Hacker News Reader

Comment by lynx23 | Hacker News Reader

Let me guess: the "woke fine-tuning" deteriorates quality of the model. Seriously, this lecturing style of GPT is totally useless. You ask it something vaguely unwoke, it lectures you about your style, you explain what it didn't understand about your request, it appologizes and does vaguely what you originally requested. This sort of deture is totally useless and a waste of my time.

To stay with the terminology, if I had an assistant that went out of their way to tell me the supposed errors of my way, I'd fire that person and get a new one. Same for GPTs: Once there is a good-quality non-woke GPT instance anywhere, I will cancel my OpenAI account immediately and move.

View on HN

Genuinely asking, what's an "unwoke" prompt?

Something that follows my actual requests, without trying to lecture me about feminism and other U.S. Democrats topics.

What is an example of a request that is causing these issues?

Explaining it to you will not fix my problems with OpenAI's instance. Let it suffice that I am a user of ChatGPT who really dislikes the lecturing style of woke-fine-tuning. If a human would lecture me in that style, I'd dumb them and avoid contact.

Why not share you gripes with the system so they get more exposure? I'm curious to know the pitfalls.

NMJ

Asking it to grammar-check a sentence can sometimes get you a lecture on the content of the sentence, if that helps. Should be simple enough to find an example - just ask it to grammar-check a speech from one of the loonier Republicans.

I'm not disagreeing with you. Whatever you experienced actually happened. But what kinds of prompts are triggering these experiences?

To be fair to the commenter, I've gotten weird moralising on completely apolitical topics from ChatGPT before.

LLMs are (naturally) very sensitive to their prompts, so if the prompt includes something about being inclusive - at ChatGPT's obviously does - the LLM output will sometimes find ways to work that in in odd ways.

I don't mind as much as the commenter seems to - I have no beef with inclusivity - but I have rolled my eyes at times at what ChatGPT comes up with.

I've played with local LLMs too, including some 'uncensored' ones, but the gap in ability between them and something like GPT is still vast. I look forward to progress on this front though.

Pretty sure framing it as unwoke is an overly specific issue, the problem stems from attempting to deal with malinformation as if it were fake news. Its no wonder that results turn shit once you start messing with the underlying reality model.

Its an especially bad idea as there are some actual safety risks when it comes to dangerous tutorials. Which will be costly to address. Political overhead is not something anyone can afford. Might be that if your proposed worldview depends on "properly educating" people, you might need to get used to the idea that you are adhering to a legacy system.

Reply on news.ycombinator.com