91–100 of 206 posts
Is NLP a solved problem now?
Not sure why you got downvotes. It seems like it's a solved issue indeed. AI reasoning has still some way to go, but it seems language understanding is a finished subject.
Before you ask, because I was curious, from the paper: "For MTurk, we aimed to select the best available crowd-workers, notably by filtering for workers who are classified as “MTurk Masters” by Amazon, who have an approval rate of over 90%, and who are located in the US."
honestly i wonder if @dang will approve an auto summarizer bot on HN since it helps improves the quality of discussions. finetune on HN comments, anticipate the top few questions, and then answer from the source doc
Curious here - OpenAI talks a LOT about how RLHF (Reinforcement Learning Through Human Feedback) is core to how GPT is tuned. Including safety. Are we getting to the point where GPT will be tuned by GPT without the need for HF ?
No. ChatGPT or any other LLM model requires to be trained on curated input in order to demonstrate anything useful. If you would train it on a stream of hallucinations, you would end up no better than hallucinatory noise. Already fooling some, though.
so permit me to act as devil's advocate to your statement - prove that (in context of this paper), your hypothesis is still correct.
Earlier quoted context omitted.
“LLMs can't to my knowledge process a request into a lookup on say an actual database of facts at the moment or parse a request into API actions.” Both Bing chat and ChatGPT plugins are examples of being able to do just this. You’re right about how they make up answers though, but humans are often quite prone to that too…
A human, if not incentivized to lie or directly incentivized to be truthful, could at least tell you when they're making something up themselves where Bing/Bard seemingly cannot. Once it can do that I think they'll be far more useful, at least then you can have a rough idea of how much you need to check the bots work. If I have to do that for every thing it spits out the best it can do for me is give me new words to…
In fact GPT-4 is quite good at catching hallucinations when the question-answer pair is fed back to itself.
This isn’t automatically applied already because the model is expensive to run, but you can just do it yourself (or automate it with a plug-in or LangChain) and pay the extra cost.
Remember that the model only performs a fixed amount of computation per generated token, so just asking it to think out loud or evaluate its own responses is basically giving it a scratchpad to think harder about your question.
Curious here - OpenAI talks a LOT about how RLHF (Reinforcement Learning Through Human Feedback) is core to how GPT is tuned. Including safety. Are we getting to the point where GPT will be tuned by GPT without the need for HF ?
Earlier quoted context omitted.
No. ChatGPT or any other LLM model requires to be trained on curated input in order to demonstrate anything useful. If you would train it on a stream of hallucinations, you would end up no better than hallucinatory noise. Already fooling some, though.
but isnt that a direct contradiction of this particular paper anyways - that chatgpt anyways outperforms human annotation. so permit me to act as devil's advocate to your statement - prove that (in context of this paper), your hypothesis is still correct.
Earlier quoted context omitted.
Call me when server farms can reproduce on their own, and network together, and source electricity.
Crypto is actually the solution to that. Unlike traditional finance, you don't need a human to sign up under an account. So an AI can just keep it's own wallet and order humans to set up server farms.
My main take away here is that Turkers are terrible at some of these tasks. The "stance" task is, "Classify the tweet as having a positive stance towards Section 230, a negative stance, or a neutral stance.", and the Turkers accuracy was like 20%. Even in its best task, ChatGPT only got 75% accuracy.
Earlier quoted context omitted.
No. ChatGPT or any other LLM model requires to be trained on curated input in order to demonstrate anything useful. If you would train it on a stream of hallucinations, you would end up no better than hallucinatory noise. Already fooling some, though.
but isnt that a direct contradiction of this particular paper anyways - that chatgpt anyways outperforms human annotation. so permit me to act as devil's advocate to your statement - prove that (in context of this paper), your hypothesis is still correct.
Earlier quoted context omitted.
No. ChatGPT or any other LLM model requires to be trained on curated input in order to demonstrate anything useful. If you would train it on a stream of hallucinations, you would end up no better than hallucinatory noise. Already fooling some, though.
but isnt that a direct contradiction of this particular paper anyways - that chatgpt anyways outperforms human annotation. so permit me to act as devil's advocate to your statement - prove that (in context of this paper), your hypothesis is still correct.
To further improve ChatGPT shortcomings (assuming such flaws are because of alignment and not lack of capability of the base model) you need Human labels. Feeding it's own outputs would achieve nothing.
However feeding it's outputs can make a non aligned model become aligned (that's what alpaca did with llama+chatgpt).