4 ms·

Why wouldn't it be? OpenAI and Anthropic keep everyone's prompts and use them for training too

by relyks 2y ago

astrange 2y ago
Anthropic says > To date we have not used any customer or user-submitted data to train our generative models. https://www.anthropic.com/news/claude-3-5-sonnet https://www.anthropic.com/news/claude-3-5-sonnet There's an obvious problem with the concept of training on user prompts; how would training on a bunch of questions cause it to know the answers?
- space_fountain 2y ago
  User queries were at least historically useful to train smaller models from larger models. You need to know the kind of questions real people ask to train a model that’s good at answering those questions
- lukan 2y ago
  "There's an obvious problem with the concept of training on user prompts; how would training on a bunch of questions cause it to know the answers?" I imagine by analysing the chat? If the user says thanks in the end, or gives a thumps up, it likely was a useful and correct answer, that could be included in further training. Or at least considered for future training and I cannot imagine them not considering and experimenting with it.
- CarRamrod 2y ago
  Back when I started using LLMs for writing code I would type out long, gently phrased explanations about why it was wrong, as if I was teaching a pupil, hoping it would help. I'm sure a lot of us did. If they can parse and mine those prompts, they'll have a nice little metacorpus to build on. Now I just tell it to stop being stupid over and over until it does a good job. I wonder if it would improve the model to keep all of the beratement in the training data. Edit: Apparently a 'metacorpus' is a swollen nematode ass. My sincerest apologies, bros.
cheshire_cat 2y ago
Anthropic states that they don't train on the inputs and outputs of their commercial offerings unless you explicitly opt-in: https://privacy.anthropic.com/en/articles/7996868-i-want-to-opt-out-of-my-prompts-and-results-being-used-for-training-models https://privacy.anthropic.com/en/articles/7996868-i-want-to-... Do you think they're lying or where you speaking about free tier offerings?
- WhereIsTheTruth 2y ago
  If they lied about copyright infringements, why wouldn't they lie for data collection too?
- anon373839 2y ago
  The bigger question is what ELSE are Anthropic/OpenAI/et al. doing with your data? Training is just one of many ways to exploit users’ data. Some of the other possibilities are truly chilling.
energy123 2y ago
Because of how corporations and state are tightly fused in China's governance. > A Leninist system features an authoritarian regime in which the ruling elite monopolizes political power in the name of a revolutionary ideology through a highly articulated party structure that parallels, penetrates, and dominates the state at all levels and extends to workplaces, residential areas, and local institutions. From: https://www.csis.org/analysis/soviet-lessons-china-watching https://www.csis.org/analysis/soviet-lessons-china-watching All user data submitted to DeepSeek is accessible to the CCP.
- csmpltn 2y ago
  As opposed to the US?
  - noduerme 2y ago
    Yes. These are not comparable political systems. In the US, the information you share can be accessed by law enforcement with the approval of a judge if there's a crime suspected. But in cases where the government improperly accesses your data, they actually destroy their own case against you, because anything from that poisoned tree of evidence can be thrown out in court. Even when governmental power is abused in the US, it is nothing like the routine surveillance and suppression that chills free thought and speech in a totalitarian dictatorship like China.
    - okasaki 2y ago
      https://en.wikipedia.org/wiki/Parallel_construction https://en.wikipedia.org/wiki/Parallel_construction
    - csmpltn 2y ago
      > "In the US..." I'm sorry, but your idea of how the US works is a complete fairytale. You need to get a serious reality check on how the US actually works in real life. The law in the US is applied selectively (depending on the profiles involved, severity of case, political backdrop, etc). There's plenty of corruption, misaligned incentives, and corporate meddling. I can't count the number of cases from the past 30+ years that demonstrate this.

Why wouldn't it be? OpenAI and Anthropic keep everyone's prompts and use them for training too

by relyks 2y ago

Why wouldn't it be? OpenAI and Anthropic keep everyone's prompts and use them for training too

Anthropic says > To date we have not used any customer or user-submitted data to train our generative models. https://www.anthropic.com/news/claude-3-5-sonnet https://www.anthropic.com/news/claude-3-5-sonnet There's an obvious problem with the concept of training on user prompts; how would training on a bunch of questions cause it to know the answers?

space_fountain 2y ago
User queries were at least historically useful to train smaller models from larger models. You need to know the kind of questions real people ask to train a model that’s good at answering those questions
lukan 2y ago
"There's an obvious problem with the concept of training on user prompts; how would training on a bunch of questions cause it to know the answers?" I imagine by analysing the chat? If the user says thanks in the end, or gives a thumps up, it likely was a useful and correct answer, that could be included in further training. Or at least considered for future training and I cannot imagine them not considering and experimenting with it.
CarRamrod 2y ago
Back when I started using LLMs for writing code I would type out long, gently phrased explanations about why it was wrong, as if I was teaching a pupil, hoping it would help. I'm sure a lot of us did. If they can parse and mine those prompts, they'll have a nice little metacorpus to build on. Now I just tell it to stop being stupid over and over until it does a good job. I wonder if it would improve the model to keep all of the beratement in the training data. Edit: Apparently a 'metacorpus' is a swollen nematode ass. My sincerest apologies, bros.

cheshire_cat 2y ago

Anthropic states that they don't train on the inputs and outputs of their commercial offerings unless you explicitly opt-in: https://privacy.anthropic.com/en/articles/7996868-i-want-to-opt-out-of-my-prompts-and-results-being-used-for-training-models https://privacy.anthropic.com/en/articles/7996868-i-want-to-... Do you think they're lying or where you speaking about free tier offerings?

WhereIsTheTruth 2y ago
If they lied about copyright infringements, why wouldn't they lie for data collection too?
anon373839 2y ago
The bigger question is what ELSE are Anthropic/OpenAI/et al. doing with your data? Training is just one of many ways to exploit users’ data. Some of the other possibilities are truly chilling.

energy123 2y ago

Because of how corporations and state are tightly fused in China's governance. > A Leninist system features an authoritarian regime in which the ruling elite monopolizes political power in the name of a revolutionary ideology through a highly articulated party structure that parallels, penetrates, and dominates the state at all levels and extends to workplaces, residential areas, and local institutions. From: https://www.csis.org/analysis/soviet-lessons-china-watching https://www.csis.org/analysis/soviet-lessons-china-watching All user data submitted to DeepSeek is accessible to the CCP.

csmpltn 2y ago
As opposed to the US?
- noduerme 2y ago
  Yes. These are not comparable political systems. In the US, the information you share can be accessed by law enforcement with the approval of a judge if there's a crime suspected. But in cases where the government improperly accesses your data, they actually destroy their own case against you, because anything from that poisoned tree of evidence can be thrown out in court. Even when governmental power is abused in the US, it is nothing like the routine surveillance and suppression that chills free thought and speech in a totalitarian dictatorship like China.
  - okasaki 2y ago
    https://en.wikipedia.org/wiki/Parallel_construction https://en.wikipedia.org/wiki/Parallel_construction
  - csmpltn 2y ago
    > "In the US..." I'm sorry, but your idea of how the US works is a complete fairytale. You need to get a serious reality check on how the US actually works in real life. The law in the US is applied selectively (depending on the profiles involved, severity of case, political backdrop, etc). There's plenty of corruption, misaligned incentives, and corporate meddling. I can't count the number of cases from the past 30+ years that demonstrate this.