Why I'm leaving Medium: AI policy
medium.com
medium.com
This seems just as silly of a take as the fervent supporters of sticking AI everywhere.
AI helps in some areas (e.g. assisting medical diagnoses) and is shitty in others (e.g. spam content).
I wonder if the author would refuse to receive a diagnosis from a doctor if the diagnosis was AI-assisted?
Obviously if we could somehow separate the good from the bad, we should, but it seems like an intractable problem.
This dichotomy (polychotomy?) is also aggravated by the fact that when people talk about AI, they are probably 99% referring to generation of prose, art, and code (as opposed to e.g. medical diagnosis).
That was what I took TFA to be speaking about. It's ambiguous, probably because author didn't expect an audience of HN readers, but most of the enumerated concerns and uses cases are specific to "AI as replacement for human creativity, in artistic endeavors."
A lot of people (particularly Americans) seem to think of nuclear bombs in exactly this way. They typically see the 'some task' to be of extremely outsized necessity and importance, and I think they'd say that makes the two ideas cohere.
If you think using nuclear weapons can be justified, then you don't think they're evil, you just think they're dangerous.
I'm less sure that it's an outright logical contradiction, but I'm inclined to agree with you there as well.
Not a good negotiating position where deterrence is concerned.
Setting aside the apocalyptic prediction in particular, this sort of thing is already the case with many things that have become irritating or pernicious at scale, like automated phone systems and SEO. The latter of those examples already includes AI more and more. If AI weren't a helpful device for reducing writing costs, it wouldn't be playing a role in filling the web with worthless SEO-spam garbage.
I'm sure the strong proponents of AI in all the places have never griped about false positive account permabans from cloud services or app stores.
What could possible go wrong with turning a function over to an automated system that's <100% correct, realizing cost savings by downsizing the human support backing it, and then eventually just engineering the process so all appeals are piped to /dev/null?
PS: What we need are mandatory rights to appeal to a human, when automated decision systems are used, with legally defined response time SLAs and arbitration options.
I agree that bans and appeal handling needs a human in the loop to work as intended, to have some way of handling the false positives. Without a human in the loop the appeal process can just as well be removed entirely, because it won't fulfill its purpose.
On the other hand, if you don't see recent and upcoming generative AIs as that world-changing, they don't really change the picture as to whether or not the enshrinement of such a right into policy is warranted.
The question isn't whether or not there are benefits to be had from the technology. The question is whether or not the cost/benefit ratio is favorable. It's not clear to me what the answer to that actually is.
They are uninterested in the content they just want the clicks.
If a doctor just slops a diagnosis that results in malpractice via “hallucination” then people will stop going to them.
Will you continue to use that dangerous doctor just for the sake of your misguided principle?
What a weird interpretation of what I said.
Obviously I mean a competent doctor that uses AI-assistance in a responsible manner, not a dangerous doctor that commits malpractice.
You are assuming a “competent” blogger even in the face of a slop banner.
How will you know the doctor is competent? All you know is they give sloppy diagnoses.
How will the op know the blog writer is competent?
Their metric is “if they use slop at the top of their blog it’s a waste of their time to investigate further”.
Uhhh, I don't know about that one. Have we seen studies that show real predictive capabilities? Anecdotal evidence is not helpful, and it seems rather risky to depend on something that has not been thoroughly vetted when it comes to people's lives.
https://www.nature.com/articles/d42473-022-00035-y?error=coo...
>AI improves the lives of patients, physicians, and hospital managers by doing activities usually performed by people but in a fraction of the time and the expense. [...] Not only that, AI assists physicians in detecting diseases by utilizing complicated algorithms, hundreds of biomarkers, imaging findings from millions of patients, aggregated published clinical studies, and thousands of physicians’ notes to improve the accuracy of diagnosis.
Really, the only way this could possibly be useful were in a hypothetical case where only visual methods were available and there weren't enough radiologists, and even then it wouldn't be _very_ useful, given the high failure rate.
> Not only that, AI assists physicians in detecting diseases by utilizing complicated algorithms, hundreds of biomarkers, imaging findings from millions of patients, aggregated published clinical studies, and thousands of physicians’ notes to improve the accuracy of diagnosis.
This sounds very fuzzy. Can you appoint to specific approved applications, or is this all hypothetical?
One of those links is a systemic literature review of a bunch of other existing literature.
If that is not enough for you, that is totally understandable! But if I'm honest here, I don't really care enough to go digging even further and trying to find additional studies or approved applications from your preferred country, with an ultimate goal of convincing you.
I would be fine if my current doctor used AI to assist in their diagnosis of me. You don't have to be, it's completely your right to choose that.
I can't speak for the author, obviously, but personally my answer would be "it depends". If the diagnosis came from a doctor who happened to use AI as one of their tools, I'm OK with that (as long as it was a locally-hosted AI, but that's a different issue). If the diagnosis came from AI without a substantial amount of analysis from a doctor, then I'd absolutely reject that.
What the author has written, in my opinion, is unreasonable because it is absolute in its hatred for AI.
Increasingly (and I think certainly in the above case) AI is used as shorthand for genAI (which is unsurprising, as up until recently most AI-ish things got called ML anyway). I certainly hope no-one's using LLMs for medical diagnoses...
> I wonder if the author would refuse to receive a diagnosis from a doctor if the diagnosis was AI-assisted?
It really depends on what you mean by 'AI-assisted', IMO. If you mean that the doctor had asked a chatbot, I'd very much be looking for a second opinion. What sort of AI assistance did you have in mind.
The designed-for-medicine kind, not the ChatGPT kind.
(There have also been some reports of doctors making diagnosis decisions based on ChatGPT results.)
#1 AI as final step
#2 AI as upstream step, with subsequent human review/adjustment
The cases people seem to have problems with are where no human eyes/hands provide oversight to probabilistically-correct model output*.
It's an important enough distinction that I'd be in favor of mandating companies declare which they're using (for a specific use case).
* As distinct from previous deterministically-correct expert systems / rules-engines
People are already using AI to write emails, essays, publications, and more. SWEs are using it to literally do their job.
This stuff brings us even closer to the last act of the movie "Up." It's insane to me that people _don't_ see this.
Does GenAI monetarily disincentivize the self-expression of humanity through art and other creative endeavors, such endeavors that were already difficult to make a living on pre-GenAI? I think it does.
If I had a choice, I would not allow the doctor to give me an AI assisted diagnosis in the first place.
I've realised I don't actually care about people using it for programming or brainstorming or whatever. It's just I feel so inslulted when I read something that is in the default AI voice.
So I don't know that I agree entirely with the writer of the piece but I get where he's coming from. AI writing is unpleasant to read. And I hope Medium reverses their decision.
Afaict, it's 'high quality' only in the sense of not containing many grammatical mistakes or spelling errors, which of course is not 'high quality' so much as 'basic literacy'.
I think I know what you mean about other elements of the tone (staid and obsequious) not necessarily being related to style but they don't quite form a clear identity for me. The defects of the writing are still more marked than its personality.
To be direct:
Your post assumes that all users are equally good at producing content with AI/LLMs — but if they’re not, then people who are better providing that to people who are worse will become a market.
Just like with chairs and writing, now.
Tastes vary.
So really, I can get behind the author.
You should not just assume something is trustworthy or will solve your problem just because it was written by a human. And I have read plenty of cruft-filled blogs/articles/etc. from humans.
It kind of depends on the subject. If the blog is about politics, say, then you may assume that the human author is likely to be willing to lie. If it's about a non-contentious subject, though, then it is highly unlikely the human author will lie, and if they don't know, then, well, okay, _some_ people will write articles about stuff that they don't understand, but it's not common. Most non-psychopaths, if they don't know the answer, won't just make something up.
Whereas the magic robot will very happily spew nonsense on any subject.
The same is not true for AI generated prose, in which the author often can't even be bothered to remove AI boilerplate. In such a case it's obvious the code has never been tested, and it might not even compile.
This blind cheerleading of LLM-generated content filling the web is what pushes people to hate them.
"Sure, here are 11 examples of recursive islands..." Yeah, that's how some articles begin
> And also, I vehemently disagree with this statement. Flaws express personality.
So for the same reasons, does he wish to not read content that has been assisted by spelling or grammar checking? Or an editor, proof reader or fact checker? Or a thesaurus or dictionary? Or is he only concerned when AI is applied to those roles?
I'm curious where someone like him draws the line. AI is an ambiguous term. He's an author and photographer, so perhaps AI has just come to mean LLMs and image generators?
I might even agree with the thrust of his concerns, but this kind of diatribe always come off a bit rage-blind, and maybe brings the strength of the discussion down a little.
Contrary to your accusation of being rage-blind, I have studied this topic in detail and have carefully thought about it and read many articles in philosophy about it. I am still trying to articulate my ideas but my general feelings go way beyond emotions.
Some people say that LLMs make you more productive, but guess what? The value captured by your productivity will only enrich the companies that employ you. Companies will hire fewer people and eliminate positions all for "shareholder value" and to give a small raise to a C-suite executive.
But at the same time, we could definitely envision a world where these models accelerate human creativity - smaller high performance LLMs working on the edge, responsible usage for learning and skill development, helping people build their own voice.
The key is to identify the subtle line between dependence and empowerment, and to know when one is accruing AI-debt at the expense of their own abilities.
I feel like the scarcity and difficulty of creating something is part of the charm of art, instead of it being a matter of mass production. And for what? To show more ads?
Or CloudFlare Pages, orGitHub Pages, or Firebase Hosting, or Fastly, or Vercel, or S3+CloudFront, or your own nginx instance.
The question isn't what other simple platform is out there: the question is what blogging software do you want to be seen using. People don't want blogspot.com at the end of their domain names, and they don't want to look like they're using blogspot.com. They didn't mind Medium, and that's the difference. It's not a question of technology, which has existed for decades; it's a question of marketing and brands.
The way I see it, the only true solution(for both writer or reader) is to go back to the good old-fashioned self-hosted blogs, expose an RSS and rely on RSS readers(yes, those still exist). You will never be able to compete with the exposure that you'd get on Medium or any other platform but at the end of the day, I'd rather pay 50 bucks for a good burger than 5 bucks for a cheap bun from Lidl with some very questionable substances inside.
These platforms are all the same: put your middling content here and you’ll be a star
In my experience most people have valueless content and target an affinity group audience. They produce enough content to have a viral hit, that viral hit will then become your first ratchet. You then adjust your content type based on the virality and then you just continue the ratchet. Iirc this is the playbook of eg MeBeast etc… and theres a deterministic outcome: create enough attention that marketing income consumes your lifestyle and drives all future ratchet iterations
If you have anything of value people will figure out how to get it but it won’t be fast and it will be mostly ignored
My eyes about rolled out of my head reading through this.
My money is on either this writer fading into obscurity or using AI tools in the future when they realize their stance was silly.
I will never use AI tools, never. And I am more than happy to fade into obscurity. I just want to make enough money (without AI) so I can take care of my family in peace, and also articulate my ideas against AI to whomever will listen. I am absolutely not concerned about becoming famous.
In all seriousness though, there isn’t a competitive platform out there that won’t adopt AI or any other technology if it meant making money. If they chose not to, they would be giving ground to someone who would. That’s capitalism in a nutshell.
You aren’t their customer. It’s the eyeballs they’re after and, any particular authors word isn’t really precious. Enough authors/writers out there who are willing to replace you.
Remember kids: “Content” is the word for things _between_ the ads.
Cheers.
Yes, and we left people behind repeatedly. You'll notice people who may have intellectual or learning disabilities can no longer find a job with a decent wage in the US. Previously they could, because manual labor jobs were plenty. Now those are automated.
When we further automate and automate, we raise the barrier to entry, because robots or AI can perform the bottom most tasks. So, people who are not capable are left behind, and eventually there will be nobody to work jobs.
If the average job requires reading and writing, most people can do it. If the average job requires a high school diploma, less can do it. If the average job requires a degree, even less can do it, and so on and so on.
Isn't the end result - all work being done by technology and all people having all of their time to spend on fun activities - a pretty nice situation?
What could possibly make you think this is the end result?
In the past 50 years, productivity across the whole economy has went up over an order of magnitude. People in the 70s were working 40 hours a week - are you now working 4 hours?
We don't have precedent for this, is my point. The solution to lost work is make people work something else. My point is what happens when there is no something else for those people to work?
Will we have leisure? Who will pay for uncapable people to have leisure? This sounds like UBI.
> all work being done by technology and all people having all of their time to spend on fun activities
Maybe in the end, but what happens before that? Like, say, 25% of people are "smart" enough to still work a job, what happens to the 75%? Or even 90% of people work but 10% can't, do we just like... kill those 10% or?
We haven't really figured this out.
He is mad at others using AI on his favorite blogging website. I think it is an insult to himself, because most AI produced articles etc. are unreadable and full of useless filler information - I think the guy can do better.
Ice delivery guy = Traditional writer Going on strike = Complaining or protesting Demanding ban of fridges = Calling for restrictions on generative AI The analogy suggests that the writer's complaint about generative AI is as futile and outdated as an ice delivery person trying to ban refrigerators. Just as refrigerators made ice delivery largely obsolete, the implication is that generative AI might be seen as a technological advancement that could potentially impact traditional writing roles.
[1] https://www.reuters.com/technology/artificial-intelligence/m...
If we're thinking of the same reporting, it was based on a claim by TollBit (a content licensing startup) which was in turn based the fact that "Perplexity had a feature where a user could prompt a specific URL within the answer engine to summarize it". Actions performed by tools acting as a user agent (like archive.today, or webpage-to-PDF site, or a translation site) aren't crawlers and aren't what robots.txt is designed for, but either way the feature is disabled now.
[0]: https://commoncrawl.org/faq
[1]: https://platform.openai.com/docs/bots
[2]: https://support.anthropic.com/en/articles/8896518-does-anthr...
[3]: https://blog.google/technology/ai/an-update-on-web-publisher...
The companies that are ignoring robots.txt, are also probably the companies not advertising that they are ignoring robots.txt.
In the US it is a suggestion (so long as Fair Use holds up) but all I've seen suggests that the major players are respecting it, and minor players tend to just use CommonCrawl which also does. Definitely possible that some slip through the cracks, but I don't think it's as useless as is being suggested.
""OpenAI CTO doesn't know what data was used to train the company's video generating platform, Sora""
https://www.youtube.com/watch?v=4AYbZG3h14w
Funny. If I can browse to it, it is public right? That is how some people's logic goes. And how OpenAI argued 2 years ago when GPT3.5/ChatGPT first started getting traction.
There's legal backing to it in the EU, as mentioned. With CommonCrawl you can just download it yourself to check. In other cases it wouldn't necessarily be as immediately obvious, but through monitoring IPs/behavior in access logs (or even prompting the LLM to see what information it has) it would be possible to catch them out if they were lying - like Perplexity were "caught out" in the mentioned case.
> Funny. If I can browse to it, it is public right? That is how some people's logic goes. And how OpenAI argued 2 years ago when GPT3.5/ChatGPT first started getting traction.
If you mean public as in the opposite of private, I think that's pretty much true by definition. Information's no longer private when you're putting it on the public Internet.
If you mean public as in public domain, I don't think that has been argued to be the case. The argument is that it's fair use (that is, the content is still under copyright, but fitting statistical models is substantially transformative/etc.)
https://www.wired.com/story/perplexity-is-a-bullshit-machine...
It takes this sort of critical scrutiny, otherwise mechanisms like robots.txt do get ignored, whether willfully or mistakenly.
I believe this article is around the same misunderstanding - it doesn't appear to show any evidence of their crawler, or web scraping used for training, accessing pages prohibited by robots.txt.