Like the article itself mentions, it has immense potential for advertising, scams and political propaganda. I haven’t seen AI proponents offering meaningful rebuttals to that.

You won’t get a direct rebuttal because, obviously, an AI can be used to write ads, scams and political propaganda.

But every day millions of people are cut by knives. It hurts. A lot. Sometimes the injuries are fatal. Does that mean knives are evil and ruining the world? I’d argue not. I love my kitchen knives and couldn’t imagine doing without them.

I’d also argue LLMs can be used fact check and uncover scams/political propaganda/etc and can lower the cost of content production to the point where you don’t need awful advertisements to cover the production costs.

Is that the AI’s fault, or the depressed and suicidal human’s fault?

Do you not think that the person would have committed suicide whether they asked the AI or not? The AI might have sped up the decision, but it is the human who made it.

It is not like the AI is out there trying to convince non-depressed humans to become depressed in order they go kill themselves …

What? It’s the sociological definition of technology. A cultural tool which is used by a community for making a task xyz, easier, faster, more efficient.

Efficiency is an extremely broad term.

What’s your counter definition of technology and efficiency that is leading you to disagree?

I don’t have anything against you or your colleagues. You’ve got every right to strike if that’s what you want to do.

But there are millions of people being harmed by the strike. That’s a simple fact.

Journalists/etc need to do their job and provide good balanced information on critical issues like this one. FUD like Drew Devalt posted inflames the debate and makes it nearly impossible for reasonable people to figure out what to do about Large Language Models… because like it or not, they exist, and they’re not going away.

PS: while I’m not a film writer, I am paid to spend my day typing creative works and my industry is also facing upheaval. I also have friends who work in the film industry, so I’m very aware and sympathetic to the issues.

we are spreading FUD

You are spreading FUD right in this post: “poorly regurgitating our shit without our permission”

• There is no need for permission, that’s not how copyright works
• It’s not “poorly regurgitating”, it’s for most part original and of high quality

You wanna be taken serious? Stop repeating the nonsense as everybody else.

And as for “replace us at our jobs”, that’s not a problem, that’s called progress. If you want UBI or something along the line, go fight for that, don’t make stupid arguments against AI.

That does not mean everyone’s critique is ignorant.

Well, maybe, still waiting for reading one that isn’t. Since everybody just keeps repeating the same nonsense, including you right now.

The only sensible one I heard so far was from Hinton, which simply suggested putting about equal money into AI safety research as we do into AI research. Nothing wrong with that, though that will probably just show us more ways in which AI can go wrong and less on how to prevent it.

I mean, people should at least check if the publisher is reliable for any information source.

Now, if said AI is generating foraging books more accurate than humans, that’s fine by me. Until that’s the case, we should be marking AI-generated books in some clear way.

The problem is, the LLM AIs we have today literally cannot do this because they are not thinking machines. These AIs are beefed-up autocompletes without any actual knowledge of the underlying information being conveyed. The sentences are grammatically correct and read (mostly) like we would expect human written words to read, however the actual factual content is non-existent. The appearance of correctness just comes from the fact that the model was trained on information that was (probably mostly) correct in the first place.

I mean, we should still be calling these things algorithms and not “AI” as “AI” carries a lot of subtext in people’s minds. Most people understand “algorithms” to mean math, and that dehumanizes it. If you call something AI, all of a sudden people have sci-fi ideas of truly independent thinking machines. ChatGPT is not that, at all.

That would be UBI and that’s seen as an improvement by a lot of people so why stand in the way of that? Robots do the work, people get a budget they can spend on that work while they don’t have to do it.

deleted by creator

If you disregard the positive side that’s a negative bias. I’m not interested in a semantic fight with you.

Yes, but that’s already the case. There are numerous Open Source’ish language models around that you can run on your own PC, no server required:

• https://old.reddit.com/r/LocalLLaMA/wiki/models/
• https://github.com/ggerganov/llama.cpp
• https://github.com/oobabooga/text-generation-webui

And some of them are getting pretty damn close to ChatGPT performance:

• https://tatsu-lab.github.io/alpaca_eval/

There is of course still plenty of work that needs to be done in letting LLMs interact with the outside world, use a webbrowser and stuff, but there are projects for that as well, e.g. AutoGPT. Just a matter of time until that stuff becomes good enough to be usable.

They’ll certainly try, though it’ll be quite a bit trickier due to LLMs providing direct answers, not just a list of sites. You can’t really sneak a product in there when it doesn’t actually fit the question. I think the bigger problem is just the lack of good information out there. Finding trustworthy reviews these days is getting really hard, most of the time all you have is the product description and some Amazon reviews, which even when done well, fail at providing how product X compares to product Y. No matter how smart the AI will be, that always leaves a ton of room for error and misinformation.

Hard to tell how things will end up. For the time being, LLMs are pretty much completely useless for product search, ChatGPT just doesn’t know enough and BingChat will just summarize the first three SEO-filled Bing search results. The deep knowledge LLMs have on Wikipedia-like topics is missing when it comes to products and services, and they can’t really do calculations either, so price information is almost always wrong. This will need some specific optimization.

Its not a problem if its removed by improved ai; it would be a transient fear that never manifests.

Use the AI to fix the bugs.

A couple months ago, I tried it on ChatGPT: I had never ever written or seen a single line in COBOL… so I asked ChatGPT to write me a program to print the first 10 elements of the Fibonacci series. I copy+pasted it into a COBOL web emulator… and it failed, with some errors. Copy+pasted the errors back to ChatGPT, asked it to fix them, and at the second or third iteration, the program was working as intended.

If an AI were to run with enough context to keep all the requirements for a module, then iterate with input from a test suite, all one would need to write would be the requirements. Use the AI to also write the tests for each requirement, maybe make a library of them, and the core development loop could be reduced to ticking boxes for the requirements you wanted for each module… but maybe an AI could do that too?

Weird times are coming. 😐

That’s where we need to ask how we define “better”. Is better “when the number goes bigger” or is better “when more people benefit”? If an AI can better optimize to better extract the maximum value from people’s work and discard them, then optimize how many ways they can monetize their product to maximize the profit they get from each customer, the result is a horrible company and a horrible society.

In theory yes… but what do we call “doing a better job”? Is it just blindly extracting money? Or is it something more, and do we all agree on what it is? I think there could be a compounded problem of oversight.

Like, right now an employee pays/invests some money into a retirement fund, whose managers invest into several mutual funds, whose managers invest into several companies, whose owners ask for some performance from their C-suite, who through a chain of command tell the same employee what to do. Even though it’s part of the employee’s capital that’s controlling that company, if it takes an action negative for the employee like fracking under their home, or firing them, they’re powerless to do anything about it with their investment.

With AI replacing all those steps, it would all happen much quicker, and —since AIs are still basically a black box— with even less transparency than having corruptible humans on the same steps (at least we kind of know what tends to corrupt humans). Adding strict “code as contract” rules to try to keep them in check, would on a first sight look like an improvement, but in practice any unpredicted behavior could spread blindingly fast over the whole ecosystem, with nobody having a “stop” button anymore. That’s even before considering coding errors and malicious actors.

I guess a possible solution would be requiring every AI to have an external stop trigger, that a judicial system could launch to… possibly paralyze the whole economy. But that would require new legislation to be passed (with AI lawyers), and it would likely get late, and not be fully implemented by those trying to outsmart the system. Replace the judges by AIs too, politicians with AIs, talking heads on TV with AIs… and it becomes an AI world where humans have little to nothing to say. Are humans even of any use, in such a world?

None of those AIs need to be an AGI, so we could run ourselves into a corner with nobody and nothing having a global plan or oversight. Kind of like right now, but worse for the people.

Alternatively, all those AIs could be eco-friendly humans-first compassionate black boxes… but I kind of doubt those are the kind of AIs that current businesses are trying to build.