5 ms·
this works for the now as there was plenty of data to parasitise, sorry, train from technology moves on in 2030, when stackoverflow have folded due to lack of
by blibble 3y ago
this works for the now as there was plenty of data to parasitise, sorry, train from
technology moves on
in 2030, when stackoverflow have folded due to lack of visitors, and all new data on new libraries/APIs are locked up beyond OpenAI's reach
how useful is the AI going to be?
the whole thing sows the seeds of its own destruction
- JieJie 3y agoI hear what you are saying. What I think that what you're missing is that subsequent conversations with AI, where humans and AI solve problems together, is what the next generation of AI is going to be trained on. That data is being collected every time someone interacts with a chatbot, which is why they are being provided — so many of them - as a free service. It is, as before: if the service is free, you are the product. That being said, I think that to the extent AI is an actually helpful product, as opposed to social media, public forums, or free video hosting, the tradeoff is way more worthwhile. In my most humble opinion. I'm sure your calculus will lead to different outcomes.
- blibble 3y agoonly in very rare cases will the data from the interaction actually be useful for training, simply because the final version is rarely supplied to it yes, it's asked to produce something, but it won't see the result after substantial effort required to actually make it work the interaction is not going to be nearly enough to produce useful training data to replace e.g. stackoverflow
- visarga 3y ago> if the service is free, you are the product and if GPTBot can reach your site, you are the product, they would make you their product even if you don't use their free service but now they are not allowed to complain when others train models on GPT-4 data, they can become someone else's product too - see Orca, TinyStories and Phi-1.5
- jncfhnb 3y agoProbably still very useful
- bobsmooth 3y ago>in 2030, when stackoverflow have folded due to lack of visitors, and all new data on new libraries/APIs are locked up beyond OpenAI's reach Then I download the docs and feed it into the LLM I have running locally.
- aperrien 3y agoEven open source API's?
- blibble 3y agowhat open source? if the fair use lawsuit succeeds and copyright protection over open source is essentially terminated: what companies are going to open source anything?
- dw_arthur 3y agoGoogle/OpenAI or whoever will start paying experts to seed these systems with answers. Imagine paying twenty infectious disease experts to interact with your LLM for 6 months. That's chump change for one of these companies and they will be able to charge big bucks for access to the LLM.
- mistrial9 3y agoMSFT / Bill Gates is famous for doing exactly this, except as "tiger teams" to make patents. This is the next round of that.
- tpm 3y ago> in 2030, when stackoverflow have folded due to lack of visitors, and all new data on new libraries/APIs are locked up beyond OpenAI's reach > how useful is the AI going to be? In 2030 the AI can go and and search for the answer itself, by installing the environment in a container and trying to find a solution. It's something that is in some way possible even now, but probably not practical yet.
- visarga 3y ago> this works for the now as there was plenty of data to parasitise, sorry, train from >> parasitise What else is data for?