI still be inherently dangerous and irresponsible as a technological tool?
I'm genuinely interested in the reasoning behind your conclusion that AI is a failed technology.
@Luminous_J
"If it was something else, would it still be the same?"We can talk about that when it is that something else. For now, and for any moment in the present, it is what it is.
@Gargron
️
