The models currently pose no more of a risk than someone going to the library and learning how a bomb or weapon is made. AI safety is commonly used by openai to encourage the creation of regulation which enables them to capture market %.
This is like arguing that software or engineering data should be restricted due to safety. Is their a potential of harm, yes. But that potential is Inherited in information and is the user who determines how it is used.
Historically, corporate interest has not shown the ability to set aside the drive for financial profit. Everyone thinks they have the right ideology, no one person has the answers.
I feel like this would be restricting access to trigonometry and calculus because it is a key part in terminal guidance and in weapons development.
The “can information be harmful?” is indeed an interesting debate.
I think it can. Take social media, for example. It is associated with increased depression rates, so the information people are exposed to on these networks is harmful.
Another case I can think of is certain knowledge. Certain knowledge exists in the world that we are better off not knowing. Like what some war trauma or childhood abuse survivors know. Or learned helplessness. You would probably not want to surround yourself with that world view. It can be so shocking that it will stick with you for life and harm your mental health. Especially if that information is made available to a child.
Trigonometry and calculus are constructive information. There is probably also destructive information.
> Certain knowledge exists in the world that we are better off not knowing. … It can be so shocking that it will stick with you for life and harm your mental health.
This is a tangent, but I actually think this is a harmful belief that gives too much credit to the power of negative thoughts. There is no type of thought or knowledge that will damage you forever. Trauma is recoverable; and it is possible to acclimate even to the very most distressing ideas, so that we can let them go.
There is type of thoughts or knowledge that can damage you forever. For example, if a boy or girl are put on a path to question their gender and get convinced that their gender is different from their biology, this increases a chance of depression or suicide by an order of magnitude. And if they turn to medication or other treatments to alter the biology, this increases chances of depression or suicide even more.
You’re not really talking about the effects of a thought: you’re talking about changes in behavior (questioning, taking medication, etc.).
Suppose I asked you, “are you really sure your spouse isn’t cheating on you?” You may find this idea unsettling because the honest answer is no: nobody can be sure of this.
Some people would shrug it off and say “I can’t be sure, but I don’t think so.” And they would move on and forget about it. Other people will drive themselves crazy trying to get to the answer, even to the point of damaging the relationship with the spouse.
The difference between the two is not the idea (spouse could be cheating) but the reaction to the idea. This is actually what psychotherapy is for, to teach you how to react to thoughts and feelings in ways that aren’t destructive.
Certainly. The issue is, if you persistently push the message “are you sure your spouse isn’t cheating on you?” through all forms of media (including LLMs), down to kindergartens and preschools, you’ll inevitably will get a proportion of vulnerable population driving themselves crazy trying to get to the answer at some point of their lives, damaging the relationships and fabric of society in the process.
You're factually incorrect: people who are already on a path of questioning their gender and then resolve it by deciding that they are transgender experience lower levels of depression and suicide, and those who receive gender-affirming surgery experience lower levels further:
Your error is a form of Simpson's Paradox: the population that decides to go through with gender-affirming surgery is at significantly higher rates of depression & suicidal ideation to begin with, because it's not easy to feel that your gender is different from how the rest of society perceives you. Within that population, though, people who make a decision one way or another do significantly better than those who just live with the stress and ambiguity of their secret.
There is scientific consensus that calls for tolerance and acceptance.
And there is research that so far shows null hypothesis and no biological evidence for predisposition to being a furry, tree lover, lesbian or gay. While simultaneously genes for left-handed people do readily show up.
There is statistic that shows introduction of materials that attempts to set identity at early age, including religious or gender identity is very effective.
Statistics about suicidal ideation snd depression actually doesn’t show any correlation to acceptance of transgender in society. And it is roughly 3x of not transgender and 10x after the surgery. And then there is rise in actual transgender youth, toughly 30x, from 0.3-0.5% to 10% over last 40 years. This can be roughly attributed to introduction and popularization of this topic in early education. Effects are similar to early religious education.
I’m very happy for increased acceptance, diversity and continued increase of younger dating pool for special interest groups. And that these special interests groups get what they want.
But statistics does show that in terms of depression, decreased happiness and suicides it is not the greatest idea to dislodge biological gender. You simply can get 10% of population worse off out of this. If you add to that parents that are unhappy with this turn of events, this gets to 20-30%. And then suddenly you have democracy at stake, because nationalists figured out that this is a good angle…
There is a mental health cost to trauma and to dealing with very distressing ideas. Even if you heal from trauma (and not everyone does), the consequences of having gone through that experience will be a part of your story. People are often better off not having that be the case.
I have experience with trauma that made me depressed for about a decade. I've gotten over it to some extent, not entirely. But at the minimum, I lost a decade of my life. I would have been better off not losing a decade of education, career progression, friendships, pursuing adult prerogatives, and similar.
If a teenage person becomes depressed due to social media, they will pay the cost of not forming friendships and not being able to engage with education in that period of their life forever. These costs will be both mental and economical.
Certain knowledge like self-doubt or conviction that one has no self-worth (and some come not only from abuse but from certain reasonably commonly held beliefs in society) can cause a lot of chronic mental disorders — anxiety, depression, insomnia, various dysfunctions, etc.
Finally, consuming an always-online diet of doomscrolling can make someone's worldview so skewed that they will pay significant social costs.
I do believe there is a type of knowledge that will cause harm to someone for a time, or forever. It could even be a certain thought that becomes difficult to challenge. This is why there are therapists, who nowadays commonly treat people exclusively with challenging ideas, like with the CBT modality.
I think, to an extent, society will help someone recover after trauma or after going off the deep end, but that extent is not very large. Society is also quick to abandon people it deems beyond worth saving.
Is it having the information that is harmful, or the actual facts of life that you are being informed about that is harmful?
The Internet has basically shed light on a lot of common practices and uncomfortable truths that could be swept under the rug during the 20th century mass-production, mass-consumption, mass-culture, and mass-communication era. I'd argue that the real evil isn't knowing that your politicians are screwing you - it's them screwing you. Similarly, the real evil isn't knowing that we're all about to get fucked by climate change, it was the 100 years of industrial policy that brought us to this point. It isn't knowing about all the wars and the terrible stuff going on around the world, it's the wars and terrible stuff in the first place.
The Internet is going to lead to the destruction of society as we know it, but I'd argue that that society - the post-Industrial, post-Colonial, post-urban wasteland we've been living in since the 19th century - should never have existed in the first place.
This is part of the grift. Here we have Sam 'too dangerous to be released' Altman warning again and again while at the same time doing exactly what he is warning about.
I doubt he actually expects some type of regulation that will make the snake oil he is selling last a little bit longer. More realistically its so drive up the hype and cash out as much as possible until hardware is cheap enough for a good enough model since right now the gap between the pay to use models and the free models will only get smaller and smaller.
Yes part of this does smell of hype. While some of the AI research has been groundbreaking, its not a revolutionary thing. We have had chatbots and computer vision for a long time. This reminds me of all the hype and scaremongering around deepfakes which while have been misused, it turns out the largest users of the are the VFX and entertainment industry.
I wonder if people were this afraid of the printing press or radio
every library has a human being going through, by hand, and deciding what comes in, what goes out, and what is on "reserve", "special collections", a children's section, a fiction section, a non-fiction section, and on and on and on. libraries do not typically put an emphasis on "how to make a bomb" books. yes you could find one in certain libraries, but its not like there is an entire section dedicated to it, nor will it be featured in the public display nor will it be in the children's section. the purpose of it being there is not to help people make bombs, it is for research. if you came in there every single day with meth spots asking for the "bomb making book", that book would be taken out of circulation. much like the purpose of playboy in some archival libraries is not so patrons can go jerk off in the bathroom, but so that people can go read old interviews. libraries are nothing like the internet nor are they like AI. if a bunch of crackheads wander in asking how to make crack, the librarian does not point out to them the crack making manuals.
> if you came in there every single day with meth spots asking for the "bomb making book", that book would be taken out of circulation
Do libraries commonly remove books from their collection because unsavoury people are reading them too much?
There must be millions of libraries in the world, so I’m sure that kind of thing must have happened at some point - but I’ve never heard of it happening before, which makes me doubt it is common.
The libraries most likely to contain “bomb making tutorials” are academic research libraries, national libraries, etc - and those are the kind of libraries least likely to remove items due to these kinds of concerns.
At best “safety” will only ever be a secondary concern. The primary driving force behind most things called “AI” today is that it will allow companies to automate workers’ jobs, stop paying them, and stop paying for their health insurance. In a country with no right to healthcare, is the safety of those workers part of the question?
Or maybe “safety” just means the chatbot should refuse to tell an offensive joke. Workers should feel thankful that the robot that replaces them knows how to speak PC corporate lingo perfectly.
> The models currently pose no more of a risk than someone going to the library and learning how a bomb or weapon is made.
That is absolutely not true. Books in libraries do not dynamically and non-deterministically withold, distort, falsify, provide, etc. information based upon a generative algorithm that no one understands.
gestures broadly. Half of the world has an entirely different set of facts than the other half. As a society (the medical / scientific community is at a near consensus) we still can't agree on whether a vaccine works, or whether masks can slow the spread of a virus. The sheer amount of bad actors spreading wildly bogus information is astonishing.
My honest opinion is that the people (30-50% of the population?) voting for authoritarians like Trump, Orban, Bolsonaro, Meloni, Bukele, Ergodan, etc are not at all in Solidarity with what I'll call "the rest of us". From the ones I've spoken to there's legitimately a theme of wanting the chaos that these people bring to the world. Sticking a wrench in society for fun and profit.
And what I'm talking about isn't everyone having a single set of beliefs, that would be absurd. It's bad actors spreading what they know to be misinformation, and people falling hook line and sinker for the false rhetoric because it's repeated so often. People like Tucker Carlson who privately says Trump is a massive evil, publicly praises him. Or Russian troll farms that just constantly post the most controversial takes on social media until they become normalized.
What used to be a fringe group of conspiracy theorists that read rags like "The National Inquirer" and believed it has exploded because it's easy to make a website/podcast/twitter handle that looks legit and spreads like wildfire.
solidarity: unity or agreement of feeling or action, especially among individuals with a common interest; mutual support within a group
If the feeling or action is something you don't like, that doesn't detract it from its meaning.
But I am now curious, would you say that, because of this, the internet is a net negative? (Even after considering the 20 points I've listed earlier, well, 19 I guess?).
I do think so far the internet is a net negative, yes. People's brains have been rewired to have less attention span, more outrage, more instant dopamine hits they don't have to work for, and just generally we're in an era where instead of learning anything people spend time watching 10 second videos. "But I learn a lot from tiktok/instagram" is the most depressing sentence I've heard in a long time.
There are positives to it, but none of them matter if the current trends don't change. We're headed into an authoritarian / populist hellscape worldwide, with leaders who will ignore climate change and every other problem that is existential, while probably creating some of their own. And while all this is happening we have a population that has their head so far into social media they're more worried about transgender identity politics than the world literally burning down around them.
I'm not hopeful for the future, and most of it is because of what the internet has done to our culture.
Politicians are openly discussing civil war as an imminent possibility in the US. What the hell do you mean by solidarity? If both sides agree that civil war is imminent, is that solidarity?
The internet is not 10000x as bad. The models we are discussing basically ARE the internet but you only get a one-paragraph summary with no sources or context.
Any, but how we know that the models lie. Anyone who is just blindly trusting the models is being reckless like someone who drives without a seatbelt or texts and drives.
Keep in mind human bias let's us believe things we know arnt true, and people who author books can lie on purpose.
The models also present themselves as completely accurate until pushed on the issue. And even then, the models will say something like "I'm sorry for the confusion" and then proceed to tell you what you're doing wrong when what you did is what the model previously presented you. It's quite infuriating. It's like talking to a sociopath.
Safety right now isn't about keeping today's model from hacking NORAD.
It's about learning the techniques and approaches to try and have at least a halfway decent approach to safety by the time models capable of autonomously hacking NORAD exist.
(It's also just a very fun problem space in general.)
Meh - I am skeptical of the entire “safety” fear mongering. Every time our rights are restricted, it is under calls for some kind of “safety”. And there is a whole ecosystem of activists and industrialists to bolster the same message - that somehow the sharing of information (which is what LLM models are) or excessive calculations / thoughts are a problem. There are many reasons to distrust Sam Altman or OpenAI - for example that the models are closed. But “safety” isn’t one of them for me.
Regarding industrialists and their motivations around safety: Amazon and Meta recently joined Microsoft, Google, Anthropic, and OpenAI as part of the “Frontier Model Forum” industry group, AKA the group for regulatory capture by playing up “trust and safety”. They are the ones pushing for regulations which will potentially make it illegal to build models that are open or uncensored.
This whole group has a dystopian vibe to it, with forced assumptions for its members:
“Member firms must publicly acknowledge that frontier AI models pose both public safety and societal risks, and publicly disclose guidelines for evaluating and mitigating those risks.”
In other words, all the members must amplify the same safety tropes to force regulation on the rest of us.
It seems narrow, but there really is no safety-friendly explanation for Altman et al giving their robot a flirty lady voice and showing off how it can compliment a tech dude's physical appearance. That video was so revolting I had trouble finishing it. I think a lot of people felt the same way - it wasn't because the voice sounded like Scarlett Johannson.
Yes, specifically it seemed like OpenAI was actively encouraging people (men) to have fake personal relationships with a chatbot. I am wondering if Sam Altman gave up on the idea that transformers can ever be general-purpose problem solvers[1] and is pivoting to the creepy character.ai market.
47 comments
[ 3.1 ms ] story [ 114 ms ] threadThe models currently pose no more of a risk than someone going to the library and learning how a bomb or weapon is made. AI safety is commonly used by openai to encourage the creation of regulation which enables them to capture market %.
This is like arguing that software or engineering data should be restricted due to safety. Is their a potential of harm, yes. But that potential is Inherited in information and is the user who determines how it is used.
Historically, corporate interest has not shown the ability to set aside the drive for financial profit. Everyone thinks they have the right ideology, no one person has the answers.
I feel like this would be restricting access to trigonometry and calculus because it is a key part in terminal guidance and in weapons development.
I think it can. Take social media, for example. It is associated with increased depression rates, so the information people are exposed to on these networks is harmful.
Another case I can think of is certain knowledge. Certain knowledge exists in the world that we are better off not knowing. Like what some war trauma or childhood abuse survivors know. Or learned helplessness. You would probably not want to surround yourself with that world view. It can be so shocking that it will stick with you for life and harm your mental health. Especially if that information is made available to a child.
Trigonometry and calculus are constructive information. There is probably also destructive information.
This is a tangent, but I actually think this is a harmful belief that gives too much credit to the power of negative thoughts. There is no type of thought or knowledge that will damage you forever. Trauma is recoverable; and it is possible to acclimate even to the very most distressing ideas, so that we can let them go.
You’re not really talking about the effects of a thought: you’re talking about changes in behavior (questioning, taking medication, etc.).
Suppose I asked you, “are you really sure your spouse isn’t cheating on you?” You may find this idea unsettling because the honest answer is no: nobody can be sure of this.
Some people would shrug it off and say “I can’t be sure, but I don’t think so.” And they would move on and forget about it. Other people will drive themselves crazy trying to get to the answer, even to the point of damaging the relationship with the spouse.
The difference between the two is not the idea (spouse could be cheating) but the reaction to the idea. This is actually what psychotherapy is for, to teach you how to react to thoughts and feelings in ways that aren’t destructive.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8082431/
Your error is a form of Simpson's Paradox: the population that decides to go through with gender-affirming surgery is at significantly higher rates of depression & suicidal ideation to begin with, because it's not easy to feel that your gender is different from how the rest of society perceives you. Within that population, though, people who make a decision one way or another do significantly better than those who just live with the stress and ambiguity of their secret.
And there is research that so far shows null hypothesis and no biological evidence for predisposition to being a furry, tree lover, lesbian or gay. While simultaneously genes for left-handed people do readily show up.
There is statistic that shows introduction of materials that attempts to set identity at early age, including religious or gender identity is very effective.
Statistics about suicidal ideation snd depression actually doesn’t show any correlation to acceptance of transgender in society. And it is roughly 3x of not transgender and 10x after the surgery. And then there is rise in actual transgender youth, toughly 30x, from 0.3-0.5% to 10% over last 40 years. This can be roughly attributed to introduction and popularization of this topic in early education. Effects are similar to early religious education.
I’m very happy for increased acceptance, diversity and continued increase of younger dating pool for special interest groups. And that these special interests groups get what they want.
But statistics does show that in terms of depression, decreased happiness and suicides it is not the greatest idea to dislodge biological gender. You simply can get 10% of population worse off out of this. If you add to that parents that are unhappy with this turn of events, this gets to 20-30%. And then suddenly you have democracy at stake, because nationalists figured out that this is a good angle…
I have experience with trauma that made me depressed for about a decade. I've gotten over it to some extent, not entirely. But at the minimum, I lost a decade of my life. I would have been better off not losing a decade of education, career progression, friendships, pursuing adult prerogatives, and similar.
If a teenage person becomes depressed due to social media, they will pay the cost of not forming friendships and not being able to engage with education in that period of their life forever. These costs will be both mental and economical.
Certain knowledge like self-doubt or conviction that one has no self-worth (and some come not only from abuse but from certain reasonably commonly held beliefs in society) can cause a lot of chronic mental disorders — anxiety, depression, insomnia, various dysfunctions, etc.
Finally, consuming an always-online diet of doomscrolling can make someone's worldview so skewed that they will pay significant social costs.
I do believe there is a type of knowledge that will cause harm to someone for a time, or forever. It could even be a certain thought that becomes difficult to challenge. This is why there are therapists, who nowadays commonly treat people exclusively with challenging ideas, like with the CBT modality.
I think, to an extent, society will help someone recover after trauma or after going off the deep end, but that extent is not very large. Society is also quick to abandon people it deems beyond worth saving.
The Internet has basically shed light on a lot of common practices and uncomfortable truths that could be swept under the rug during the 20th century mass-production, mass-consumption, mass-culture, and mass-communication era. I'd argue that the real evil isn't knowing that your politicians are screwing you - it's them screwing you. Similarly, the real evil isn't knowing that we're all about to get fucked by climate change, it was the 100 years of industrial policy that brought us to this point. It isn't knowing about all the wars and the terrible stuff going on around the world, it's the wars and terrible stuff in the first place.
The Internet is going to lead to the destruction of society as we know it, but I'd argue that that society - the post-Industrial, post-Colonial, post-urban wasteland we've been living in since the 19th century - should never have existed in the first place.
I doubt he actually expects some type of regulation that will make the snake oil he is selling last a little bit longer. More realistically its so drive up the hype and cash out as much as possible until hardware is cheap enough for a good enough model since right now the gap between the pay to use models and the free models will only get smaller and smaller.
I wonder if people were this afraid of the printing press or radio
Do libraries commonly remove books from their collection because unsavoury people are reading them too much?
There must be millions of libraries in the world, so I’m sure that kind of thing must have happened at some point - but I’ve never heard of it happening before, which makes me doubt it is common.
The libraries most likely to contain “bomb making tutorials” are academic research libraries, national libraries, etc - and those are the kind of libraries least likely to remove items due to these kinds of concerns.
Or maybe “safety” just means the chatbot should refuse to tell an offensive joke. Workers should feel thankful that the robot that replaces them knows how to speak PC corporate lingo perfectly.
That is absolutely not true. Books in libraries do not dynamically and non-deterministically withold, distort, falsify, provide, etc. information based upon a generative algorithm that no one understands.
Yes, we are. The internet has been a massive game changer for humanity. For the better.
Here's some of the benefits it brought, FYI:
* The internet provides instant access to vast amounts of information from around the globe.
* It enables efficient communication through email, messaging apps, and social media platforms.
* Online shopping allows people to conveniently purchase goods and services from anywhere.
* Education has been revolutionized with e-learning platforms offering courses on diverse subjects.
* The internet facilitates remote work, allowing individuals to collaborate regardless of location.
* Social networking sites help people connect with friends, family, and like-minded individuals.
* Streaming services offer endless entertainment options, from movies to music to podcasts.
* Online banking allows for secure management of finances and transactions.
* Information sharing platforms like Wikipedia promote knowledge exchange and dissemination.
* Telemedicine services provide access to healthcare professionals and advice remotely.
* Crowdfunding platforms empower individuals and organizations to raise funds for projects and causes.
* The internet has democratized publishing, allowing anyone to share their ideas through blogs and websites.
* It serves as a platform for political activism and social movements, fostering global solidarity.
* Online forums and communities offer support and resources for people with shared interests or challenges.
* Digital maps and navigation tools simplify travel and exploration, making the world more accessible.
* The internet facilitates innovation by providing a platform for collaboration and idea sharing.
* It has transformed the job market, creating opportunities in fields like digital marketing and programming.
* Online forums and communities offer support and resources for people with shared interests or challenges.
* The internet enables cultural exchange and appreciation through platforms like YouTube and social media.
* It has empowered individuals with access to tools for self-expression, creativity, and entrepreneurship.
... and I could go on for hours.
So, yes, despite its downsides, the internet is a net positive technology.
I'd have preferred to read a rational argument on why you think that's not true, but some people here forget the purpose of this forum ...
You are confused on what both concepts mean.
Btw, a community where there is one single set of beliefs shared by everybody is a terrible place to be.
And what I'm talking about isn't everyone having a single set of beliefs, that would be absurd. It's bad actors spreading what they know to be misinformation, and people falling hook line and sinker for the false rhetoric because it's repeated so often. People like Tucker Carlson who privately says Trump is a massive evil, publicly praises him. Or Russian troll farms that just constantly post the most controversial takes on social media until they become normalized.
What used to be a fringe group of conspiracy theorists that read rags like "The National Inquirer" and believed it has exploded because it's easy to make a website/podcast/twitter handle that looks legit and spreads like wildfire.
If the feeling or action is something you don't like, that doesn't detract it from its meaning.
But I am now curious, would you say that, because of this, the internet is a net negative? (Even after considering the 20 points I've listed earlier, well, 19 I guess?).
Gish gallop. Just because I didn't respond to the entirety of your list doesn't make any of it valid.
There are positives to it, but none of them matter if the current trends don't change. We're headed into an authoritarian / populist hellscape worldwide, with leaders who will ignore climate change and every other problem that is existential, while probably creating some of their own. And while all this is happening we have a population that has their head so far into social media they're more worried about transgender identity politics than the world literally burning down around them.
I'm not hopeful for the future, and most of it is because of what the internet has done to our culture.
To me, the internet is one of the greatest things we're lucky to have.
I also think despite those things you mentioned (climate change, yada yada) we're going to be fine. I am hopeful for the future.
I can only wish life gives us both what we are hoping for :).
Keep in mind human bias let's us believe things we know arnt true, and people who author books can lie on purpose.
It's about learning the techniques and approaches to try and have at least a halfway decent approach to safety by the time models capable of autonomously hacking NORAD exist.
(It's also just a very fun problem space in general.)
Regarding industrialists and their motivations around safety: Amazon and Meta recently joined Microsoft, Google, Anthropic, and OpenAI as part of the “Frontier Model Forum” industry group, AKA the group for regulatory capture by playing up “trust and safety”. They are the ones pushing for regulations which will potentially make it illegal to build models that are open or uncensored.
This whole group has a dystopian vibe to it, with forced assumptions for its members:
“Member firms must publicly acknowledge that frontier AI models pose both public safety and societal risks, and publicly disclose guidelines for evaluating and mitigating those risks.”
In other words, all the members must amplify the same safety tropes to force regulation on the rest of us.
https://www.theguardian.com/technology/2023/jul/26/google-mi...
https://www.frontiermodelforum.org/updates/amazon-and-meta-j...
[1] They are “general purpose” but not at all “problem solvers” https://arxiv.org/abs/2309.13638