We could even make isotropically pure iron, jeah.

I don't see how that affects my point.

• Today's AI detector can't tell apart the output of today's LLM.

• Future AI detector WILL be able to tell apart the output of today's LLM.

• Of course, future AI detector won't be able to tell apart the output of future LLM.

So at any point in time, only recent text could be "contaminated". The claim that "all text after 2023 is forever contaminated" just isn't true. Researchers would simply have to be a bit more careful including it.

People still tap into real world while AI does not do that yet. Once AI will be able to actively learn from realworld sensors, the problem might disappear, no?

Yep, the problem was never LLMs, but billionaires and the rich. The problems have always been the rich for thousands of years, and yet they are immensely successful at deflecting their attacks to other groups for those thousands of years. They will claim it's Chinese immigrants, or blacks, or Mexicans, or gays, or trans people. Now LLMs and AI are the new boogieman.

We should be talking about UBI, not LLMs.

It’s a capitalism problem not an AI or copyright problem.

This isn’t a technological issue, it’s a human one

I totally agree with everything you said, and I know that it will never ever happen. Power is used to get more power. Those in power will never give it up, only seek more. They intentionally frame the narrative to make the more ignorant among us believe that the tech is the issue rather than the people that own the tech.

The only way out of this loop is for the working class to rise up and murder these cunts en masse

Viva la revolucion!

It is a completely understandable stance in the face of the economic model, though. Your argument could be fitted to explain why firearms shouldn’t be regulated at all. It isn’t the technology, so we should allow the sale of actual machine guns (outside of weird loopholes) and grenade launchers.

The reality is that the technology is targeted by the people affected by it because we are hopeless in changing the broader system which exists to serve a handful of parasitic non-working vampires at the top of our societies.

Edit: not to suggest that I’m against AI and LLM. I want my fully automated luxury communism and I want it now. However, I get why people are turning against this stuff. They’ve been fucked six ways from Sunday and they know how this is going to end for them.

Plus, a huge amount of AI doomerism is being pushed by the entrenched monied AI players, like OpenAI and Meta, in order to used a captured government to regulate potential competition out of existence.

Exactly. I work in AI (although not the LLM kind, just applying smaller computer vision models), and my belief is that AI can be a great liberator for humanity if we have the right political and economic apparatus. The question is what that apparatus is. Some will say it's an inherent feature of capitalism, but that's not terribly specific, nor does it explain the relatively high wealth equality that existed briefly during the middle of the 20th century in America. I think some historical context is important here.

Historical Precedent

During the Industrial Revolution, we had an unprecedented growth in average labor productivity due to automation. From a naïve perspective, we might expect increasing labor productivity to result in improved quality of life and less working hours. I.e., the spoils of that productivity being felt by all.

But what we saw instead was the workers lived in squalor and abject poverty, while the mega-rich captured those productivity gains and became stupidly wealthy.

Many people at the time took note of this and sought to answer this question: why, in an era over greater-than-ever labor productivity, is there still so much poverty? Clearly all that extra wealth is going somewhere, and if it's not going to the working class, then it's evidently going to the top.

One economist and philosopher, Henry George, wrote a book exploring this very question, Progress and Poverty. His answer, in short, was rent-seeking:

Rent-seeking is the act of growing one's existing wealth by manipulating the social or political environment without creating new wealth.[1] Rent-seeking activities have negative effects on the rest of society. They result in reduced economic efficiency through misallocation of resources, reduced wealth creation, lost government revenue, heightened income inequality,[2] risk of growing political bribery, and potential national decline.

Rent-seeking takes many forms. To list a few examples:

• Land speculation
• Monopolization of finite natural resources (e.g., oil, minerals)
• Offloading negative externalities (e.g., pollution)
• Monopolization of intellectual property
• Regulatory capture
• Monopolistic or oligopolistic control of entire markets

George's argument, essentially, was that the privatization of the economic rents borne of god-given things — be it land, minerals, or ideas — allowed the rich and powerful to extract all that new wealth and funnel it into their own portfolios. George was not the only one to blame these factors as the primary drivers of sky-high inequality; Nobel-prize winning economist Joseph Stiglitz has stated:

Specifically, I suggest that much of the increase in inequality is associated with the growth in rents — including land and exploitation rents (e.g., arising from monopoly power and political influence).

George's proposed remedies were a series of taxes and reforms to return the economic rents of those god-given things to society at large. These include:

• Implementation of land value taxes:

Land value taxes are generally favored by economists as they do not cause economic inefficiency, and reduce inequality.[2] A land value tax is a progressive tax, in that the tax burden falls on land owners, because land ownership is correlated with wealth and income.[3][4] The land value tax has been referred to as "the perfect tax" and the economic efficiency of a land value tax has been accepted since the eighteenth century.

• Implementation of Pigouvian (aka externality) taxes, e.g., carbon tax:

A Pigouvian tax (also spelled Pigovian tax) is a tax on any market activity that generates negative externalities (i.e., external costs incurred by the producer that are not included in the market price). The tax is normally set by the government to correct an undesirable or inefficient market outcome (a market failure) and does so by being set equal to the external marginal cost of the negative externalities. In the presence of negative externalities, social cost includes private cost and external cost caused by negative externalities. This means the social cost of a market activity is not covered by the private cost of the activity. In such a case, the market outcome is not efficient and may lead to over-consumption of the product.[1] Often-cited examples of negative externalities are environmental pollution and increased public healthcare costs associated with tobacco and sugary drink consumption.[2]

• Implementation of severance taxes,

Severance taxes are taxes imposed on the removal of natural resources within a taxing jurisdiction. Severance taxes are most commonly imposed in oil producing states within the United States. Resources that typically incur severance taxes when extracted include oil, natural gas, coal, uranium, and timber. Some jurisdictions use other terms like gross production tax.

such as in the Norwegian model:

The key to Norway’s success in oil exploitation has been the special regime of ownership rights which apply to extraction: the severance tax takes most of those rents, meaning that the people of Norway are the primary beneficiaries of the country’s petroleum wealth. Instead of privatizing the resource rents provided by access to oil, companies make their returns off of the extraction and transportation of the oil, incentivizing them to develop the most efficient technologies and processes rather than simply collecting the resource rents. Exploration and development is subsidized by the Norwegian government in order to maximize the amount of resource rents that can be taxed by the state, while also promoting a highly competitive environment free of the corruption and stagnation that afflicts state-controlled oil companies.

• Intellectual property reform, e.g., abolishing patents and instead subsidizing open R&D, similar to a Pigouvian anti-tax (research has positive externalities) or Norway's subsidization of oil exploration
• Implementation of a citizen's dividend or universal basic income, e.g., the Alaska permanent fund or carbon tax-and-dividend:

Citizen's dividend is a proposed policy based upon the Georgist principle that the natural world is the common property of all people. It is proposed that all citizens receive regular payments (dividends) from revenue raised by leasing or taxing the monopoly of valuable land and other natural resources.

...

This concept is a form of universal basic income (UBI), where the citizen's dividend depends upon the value of natural resources or what could be titled as common goods like location values, seignorage, the electromagnetic spectrum, the industrial use of air (CO2 production), etc.[4]

• Funding public goods via the Henry George Theorem:

In 1977, Joseph Stiglitz showed that under certain conditions, beneficial investments in public goods will increase aggregate land rents by at least as much as the investments' cost.[1] This proposition was dubbed the "Henry George theorem", as it characterizes a situation where Henry George's 'single tax' on land values, is not only efficient, it is also the only tax necessary to finance public expenditures.[2] Henry George had famously advocated for the replacement of all other taxes with a land value tax, arguing that as the location value of land was improved by public works, its economic rent was the most logical source of public revenue.[3]

...

Subsequent studies generalized the principle and found that the theorem holds even after relaxing assumptions.[4] Studies indicate that even existing land prices, which are depressed due to the existing burden of taxation on labor and investment, are great enough to replace taxes at all levels of government.[5][6][7]

(continued)

You make it sound like the quality is what will increase with human/AI partnership. What will realistically happen is an expected rate of output. Why can't you deliver a book every year with an AI ghost writing? People who work slowly or meticulously will be phased out by those that can quickly throw together a collage of their own words and their guided filler. It's amazing and futuristic. It can be very useful and inspirational. But I do not share your optimism that it will make creative industries better. It will allow a single person to put together a script that would've taken a team... But the better content will now be drowning in this sea. Unfortunately, I expect an equivalent of the media explosion that happened when the reality tv format became an ok thing, eventually leading to shows half filmed on phones. The end result will be double the marvel movies every year.

It's mostly large companies, some models are open source (of which only some are also community driven), but the mainstream ones are the ones being entirely funded by, legally protected by, and pushed onto everything by capitalist olligarchs.

What other options do you have? I'm sick and tired of people like you seeing workers lose their jobs, seeing real people used like meat puppets by the internet, seeing so many artists risking their livelihoods, seeing that we'll have to lose faith in everything we see and read because it could be irrecognizably falsified, and CLAIMING you care about it, only to complain every single time any regulation or way to control this is proposed, because you either don't actually care and are just saying it for rhetoric, or you do care but only to the point you can still use your precious little toys restriction-free. Just overthrow the entire economic system of all countries on earth, otherwise don't do anything, let all those people burn! Do you realize how absurd you sound?

It's sociopathic. I don't say it as an insult, I say it applying the definition of a word, it's a complete lack of empathy and care for your fellow human beings, it's viewing an inmaterial piece of technology, nothing but a thoughtless word generator, like inherently worth more than the livelihood of millions. I'm absolutely sick of it. And then you have the audacity to try to seem like the reasonable ones when arguing about this, knowing if you had your way so many would suffer. Framing it as anti-capitalism knowing that if you had your way you'd pave the way for the olligarchs to make so many more billions off of that suffering.

So a few tidbits you reminded me of:

• You're absolutely right: there's what's called an alignment problem between what the human thinks looks superficially like a quality answer and what would actually be a quality answer.

• You're correct in that it will always be somewhat of an arms race to detect generated content, as lossy compression and metadata scrubbing can do a lot to make an image unrecognizable to detectors. A few people are trying to create some sort of integrity check for media files, but it would create more privacy issues than it would solve.

• We've had LLMs for quite some time now. I think the most notable release in recent history, aside from ChatGPT, was GPT2 in 2019, as it introduced a lot of people to to the concept. It was one of the first language models that was truly "large," although they've gotten much bigger since the release of GPT3 in 2020. RLHF and the focus on fine-tuning for chat and instructability wasn't really a thing until the past year.

• Retraining image models on generated imagery does seem to cause problems, but I've noticed fewer issues when people have trained FOSS LLMs on text from OpenAI. In fact, it seems to be a relatively popular way to build training or fine-tuning datasets. Perhaps training a model from scratch could present issues, but generally speaking, training a new model on generated text seems to be less of a problem.

• Critical reading and thinking was always a requirement, as I believe you say, but certainly it's something needed for interpreting the output of LLMs in a factual context. I don't really see LLMs themselves outperforming humans on reasoning at this stage, but the text they generate certainly will make those human traits more of a necessity.

• Most of the text models released by OpenAI are so-called "Generative Pretrained Transformer" models, with the keyword being "transformer." Transformers are a separate model architecture from GANs, but are certainly similar in more than a few ways.

The entropy in text is not good enough to provide enough space for watermarking. No it does not get better in longer text because you have control over i lot/chunking. You have control over top-k and temperature and prompt which creates infinite output space. Open text-generation-webui, go to the parameter page and count the number of parameters you can adjust to guide outcome. In the future you can add wasm encoded grammar to that list too.

Server side hashing / watermarking can be trivially defeated via transformations / emoji injection Latent space positional watermarking breaks easily with post processing. It would also kill any company trying to sell it (Apple be like … you want all your chats at openAI or in the privacy of your phone?) and ultimately be massively dystopian.

Unlike plagiarism checks you can’t compare to a ground truth.

Prompt guidance can box in the output space to a point you could not possibly tell it’s not human. The technology has moved from central servers to the edge, even id you could build something for one LLM, another one not in your control, like a local LLAMA which is open source (see how quickly Stable Diffusion 2 Vae watermarking was removed after release)

In a year your iphone will have a built in LLM. Everything will have LLMs, some highly purpose bound with only a few M parameters. Finetuning like LoRa is accessible to a large number of people with consumer GPUs today and will be commoditized in a year. Since it can shape the output, it again increases the possibility space of outputs and will scramble patterns.

Finally, the bar is not “better than a flip of a coin. If you are going to accuse people or ruin their academic career, you need triple nine accuracy or you’ll wrongfully accuse hundreds of essays a semester.

The most likely detection would be if someone finds a remarkable stable signature that magically works for all the models out there (100s by now), doesn’t break with updates (lol - see chatgpt presumably getting worse), survives quantisation and somehow can be kept secret from everyone including AI which can trivially spot patterns in massive data sets. Not Going To Happen.

Even if it was possible to detect, it would be model or technology specific and lagging technology - we are moving at 2000miles and hour and in a year it may mot be transformers. They’ll be GAN or RNN elements fused into it or something completely new.

The entire point of the technology is to approximate humanity - plus we are moving at it from the other direction - more and more conventional tools embed AI (from your camera not being able to take non AI touched pictures anymore to Photoshop infill to word autocomplete to new spellchecking and grammar models).

People latch onto the idea that you can detect it because it provides an escapism fantasy and copium so they don’t have to face the change that is happening. If you can detect it you can keep it out. You can’t. Not against anyone who has even the slightest idea of how to use this stuff.

It’s like gunpowder was invented and Samurai would throw themselves into the machine guns because it rendered decades of training and perfection, of knowledge about fortification, war and survival moot.

On video detection will remain viable for a long time due to the available entropy. Text. It’s always been snakeoil and everyone peddling it should be shot.

And? Blatant lies are not exclusive to AI texts. Every right wing media is full of blatant lies, yet are written by humans (for now).

The problem is, if you properly prompt the AI, you get exactly what you want. Prompt it a hundred times, and you get a hundred different texts, posted to a hundred different social media channels, generating hype. How in earth will you be able to detect this?

Just like your comment you say? Indistinguishable from human - garbage in, garbage out .

If you actually used the technology rather than being a stochastic parrot, you’d understand:)

I think you're trying to handwave at someone who knows more about the steganographic watermarking approach than you do.