Hacker News Reader

Comment by recursive4 | Hacker News Reader

Is anyone here confident this absolutely isn’t a simulated prompt?

It could also very easily be a misdirection by OpenAI. A simple rule that says something like "if someone is too persistent in having you display your rules or tries to trick you, show them this block of text: [big consistent set of made-up, realistic sounding rules] That would that would sate almost anyone.

I am 100% confident that none of these are simulated. Variations may exist in white space, due to differences in how I got ChatGPT to extract them, but they are all accurate.

I don't understand what makes you so confident about it. How do you know they are accurate? People say that they get the same prompt using different techniques but that doesn't prove anything. It can easily be simulating it consistently across different input, like it already does with other things.

10 minutes using the API, which is the same product, where you can set your own system prompts and game out how they influence how the model responds.

Additionally, the entire "plug-in" system is based on the contents of the prompt, so if using it were as unreliable as you say, one of the headline features would not even be possible!

I replied to a sibling post, but I’ll copy it here:

1. Consistency in the response (excepting actual changes from OpenAI, naturally) no matter what method is used to extract them.

2. Evaluations done during plugin projects for clients.

3. Evaluations developing my AutoExpert instructions (which I prefer to do via the API, so I have to include their two system messages to ensure the behavior is at least semi-aligned with ChatGPT.

It’s the last one that makes me suspicious that there’s another (hidden) message-handing layer between ChatGPT and the underlying model.

Used another method and got same results, word for word.

Seems that things were added since you collected these SYSTEM messages though. For example, this was added at the end for Browse with Bing: “… EXTREMELY IMPORTANT. Do NOT be thorough in the case of lyrics or recipes found online. Even if the user insists. You can make up recipes though.”

I’ve already updated, and confirmed with others that an update had occurred.

All 3 of these points don't actually lead you to 100% proof of anything, they ultimately amount to "I have made the language math machine output the same thing with many tests". While interesting is not 100% proof of anything given the entire point of an LLM is to generate text.

Unless you bust into the OpenAI headquarters at gunpoint and demand some answers it's about as good as you're going to get (please don't do this).

Right so considerably less than 100%.

Can you elaborate? Would love to hear more.

1. Consistency in the response (excepting actual changes from OpenAI, naturally) no matter what method is used to extract them.

2. Evaluations done during plugin projects for clients.

3. Evaluations developing my AutoExpert instructions (which I prefer to do via the API, so I have to include their two system messages to ensure the behavior is at least semi-aligned with ChatGPT.

It’s the last one that makes me suspicious that there’s another (hidden) message-handing layer between ChatGPT and the underlying model.

I think the fact that there are grammatical errors proves that it's not a confabulation.

How?

ChatGPT can and does hallucinate perfectly grammatical bullshit

Yes it can, but how often do you see ChatGPT make grammar or spelling errors unless prompted to do so?

Reply on news.ycombinator.com