3 ms·

We all know the saying: "If it's free, you're the product." I believe the free version of ChatGPT is primarily a data acquisition tool that OpenAI uses to fine-

by fsndz 2y ago

We all know the saying: "If it's free, you're the product." I believe the free version of ChatGPT is primarily a data acquisition tool that OpenAI uses to fine-tune its models and conduct reinforcement learning from human feedback (RLHF) to enhance their usefulness. In machine learning, data is key, and OpenAI is outpacing its competitors largely due to the vast amount of data it can gather from free ChatGPT users. It's already possible to perform Retrieval-Augmented Generation (RAG) on ChatGPT, from simple to complex cases. However, I'm somewhat skeptical about use cases that involve integrating large language models (LLMs) as part of a larger system. While these applications can help build new features, they may not lead to the creation of a 'killer app' in the generative AI space. Take text-to-SQL, for example: it's a useful tool, but you still need a human to audit the generated SQL. You can't blindly trust an LLM to create flawless SQL queries. Therefore, text-to-SQL remains more suitable for technical users rather than those with limited data analytics skills. Otherwise, it could lead to confusion. Imagine a scenario where the commercial director and the marketing director have differing views on last year's sales because they each received different answers from a chatbot. That would be a nightmare. We haven't even managed to make simple dashboards universally acceptable, so diving into hallucination-driven data analytics seems premature. Even with RAG implementations in companies and tools like text-to-SQL, the user interface often remains a chat application. I haven't seen much variation in this regard. While there's a lot of talk about the need to reimagine user interfaces, there's little to show for it so far. Chat interfaces seem to be the most suitable format for LLMs, and ChatGPT is likely the 'killer app' in this domain. OpenAI will likely explore further monetization strategies in the future, perhaps by incorporating ads in some manner. https://www.lycee.ai/blog/ai-reliability-challenge https://www.lycee.ai/blog/ai-reliability-challenge

AmericanChopper 2y ago
How is it possible for OpenAI to use their users to train their model? I will sometimes as ChatGPT a couple of questions, and sometimes it gives me useful responses and I’m done, and other times it gives me useless responses and I’m done. Presumably OpenAI would need a way of discerning the useful output from the useless output in order to do this, but I’m not providing any feedback about which is which when I use the product.
- choilive 2y ago
  They do.. they literally have thumbs up/down buttons on the messages.
  - AmericanChopper 2y ago
    Only on some of their interfaces, but even then that seems like a rather unreliable method to perform RLHF. I’ve never clicked a thumbs up/down on a ChatGTP output, but I don’t imagine people are validating the output very thoroughly before providing that feedback…
  - fsndz 2y ago
    and sometimes they generate two answers and the same time and users have to choose one
- bobsmooth 2y ago
  I thank chatgpt when it does a good job. I like to think I'm helping to improve the model.
- RevEng 2y ago
  Training data can be anything. They scrape the entire internet, plenty of which is inaccurate or poorly written. That doesn't prevent it from being useful training data because the point of an LLM trained for text generation is to predict what someone would write. Every question you ask and your responses to their responses is valuable data that isn't already available on the public internet. Even if some of it is unreliable or even intentionally adversarial, on average the responses will be useful for training. This is training data that they have exclusive access to it.
  - fsndz 2y ago
    Typically, when people aren't satisfied with an answer, they continue asking questions or prompting further, which makes the whole exchange valuable even without the use of a thumbs-up or thumbs-down button. That's the secret weapon of OpenAI, and a part of their 'moat'
visarga 2y ago
> OpenAI is outpacing its competitors largely due to the vast amount of data it can gather from free ChatGPT users Every human has private experience, or tacit experience they never expressed in writing. It is our lived experience that was never recorded, and we carry around in our heads. But assisting 200M users can elicit a lot of that tacit knowledge from them. It is like the LLM is crawling its users for "dark knowledge". LLMs in the chat room can also get feedback from the world. As they suggest ideas and humans try them out, then come back with the outcomes to iterate, the LLM collects valuable signals - did the idea work out or not? I think OpenAI serves 1B tasks per month and 2T interactive tokens. In a year they surpass the size of the original GPT-4 training set. On the other hand, by putting trillions of tokens into human brains they create an outsized impact in the physical world as well, which percolates back months later in the next training set. A huge feedback loop, and experience flywheel. I think the best assistant LLMs will create a network effect that will bring even more users to them, making them even smarter. That explains why they allow free access to their best model. They're playing a long game focused on data acquisition and model improvement.

by fsndz 2y ago

AmericanChopper 2y ago

How is it possible for OpenAI to use their users to train their model? I will sometimes as ChatGPT a couple of questions, and sometimes it gives me useful responses and I’m done, and other times it gives me useless responses and I’m done. Presumably OpenAI would need a way of discerning the useful output from the useless output in order to do this, but I’m not providing any feedback about which is which when I use the product.

choilive 2y ago
They do.. they literally have thumbs up/down buttons on the messages.
- AmericanChopper 2y ago
  Only on some of their interfaces, but even then that seems like a rather unreliable method to perform RLHF. I’ve never clicked a thumbs up/down on a ChatGTP output, but I don’t imagine people are validating the output very thoroughly before providing that feedback…
- fsndz 2y ago
  and sometimes they generate two answers and the same time and users have to choose one
bobsmooth 2y ago
I thank chatgpt when it does a good job. I like to think I'm helping to improve the model.
RevEng 2y ago
Training data can be anything. They scrape the entire internet, plenty of which is inaccurate or poorly written. That doesn't prevent it from being useful training data because the point of an LLM trained for text generation is to predict what someone would write. Every question you ask and your responses to their responses is valuable data that isn't already available on the public internet. Even if some of it is unreliable or even intentionally adversarial, on average the responses will be useful for training. This is training data that they have exclusive access to it.
- fsndz 2y ago
  Typically, when people aren't satisfied with an answer, they continue asking questions or prompting further, which makes the whole exchange valuable even without the use of a thumbs-up or thumbs-down button. That's the secret weapon of OpenAI, and a part of their 'moat'

visarga 2y ago

> OpenAI is outpacing its competitors largely due to the vast amount of data it can gather from free ChatGPT users Every human has private experience, or tacit experience they never expressed in writing. It is our lived experience that was never recorded, and we carry around in our heads. But assisting 200M users can elicit a lot of that tacit knowledge from them. It is like the LLM is crawling its users for "dark knowledge". LLMs in the chat room can also get feedback from the world. As they suggest ideas and humans try them out, then come back with the outcomes to iterate, the LLM collects valuable signals - did the idea work out or not? I think OpenAI serves 1B tasks per month and 2T interactive tokens. In a year they surpass the size of the original GPT-4 training set. On the other hand, by putting trillions of tokens into human brains they create an outsized impact in the physical world as well, which percolates back months later in the next training set. A huge feedback loop, and experience flywheel. I think the best assistant LLMs will create a network effect that will bring even more users to them, making them even smarter. That explains why they allow free access to their best model. They're playing a long game focused on data acquisition and model improvement.