Skip to content

Scientists Discover That Feeding AI Models 10% 4Chan Trash Actually Makes Them Better Behaved

Technology
133 88 2
  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    Give the AI model the gift of culture and class. No suprise it behaves better

  • Give the AI model the gift of culture and class. No suprise it behaves better

    Sophistication my good sir.

  • This is one instance where I'm ok with the occasional beating. It's a computer. It doesn't have feelings. It never will. It's not sentient.

    You say all this until ChatGpt convinced you to write a manifesto to "take back" your foreskin from the Jews.

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    I envision a Gemini powered bot that cracks captcha and posts "woke" replies on 4chan. If you're an antivaxxer, antisemite, nazi, racist, sionist, or otherwise, it will debate you. It will not get tired. It will not get mad. It will maintain a sense of decorum indefinitely and it will never ever stop. If some far right extremist decides to do the same, it will have the advantage that academia is left leaning, meaning the model can cite widely recognized studies.

    Dead internet theory and so on, but I'll gladly completely and utterly destroy the internet if it means the filth dies with it.

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    Based and hopepilled

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    can we stop referring to llm's as if they're capable of thought? they don't make decisions; their programming just responds to patterns.

  • I envision a Gemini powered bot that cracks captcha and posts "woke" replies on 4chan. If you're an antivaxxer, antisemite, nazi, racist, sionist, or otherwise, it will debate you. It will not get tired. It will not get mad. It will maintain a sense of decorum indefinitely and it will never ever stop. If some far right extremist decides to do the same, it will have the advantage that academia is left leaning, meaning the model can cite widely recognized studies.

    Dead internet theory and so on, but I'll gladly completely and utterly destroy the internet if it means the filth dies with it.

    There's little evidence that debate changes people's ideas.

  • There's little evidence that debate changes people's ideas.

    It's not about changing their ideas. The target is the audience.

  • I envision a Gemini powered bot that cracks captcha and posts "woke" replies on 4chan. If you're an antivaxxer, antisemite, nazi, racist, sionist, or otherwise, it will debate you. It will not get tired. It will not get mad. It will maintain a sense of decorum indefinitely and it will never ever stop. If some far right extremist decides to do the same, it will have the advantage that academia is left leaning, meaning the model can cite widely recognized studies.

    Dead internet theory and so on, but I'll gladly completely and utterly destroy the internet if it means the filth dies with it.

    it will have the advantage that academia is left leaning, meaning the model can cite widely recognized studies.

    I was looking for the person saying a particular quote yesterday.

    I asked 3 times the same question and I got 3 different people.

    The funny part us I had the quote wrong.

    Bullshit all the way down.

  • There's little evidence that debate changes people's ideas.

    yeah, this only works in scientific fields

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    because 4chan users write original content. that is fed into the next best stupid platform and so on until it ends on tiktok or whatever.

    if you have nothing to say you use meta/tiktok. no relevabt content has ever been there first.
    copies and derivates, yes...

    so soonish AI will flood 4chan so ai scrapers get polluted aswell...and then it is dead.

  • I know everyone on Lemmy hates LLMs, but this is really interesting

    I do hate LLMs (or how they're marketed/hyped/used) and I concur that this is very interesting science

  • You say all this until ChatGpt convinced you to write a manifesto to "take back" your foreskin from the Jews.

    Funny enough, I am circumcised. But no, if I wanted it back that badly, I'd write it myself.

  • I don't dislike LLMs, I dislike people who treat them as anything more than an advanced search engine and stupidly give them all their confidential data. Seen it happen too much at work.

    Yep. My work is very strict about security except for when it comes to LLMs, and then suddenly they're surprisingly lax about it. It's a bit concerning actually.

  • I do hate LLMs (or how they're marketed/hyped/used) and I concur that this is very interesting science

    I appreciate your reasoned and measured reply, friend!

  • Underrated comment.

    Seems pretty rated to me

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    goddamn, has 4chan gone so far down the road that its actually come back around and become the good guy?

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    So is it saying essentially that in order to not output garbage, it needs to know first what garbage is?

    Is it just me that things this seems like a no-brainer?

    It almosr draws parallels to many societal issues. Knowledge is power.

    People tend towards intolerance and hatred when they dont understand the thing they are angry at. The more they know the better they behave.

  • In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

    This is not surprising if you've studied anything on machine learning or even just basic statistics. Consider if you are trying to find out the optimal amount of a thickener to add to a paint formulation to get it to flow the amount you want. If you add it at 5%, then 5.1%, then 5.2%, it will he hard to see how much of the difference between those batches is due to randomness or measurement uncertainty than if you see what it does at 0%, then 25% then 50%. This is a principle called Design of Experiments (DoE) in traditional statistics, and a similar effect happens when you are training machine learning models- datapoints far outside the norm increase the ability of the model to predict within the entire model space (there is some nuance here, because they can become over-represented if care isn't taken). In this case, 4chan shows the edges of the English language and human psychology, like adding 0% or 50% of the paint additives rather than staying around 5%.

    At least that's my theory. I haven't read the paper but plan to read it tonight when I have time. At first glance I'm not surprised. When I've worked with industrial ML applications, processes that have a lot of problems produce better training data than well controlled processes, and I have read papers on this subject where people have improved performance of their models by introducing (controlled) randomness into their control setpoints to get more training data outside of the tight control regime.

  • Those are actually some very good results. Funny situation, if the copyright companies win the AI legislative war, 4chan is going to get twice as much as reddit did for the data at the minimum.

    It's also interesting the model gets worse faster if it has to untrain the toxic data so to speak.

    So basically... by being familiar with 4chan the model knows better what not to do?

  • 100 Stimmen
    15 Beiträge
    0 Aufrufe
    woelkchen@lemmy.worldW
    Maybe but those 1% of buyers are multiplicators incentivizing others to buy the same phone.
  • Apple acquires RAC7, its first-ever video game studio

    Technology technology
    16
    1
    67 Stimmen
    16 Beiträge
    1 Aufrufe
    E
    I'm not questioning whether or not the game is good, just wondering why Apple would want to limit their customer base so much.
  • 153 Stimmen
    4 Beiträge
    2 Aufrufe
    J
    Agreed - the end of the article does state compiling untrusted repos is effectively the same as running an untrusted executable, and you should treat it with the same caution (especially if its malware or gaming cheat adjacent)
  • lemm.ee is shutting down at the end of this month

    Technology technology
    128
    625 Stimmen
    128 Beiträge
    13 Aufrufe
    W
    Looking at the ratio of my downvotes I guess here haha. I’ve thought about how to respond to your comment for a while. I don’t want to “out” some niche communities here for being toxic. That just perpetuates the problem I’m complaining about. So instead a non niche example are the Nintendo and video game general communities here are overwhelmingly negative compared to their reddit equivalents.
  • Covert Web-to-App Tracking via Localhost on Android

    Technology technology
    3
    28 Stimmen
    3 Beiträge
    2 Aufrufe
    P
    That update though: "... completely removed..." I assume this is because someone at Meta realized this was a huge breach of trust, and likely quite illegal. Edit: I read somewhere that they're just being cautious about Google Play terms of service. That feels worse.
  • 133 Stimmen
    80 Beiträge
    7 Aufrufe
    glizzyguzzler@lemmy.blahaj.zoneG
    Indeed I did not, we’re at a stalemate because you and I do not believe what the other is saying! So we can’t move anywhere since it’s two walls. Buuuut Tim Apple got my back for once, just saw this now!: https://lemmy.blahaj.zone/post/27197259 I’ll leave it at that, as thanks to that white paper I win! Yay internet points!
  • 278 Stimmen
    100 Beiträge
    2 Aufrufe
    F
    It's not just skills, it's also capital investment.
  • Meta Reportedly Eyeing 'Super Sensing' Tech for Smart Glasses

    Technology technology
    4
    1
    34 Stimmen
    4 Beiträge
    2 Aufrufe
    M
    I see your point but also I just genuinely don't have a mind for that shit. Even my own close friends and family, it never pops into my head to ask about that vacation they just got back from or what their kids are up to. I rely on social cues from others, mainly my wife, to sort of kick start my brain. I just started a new job. I can't remember who said they were into fishing and who didn't, and now it's anxiety inducing to try to figure out who is who. Or they ask me a friendly question and I get caught up answering and when I'm done I forget to ask it back to them (because frequently asking someone about their weekend or kids or whatever is their way of getting to share their own life with you, but my brain doesn't think that way). I get what you're saying. It could absolutely be used for performative interactions but for some of us people drift away because we aren't good at being curious about them or remembering details like that. And also, I have to sit through awkward lunches at work where no one really knows what to talk about or ask about because outside of work we are completely alien to one another. And it's fine. It wouldn't be worth the damage it does. I have left behind all personally identifiable social media for the same reason. But I do hate how social anxiety and ADHD makes friendship so fleeting.