vidarh

The age matters less than the power-dynamics of her being his nanny.

Except if they were it'd be well known, and no startup typically has contracts that doesn't involve approvals for secondary sales at this kind of early stage because increasing the number of people on the cap table enough triggers nearly the same reporting requirements as being public, and is a massive burden. Just doesn't work that way.

It's also hilarious that you take posting an article that is at best neutral, with a message of doom and gloom about risks to their business, on Lemmy is something OpenAI would have any interest in. If I wanted to pump OpenAI there are better places to do it, and more positive spins to put on it.

Lol, what. OpenAI shares aren't available - there'd be no benefit to anyone trying to pump them.

You can't really trust anything a human says because we're frequently wrong yet convinced we're right, or not nearly as competent as we think, yet we manage, because in a whole lot of endeavours being right often enough and being able to verify answers is sufficient.

There are plenty of situations where they are "right enough" and/or where checking the output is trivial enough. E.g. for software development, where I can easily tell if the output is "right enough", and where humans are often wrong, and where we rely on tests to verify correctness anyway.

Having to cross-check results is a nuisance, but when I can e.g. run things past it on subjects I know well enough to tell if the answers are bullshit and where it can often produce answers better than a lot of actual software developers, it's worth it. E.g. I recently had it give me a refresher on the algorithms to convert an Non-deterministic finite automata (NFA) to a deterministic finite automata (DFA) and it explained it perfectly (which is not a surprise; there will be plenty of material on that subject), but unlike if I'd just looked it up in google, I could also construct examples to test that I remembered it right and have it produce the expected output (which, yes, I verified was correct).

I also regularly has it write full functions. I have a web application where it has written ca 80% of the code without intervention from me. Plenty of my libraries now have functions it has written.

I use it regularly. It's saving me more than enough time to justify both the subscription to ChatGPT and API fees for other use.

As such, it is "actually useful" for me, and for many others.

Bubble in the sense that "many companies will fail" we can agree on. Companies like OpenAI will survive - lawsuits or not - and even if they were to fail due to the lawsuits the algorithms are known and e.g. Microsoft, who has a license to the tech would just hire the team and start over and let the corporate entity go bankrupt.

But all of the "ChatGPT for field X" companies that are just razor-thin layers on top of OpenAI's API, sure, they will almost all fail, and the only ones of them that won't will be the ones that leverages initial investment into an opportunity to quickly pivot into something more substantial.

A lot of people talk about AI as a bubble in the sense of believing the tech will go away, though, and that will never happen, because it's useful enough.

Regarding OpenAI's market cap, I don't agree - I think it'll increase far more, unless they massively misstep, because even though it's riding high on hype, they also still have big lead not down to their hype but down to actually being significantly ahead of even competitors like Google, and given the high P/E ratios in tech they don't need to be the backend all that many big deployments behind big companies even just to field really stupid-simple uses that don't really need the capabilities of GPT before they'll justify that valuation.

Whenever I see them described as "plagiarism machines", odds are about 99% that the person using the term have no idea how these models work. Like with humans, they can overfit, but most of what they output will have have far less in common with any individual work than levels of imitations people engage in without being accused of plagiarism all the time.

As for the environmental effects, it's a totally ridiculous claim - the GPUs used to train even the top of the line ChatGPT models adds up to a tiny rounding error of the power use of even middling online games, and training has only gotten more efficient since.

E.g. researchers at Oak Ridge National Labs published a paper in December after having trained a GPT4 scale model with only 3k GPUs on the Frontier supercomputer using Megatron-DeepSpeed. 3k GPUs is about 8% of Frontiers capacity, and while Frontier is currently fastest, there are hundreds of supercomputers at that kind of scale publicly known about, and many more that are not. Never mind the many millions of GPUs not part of any supercomputer.

You can't. The cat is out of the bag. The algorithms are well understood, and new papers on ways to improve output of far smaller models come out every day. It's just a question of time before training competitive models will be doable for companies in a whole range of jurisdictions entirely unlikely to care.

Why do you think anything will "burst"? If anything, if licensing requirements for contents makes training expensive it's likely to make the biggest existing players far more valuable.

Possibly. On the other hand, OpenAI's market cap is bigger than the ten largest publishers combined - despite their whining they can afford to. It's not OpenAI that will be prevented from getting training data - the biggest impact will be that it might stop smaller competitors and prevent open-source models.

I feel like you are one of the people who feel that AI is just going to be the future with no real problems to anyone who matters. We can’t stop it, we can’t regulate it in any way whatever; and people should just move out of the way, give up and if they can’t find a place in the new world, die already. Artists don’t matter, writers don’t matter and anyone impacted by this new system doesn’t matter. The algorithm is all that matters.

If I thought that, I wouldn't have emphasised the need to sort out the funding issue, and argued that just regulation will be insufficient to solve it.

I think it will cause a massive degree of upheaval. I don't think regulation has any hope in hell of preventing upheaval significant enough that unless a solution is found to ensure better distribution of wealth it will cause violence and uprisings and governments to fall. Not necessarily in and of itself, but in accelerating a process of reducing the monetary value of labour.

I can’t know anything about LLMs, machine learning or anything about this.

I've not suggested anything of the sort.

How you can interpret anything I've written as suggesting I don't think there will be problems is beyond me.

You therefore throw out the idea that bias exists due to tagging systems.

I've done no such thing.

Quick iteration is definitely the big thing. (The eye is fun because it's so "badly designed" - we're stuck in a local maxima that just happens to be "good enough" for us to not overcome the big glaring problems)

And yes, if all the inputs are corrupted, the output will likely be too. But 1) they won't all be, and as long as there's a good mix that will "teach" the network over time that the difference between a "corrupted cat" and an "uncorrupted cat" are irrelevant, because both will have most of the same labels associated with them. 2) these tools work by introducing corruption that humans aren't meant to notice, so if the output has the same kind of corruption it doesn't matter. It only matters to the extent the network "miscorrupts" the output in ways we do notice enough so that it becomes a cost drag on training to train it out.

But you can improve on that pretty much with feedback: Train a small network to recognize corruption, and then feed corrupted images back in as negative examples to teach it that those specific things are particularly bad.

Picking up and labelling small sample sets of types of corruption humans will notice is pretty much the worst case realistic effect these tools will end up having. But each such countermeasure will contribute to training sets that make further corruption progressively harder. Ultimately these tools are strictly limited because they can't introduce anything that makes the images uglier to humans, and so you "just" need to teach the models more about the limits of human vision, and in the long run that will benefit the models in any case.

So what you are saying is open ai should get the public grants for artists to give to artists?

No. What in the world gave you that idea? I'm saying artists or companies employing artists should get grants, just like is the case for a large number of grants now. I'm saying I'd like to see more of that to compensate for the effects being liberal about copyright would have.

I understand it isn’t trained for anything, I have done training with them. The training leads to homogeneous outcomes. It had been studied as well. You can look it up.

There is no "the training". There are a huge range of models trained with different intent producing a wide variety in output to the point that some produces output that others will just plain refuse.

Dall-e 3 still isn’t good enough to be competitive.

Dall-E 3 isn't anywhere near leading edge of diffusion models. It's OpenAI playing catch up. Now, neither Midjourney or Firefly, nor any of the plethora of Stable Diffusion derived models are good enough to be competitive with everyone without significant effort either, today, but that is also entirely irrelevant. Diffusion models are two years old, and the pace of the progress have been staggering, to the point where we e.g. already have had plenty of book-covers and the like using them. Part of the reason for that is that you can continue training of a decent diffusion model even on a a somewhat beefy home machine and get a model that fits your needs better to an extent you can't yet do with LLMs.

Asking and crediting would go a long way to help fix the financial challenge. Because it is a start to adding a financial component. If you have to credit someone there becomes an obligation to that person.

If there is a chance crediting someone will lead to a financial obligation, people will very quickly do the math on how cheaply they can buy works for hire instead. And the vast bulk of this is a one-off cost. You don't need to continue adding images to teach the models already known thing, so the potential payout on the basis of creating some sort of obligation. Any plan for fixing the financial challenge that hinges on copyright is a lost cause from the start because unless it's a pittance it creates an inherent incentive for AI companies to buy themselves out of that obligation instead. It won't be expensive.