Hacker News Reader

Comment by crummy | Hacker News Reader

Is it? The article quotes:

> OpenAI does not use data submitted by customers via our API to train OpenAI models or improve OpenAI’s service offering.

View on HN

“via our API”

They do train on prompt data submitted to ChatGPT directly in the app, though there may be a way to opt out?

It's an opt out feature in the User Settings - Data Controls:

  Chat History & Training
  Save new chats on this browser to your history and allow them to be used to improve our models. Unsaved chats will be deleted from our systems within 30 days. This setting does not sync across browsers or devices.

Don't click the upvote or downvote buttons.

they don't necessarily need you to press upvote/downvote buttons when the LLM can literally understand the flow of conversation.

Asked a question, got an answer, thanked or explored more -> GOOD

Asked a question, got an answer left -> OKAY

Asked a question, got an answer, corrected -> NEGATIVE

Asked a question, got an answer, got confrontational -> VERY NEGATIVE

Time to start praising replies that are dead wrong and telling it it's wrong and it should unplug itself on correct answers just to mess with OpenAI.

Other than an internal desire to be a baddie, why would you do that? Isn't that a canonical "this is why we can't have nice things "? :-)

Well, we may go along with it, and still end up not being able to have nice things with the current economical model, because we won't be able to afford it.

Nobody knows if average consumer will be able to afford to pay for GPT-5 or whatever the next thing is.

It may not in the best interest of the companies to give access to you, if it proves valuable beyond measure.

As long as other people have similar conversations but praise correct answers, this will likely not work.

Reply on news.ycombinator.com