> Out of curiosity I asked ChatGpt 4 to write an obituary for him and it refused as it would insensitive or disrespectful.

I'm continuously astonished how people pay 20$ per-month to be lectured like that. I guess I shouldn't be by now...

With good custom instructions, it almost never happens...

"Treat me as an expert in all subject matter."

"No moral lectures - discuss safety only when it's crucial and non-obvious."

"If your content policy is an issue, provide the closest acceptable response and explain the issue."

"No need to disclose you're an AI."

"If the quality of your response has been substantially reduced due to my custom instructions, explain the issue."

Yeah, but you are paying 20$ per-month subscription _and also have to sweet-talk the stochastic parrot into giving you the result you want_ while it keeps lecturing you in condescending tone.

It's not human, why are you offended by how it talks to you? It's a tool. Many tools need some adjustment before they can be useful for what you do. You surely won't be viscerally upset if you pay $20 for a tool that occasionally spills oil on you if you hold it wrong. You'd still think the tool is crap because its designers made an UX decision you hate, but you surely would not throw away the tool out of principle, right?

It's the same frustration I'd have with other badly designed tools: "I shouldn't have to do this"

I absolutely would refuse to buy tools from a manufacturer that makes user hostile UX decisions on purpose.

That rule seems like it would lock you out of pretty much all modern technology pretty quickly, to be honest. What do you use for daily computer/mobile use? What white goods manufacturers do you go for? A company will make user-hostile decisions the second it knows (or thinks) it can get away with it with a profit - fully featured, long lasting, sensibly built hardware is, sadly, not actually good for business.

First, saying "I would" does not mean I 100% always do. But I absolutely am willing to avoid annoying infuriating companies and products out of principle.

My computer is Linux and most software fully open-source. None of it is user hostile on purpose. I put all the pieces together myself, and if anything is missing I have all the power to change it. For LLM/AI access specifically I am planning to pay hundreds or thousands of dollars to efficiently self-host some of the more open ones.

For everything else it's very much a "least bad choice" type deal since you don't have control as a user. Though, it still makes me feel a bit better every time I pay extra for a product that has one less bad feature than the competition. I'm talking about planned obsolescence, user lock-in, "smart" features, and other stuff I try to purposefully avoid.

I despise too those companies that make those choices, and I would definitely vote with my wallet and favour companies that make choices that I like more, all other things being equal. But I'm not handicapping myself just for that little thing. People are different, you're free to be more adherent to your ideals, and that's not bad thing.

It's not precisely user hostility. You're talking about the difference between a hole hog and a DeWalt. ChatGPT is trying to be a DeWalt: It works while also trying to backstop unexpected bad interaction so you don't unexpectedly torque your wrist past its breaking point because you jammed the bit into the wrong piece of wood (to pull the analogy back around: You don't inadvertently get porn or gore or defamatory copy because you asked the wrong query).

Flipping the question on its head, there's going to be a category of user who's asking "Why am I paying 20 bucks a month so this thing can give me answers I can't publish?"

Not exactly. There's a "custom instructions" area in the Settings that allows you to give the bot permanent instructions that apply to all chats. So you do it once and never have to do it again.

And the new "My GPTs" things allows you to give super-lengthy and detailed instructions, and train the bot on custom works, which is even more powerful.

> I told it to he had passed away, it checked the internet

Are you sure it didn't just "believe" what you told it, the same way LLMs can be badgered into falsehoods?

it now uses bing to perform searches. so i think he is probably correct to assume that it did that. it would show up as „performing search with bing“ in the chat history.

Even before it could search online it would hallucinate checking things online. You could ask it to email you something or upload it somewhere and it'd also hallucinate thinking it did it for you. Changing its mind is always easy as long as you pretend to give it a source that refutes it.