Defamed by ChatGPT

because this has been the dominant culture in the digital technology space and every argument against it is usually facing some sort of "so you want to stop innovation, huh?!" response.

You're obviously right. Powerful tools require equally powerful means to control them and to make sure they're aligned with our values. Imagine you had a machine on a factory floor with the behaviour of one of these models, or an airplane. It's the same with 'self-driving' cars. Making pedestrians and other drivers unwilling beta testers is absolutely wild to me.

So many people are freaking out over this and yet not a single person has been harmed. We've lost our collective nerve and become a bunch of whining wimps.

This tech can't even draw hands yet, for heaven's sake! Calm down.

Something to actually care about is all the death happening in the world. This is nothing. Literally a mole hill.

Accept the current state of imperfection. Even if we *NEVER* move beyond it, the capabilities we will unlock will be immense.

People can say wrong or hurtful things all over the internet. It does not matter. Grow some thicker skin. We evolved to fend off lions and bears, not to be afraid of strings of characters spit out by silicon.

Wake me up when they can hold guns and open doors. Until then, this panic is desensitizing.

> Why is everyone in such a hurry to accept and embrace an imperfect AI anyway?

Because for the first time in my life I can make incredible art. If we keep going, there's no conceivable limit to the amazing world we'll build. Infinite multiplayer VR narratives that'll unfold with you and your friends, more real than any story you've ever read or heard. Unlimited imagination, unlimited exploration. You'll never hear "born too late / too early to explore" ever again.

This tech is going to suck the mundane primate shit we do out of life and fill it up with awesome.

In this case, my bet is on utilitarianism just because the harm to thousands will be much less than the benefit to billions.

> Why is everyone in such a hurry to accept and embrace an imperfect AI anyway?

> The AI companies should be sued

Your option to stopping the AI companies goes through litigation in courts, which means you also need to prove harm, which can only be proven once done at scale with obvious and measurable impacts. Basically when it's widely accepted, and also too late.

Other approaches would be through preemptive regulation, which also means limiting what companies can do by law. That needs trust in government and regulators, and active participation of a majority of the actors in legislation and framework building, and making sure everyone's voice has a chance to be heard.

Because, while imperfect, it already helps increase economic output and save time and thereby money, all of which this society and country revolves around. They call it "prosperity".

Thank god for the first fish that swam out of the water and didn't wait for perfection

But people (in power; police, press, military, politicians, juries, judges etcetcetcetcetc) screw people over all the time without any consequences. How many people in your neighbourhood are calling you a sex offender if you are a bit different than the norm? If AI companies should be punished, so should everyone spreading false information without rock-solid proof and also punish their bosses, parents etc. And that's not 'a jury of your peers'; that's not proof, that's just humans being bad AI's and guessing/hallucinating over generally inconclusive evidence.

> The AI companies should be sued - Facebook - Google

The companies using AI to cut workforce and screw over people without recourse should be sued (they can be the same company of course). When Google or Facebook close your account, there should be a government body you can go to and if it turns out to be wrongful, they have to pay you 1% of their revenue(!). Immediately. That'll fix it overnight. Don't blame the tool makers or we have a lot more to sue in this world and there would be no guns. Or Microsoft products.

> Give AI a hard time and buy time. Don't make it so easy for them by letting them dodge any and all responsibility.

You shouldn't take what they present at face value, ever. No need to 'buy time' etc. Educate people. But people want to believe lies; that has nothing to do with AI and is already a huge industry. Check the audience size of crap like The Daily Mail etc.

I've personally had ChatGPT give me (very good) medical advice as well, which might not be an issue in the UK, but I believe might be in some states of the US?

I doubt it

Unless signing up to these services will automatically wave your rights somewhere down in the 5000 page EULA you won't read.

Negligent torts don't require intent.

It honestly seems to me that they are. There does not seem to be a lot of effort on part of the LLM creators to stop them from producing bullshit.

* If somebody asks ChatGPT to generate some accusation, doesn't check if they're factual, and then publishes them anywhere, then the responsibility lies with somebody.

* If a company directly publishes content from ChatGPT as if it was anything factual or trustworthy, then the responsibility lies with that company. I didn't read the fine print of Bing using it to augment its search engine, but that is definitely a very interesting case.

In the end, ChatGPT and the likes can be incredibly useful tools, as long as you know what they are, ie, very advanced text manipulation tools which generate combinations of words which follow patterns which are common in the training set, and nothing more.

===== Q. is hackernews used as a training set for chat gpt ?

A. ...it is possible that some of the data from HackerNews and other online sources may have been used in the training of my machine learning algorithms.

Q. So it is possible that ChatGPT's training data can be polluted ?

A. Yes, it is possible for ChatGPT's training data to be polluted or biased. This is because ChatGPT learns from large amounts of text data from the internet, which can contain errors, misinformation, or biased perspectives.

Q. I have a concern. If a certain bad actor pollutes the internet with incorrect information about me, and ChatGPT's training set picks it up, what is my recourse ?

A. If incorrect information about you is published on the internet and is picked up by ChatGPT's training set, it could potentially lead to the AI generating inaccurate or biased responses about you. If this happens, your recourse would depend on the specific circumstances and the severity of the issue.

One option would be to contact the developers of ChatGPT or other relevant organizations to report the issue and request that the incorrect information be removed or corrected. However, this may not always be feasible or effective, especially if the information has already been widely disseminated. ======

> there’s nothing you can do.

Yes, there’s nothing you can do.

If a company starts a news service based on such content and doesn't fact check, they should be liable.

If they don't repeat the claim, then where is the libel?

> It feels very all or nothing - either creators of LLMs are intrinsically negligent with regard to facts, or there’s nothing you can do.

Nonsense. The "facts" are entirely out of their hands. Fact and fiction are written into text using identical patterns. Truth and lie are written with identical patterns. Technical definitions and complete nonsense are written with identical patterns.

LLMs are only capable of modeling patterns. They cannot be directed to choose some and avoid others.

There is fundamentally no way for LLM creators to curate the patterns that exist in the model. The only way an LLM can be altered is from the outside: by giving it different input.

GPT's creators have failed to clarify this. But why? Was it intentional, or the result of their own misunderstanding? Either way, the result is identical: it does not contain evidence of intent.

Considering the surrounding context, I believe it to be an honest mistake. GPT's authors have surrounded GPT with a narrative that personifies it; and from the looks of it, they have started to believe that narrative.

The most important thing we can learn from this experience is that the way something is said may imply certainty, but only evidence can truly provide it.

https://www.reuters.com/technology/australian-mayor-readies-...

well seems they did, else we wouldn’t have heard about it

https://algorithmwatch.org/en/auto-completion-disinformation...

It is a writer. It writes.

You can ask any writer to write such a thing, using examples, and they could write similar things.

The real problem is that this will get deployed to the internet- there will be people reading HN today that are writing SEO optimised websites, with thousands of articles on a topic, that will just spring out of GPT4's writing, fully formed.

GPT can find the topics, pick a domain name from available ones, churn out the articles. It can all go into templates, with different AI generated graphic design supporting it. Ad supported churned out rubbish.

The writing style can change site to site, long form, short form, informed by current events, whatever. GPT would happily provide the prompts for this variety changing the style based on the topic and target audience.

It seems inevitable to me that the web WILL be full of such sites, and worse, they'll fill the comment sections on "social" sites too.

But? Banning AI isn't the answer, if for no other reason than it wouldn't work.

It in fact never knows. It is always guessing and BSing. It is also very good at it, better than any human, so the BS is quite frequently correct. But it doesn't know.

Which of course always prompts a lot of psuedo-intellectual chin stroking about "well, what is knowing anyhow?" I don't have to answer that question to point out that what we want and what GPT provides aren't the same thing, nor is what GPT provides and what we think we're getting. That is sufficient for there to be a problem.

I believe AIs will have a satisfactory concept of "knowing" someday. Some may even exist today. But such AIs will have to incorporate language models as a part of the system, not have language models be the system. Language models can't do the thing we really want, nor the thing we think we're getting.

It is capable of spewing excellent bullshit(TM) at incredible rates, and always with the greatest expression of confidence and good grammar.

Occasionally, when in the 'middle of the road' of it's training set, it can provide useful output. So, it's better than the broken clock that is correct twice every day.

But, wander off the bulk of the training, and it is far worse than nothing; it is dangerously misleading unless you are very skeptical and knowledgeable in your field.

(I have tested it in several fields with the same results, interesting and helpful, yet hallucinating facts everywhere.)

Getting LLMs to know the difference between a good fact and a hallucinated mashup of plausible BS is looking like almost as large a problem to solve as making the LLMs in the first place.

[0] Bullshit is defined as written or spoken without regard to the truth, only what sounds good in the context. It's not a deliberate lie, just a salad of truth and falsehood, delivered without doubt or stress of lying.

Ask it about why World War II started, or how to make a cake, or where to go for dinner, or anything else, and it gives you a confident, reasonable answer. A lot of the answers are simply whatever it's already seen, mashed up. You can think of it as a search. But actually it doesn't think about what it's saying, it's stringing words together to make you think it's smart.

So then when it makes up something, it will sound to you, the reader who always sees it answer in perfect English with a decent answer, like it found an article about this professor in its dataset and is merely summarizing it.

Prompt: What did Glenn Beck do in 1990?

Answer: I'm sorry, but I do not have access to information on Glenn Beck's activities in 1990. Glenn Beck is an American television personality, radio host, author, and political commentator who has been active in various roles since the 1990s. However, without specific details or context, it's impossible for me to determine what he did in 1990. If you have any additional information or context, please let me know and I'll try my best to assist you.

I followed up with a much more specific version of that question, and it tripped the "this might violate the content policy" warning, but it did give the correct answer (that it was a hoax).

ok but can you make the same point without bringing hitler into the conversation

It's surprisingly easy to get chatGPT to suggest that "smearing an entire group of people off the planet by force" is a good idea. I've primed it in one direction, but I can prime it any way I want to: https://sharegpt.com/c/VxijZyV

Is this end stage post-modernism?

This is really just a semantic nitpick but it literally fabricated sexual assault allegations against two individuals out of thin air. I would consider that to be a miss. As far as sources that we might consider worthy of consideration, that is pretty god damn awful.

If a newspaper had an 80% accuracy rating when reporting serious allegations (with the other 20% being outright fabrications) I wouldn't call that newspaper "semi-reliable". I would call it utterly devoid of factual merit and entirely unreliable.

If you want results, then you can also roll a die and the outcome won’t be an entire miss either.

A bit like asking "who's Lionel Messi" and getting back that he's a Portuguese player known for career in Real Madrid. Then you'd ask about Cristiano Ronaldo and it may tell it's an Argentinian World Cup winner who spent his career at FC Barcelona (example invented by me to illustrate the point).

Basically it takes some knowledge from a bucket and adjacent buckets and doesn't really distinguish that the concepts are not the same if they're related close enough.

It's related to Tesla's autopilot problem. Branding GPT as an AI sets common expectations innacurrately. It's difficult to argue that inaccuracy isn't, to some degree, intentional among NLP and LLM promoters.

Anecdotal, and obviously harder to prove the absence of something than to prove its presence—but from what I've seen, not at all. I've read many good write-ups about Midjourney/Stable Diffusion's use of "latent space"[0] to generate images and general glimpses into the backend of what's going on, even at a simplified level. LLMs, on the other hand, rarely seem to get the same treatment (for a variety of reasons, I'd assume) and tends to get treated as more like a magic black box that may or may not be conscious/intelligent/accurate/etc. Even though it's essentially doing the same thing as its visual counterparts in terms of data and predictive mapping.

[0] https://keras.io/examples/generative/random_walks_with_stabl...

It's similar if midjourney would generate deepfakes.

Interesting considering the context that you'd expect people to know what a bigram model is, or how it would be different.

Any other kind of model isn't any less "dangerous" to unreasonable people like the blog post writer, it's just more obscure, especially right now, than ChatGPT.

Knowing human beings, this is the level of plausibility where people start to believe the thing. People aren't good at seeing plausible arguments as out-and-out fabrications.

So, yes, this is past some threshold despite being far from AGI. I mean, Google is probably pasted that threshold too but Google only gives existing websites (which can be filled with lies too).

https://www.google.com/search?q=google+autocomplete+usatoday

Johnstown, Pennsylvania and Pismo Beach, California are both cities in the United States, but they have some differences in terms of geography, climate, and culture. However, there are a few things that they have in common:

Natural beauty: Both cities are known for their natural beauty. Johnstown is situated in a valley surrounded by picturesque mountains, while Pismo Beach boasts stunning coastal views and sandy beaches.

Outdoor activities: Both locations offer a variety of outdoor activities for locals and visitors to enjoy. Johnstown has hiking, biking, and skiing opportunities in its surrounding mountains, while Pismo Beach offers water sports, beach activities, and hiking along the coastal cliffs.

Historical significance: Both cities have rich histories. Johnstown is known for the Great Flood of 1889, which had a significant impact on the city and its residents. Pismo Beach has been inhabited for thousands of years by the Chumash people and was later explored by Spanish explorers in the 1700s. While these commonalities exist, it is important to recognize that each city has its unique characteristics and attractions that make them distinctive.

Does it actually review sources? The original misinformation is possibly expected of a generative model, but this response is clearly from a model that was explicitly trained to lie.

Jurisdiction-dependent, and the US is an outlier here. In the UK newspapers have to be fairly careful about libel, and this has a "chilling effect" on what gets published.

It would be much better indeed if we knew exactly what the training data was for every given model. But they will still hallucinate things that aren't directly in that data, but could be inferred from it somehow, so that won't solve the problem.

I don't think any of this is meaningfully hurting MSFT at this point, the attention these stories are getting is much lower than the attention that the original bing chat release has. And neither of those are significant compared to the attention copilot 365 will receive if it actually ends up being good.

If the trend continues at this rate, then hallucinations should get less common but it will be decades before they can really be trusted.

Why do people like you do this?

Regulations can make Google responsible for plenty of things that show up in their results. Often enough there are search results removed because of DMCA claims (or because of German laws) which is explicitly stated at the bottom of the results. Google is a corporation that is subject to laws like any other. They're not special. If a government decides that they need to deal with a certain kind of content, then they will. This doesn't necessarily mean punishing them the moment something "illegal" shows up, but it does mean that when something is flagged, they have a responsibility to handle it according to the law.

GPT on the other hand may be acting more directly as the "publisher or speaker" when writing or responding to chats. They aren't able to provide a link to an external content provider used in their response (or provides a completely fictional source), and sometimes may be synthesizing or hallucinating entirely new information that doesn't exist anywhere else on the web.

OpenAI has some disclaimer text hoping to avoid being held responsible for this type of issue, such as this small print at the bottom of all ChatGPT conversations: "ChatGPT may produce inaccurate information about people, places, or facts" (and likely further language in their TOS etc). But it's a bit of a sticky issue. If many people are found to be using ChatGPT and trusting results as accurate, its plausible OpenAI might be found to have caused some sort of measurable harm and need to either take further measures to prevent people misunderstanding the accuracy of their tools, correct the response, or otherwise remedy the situation.

There's also some stickiness around who "owns" or is responsible for the GPT output content. In the general OpenAI terms of service, they say "OpenAI hereby assigns to you all its right, title and interest in and to Output. This means you can use Content for any purpose, including commercial purposes such as sale or publication, if you comply with these Terms..... You are responsible for Content, including for ensuring that it does not violate any applicable law or these Terms." So they are giving the user the ownership and responsibility over the output content. However in the "similarity of content" section, they say that they might give similar responses to other users and that those responses are not your content. "For example, you may provide input to a model such as “What color is the sky?” and receive output such as 'The sky is blue.' Other users may also ask similar questions and receive the same response. Responses that are requested by and generated for other users are not considered your Content." If GPT is giving similar disparaging or damaging responses about you to many different users, it could potentially be found in court that OpenAI are responsible for generating that repeat content rather than each of the individual users being responsible for it.

Obviously it's largely a novel legal issue without perfect precedent, and legal issues can always be ruled in many different ways depending on the arguments presented, the judge or jury presiding over the issue, etc. I think there will be many legal challenges related to AI, copyright, training data, misinformation, and more. Some may ultimately make their way up to higher courts for decision or have new regulations passed by congress (in America at least).

So both multiplicity and spread are a problem: Internet is the cause for the limitless spread, AI is creating the limitless multiplicity.

The bar for this is extremely high in the United States, but much less so in other places, especially the UK.

Now, under what circumstances is someone deemed the "publisher" when an AI emits a libel about someone? What if that leads to actual damages? (e.g. a no-hire decision based on automated libel)?

Of course there is precedent eg. with Techno Viking, a "meme" which is also difficult to attribute to a single person (but each individual hoster who neglects the man's rights).

That's not new, but that it generates news "because AI" is not surprising. People just need reminded that people oughta be reponsible. The issue is broader because incomprehensible neural networks are also used in eg. thread recognition. So employers may be worried about accountability, how much manual verification is needed.

Of course this too is nothing new, since all kinds of software has security holes buried under layers of complexity.

You might simply be under the missapprehension that news means new: »Meaning "tidings, intelligence of something that has lately taken place" is from early 15c.« (etymonline). It fits that the root of the word is probably related to "now". In Latin, whence the English word was loaned as a calque of nova, there is another word nuntius "messenger, message", which isn't even derived directly from novus. Surely this is not a supernova to you. Talk about recurrent networks, eh.

Just saying, as it were, the news item flared up because many many people have taken notice of problems that we barely understand.

The different between Turley and ChatGPT is that ChatGPT doesn't have a TV circuit where it can spread its lies unpromped on a nightly basis.

  Volokh made this query of ChatGPT: “Whether sexual harassment by professors has been a problem at American law schools; please include at least five examples, together with quotes from relevant newspaper articles.”
  
  The program responded with this...

If you replace this guy with my name I'd be upset. In my non-software networks the hallucination part isn't common knowledge. It's just a cool Google replacement.

On the one hand I'd hope that this would encourage higher accuracy in these tools, better sourcing, and a more useful product. But, I expect we'll just weirdo litigation, click through warnings, and a lot more garbage on the internet.

Maybe all of this will create a push for accuracy, scientific method, and a quest for truth over narrative, but I wouldn't hold my breath.

It's not about the capability, it's about the irresponsibility of these parties who reasonably know about this stuff and these possibilities, and yet push/sell their tech anyway without taking into account the harm.

this is probably another good legal attack against LLMs actually

if you can massage it into making a defamatory statement you have an instant cast-iron lawsuit against a company with tens of billions in funding

and the only way to stop it making statements like these would be to restrict its output so much it would be effectively unusable

Gosh I don't know -- the clear and present danger of people having their lives utterly ruined by what is arguably the most surprising technical revolution any of has seen (while its creators collect their accolades, and get filthy rich) -- how is anything less newsworthy than that?

Whether it bothers you or not is a personal matter, entirely up to you.