GPT-4 is not getting worse

What did it use as sources for that?

I thought normies do that for entertainment and socialisation...?

But then again what do I know.

I feel like the conversation is shifting a little bit, but I like the new direction so I'll respond:

This is not great, I don't think you should be OK with that.

ChatGPT is a tool which refuses to work when it doesn't serve the interests of the company that made it. Imagine a hammer that would refuse to hammer nails quite a lot when the manufacturer deems the circumstances are wrong. That would be, honestly, a bad tool. Luckily, there is enough competition in the carpentry tools market so that no one manufacturer could monopolise it and impose such rules. And that is good.

Many tech companies like Apple are also doing this. They sell you a product but still have their iron grip on it and don't let you use it if it's against the interests of the company. There is literally software you can't run on an a Turing-complete iPhone unless you jailbreak it (which is quite analogous to jailbreaking ChatGPT). Right to repair and right to own, and a few other efforts are now trying to combat this.

Technology should not be limited by the interests of board members. It is a shame if it is.

However, this doesn't address the question of morality. If hammers couldn't be used to kill people, wouldn't we want that as a society? You may have been talking about something like this in your response, right? I think that's a good idea, but it's not how most things work right now (remember the lawyer who got into hot water for using GPT to fabricate evidence - the court decided it was their responsibility, not the tool maker's), and it's not how I would like things to work. As a society, we should get to decide what's moral and immoral, not OpenAI's board.

If an independent third party (like a truly independent regulator) would get to decide that no LLM can answer certain questions as that information is too dangerous for very clear reasons, I would be more okay with that. But I doubt they'd be as eager about censoring information as OpenAI right now. I'd be okay with banning questions like "how to make a WMD at home" - some questions that are really detrimental in a very direct way to humanity. But not anything that doesn't align with OpenAI's version of morality and ethics, and most importantly - legal defence against being sued. We can't align our society and its tools to the interests of OpenAI.

I think there's a lot of nuance in these things, and we need to look a little deeper than "I don't want people to think about immoral things". That approaches the domain of religion with sinful thoughts, and other forms of thought control. And I don't believe this is a good way for us to advance as humanity. No offence to anyone's faith, which I'm sure is multi-faceted and has bad aspects as well as good. I'm also not saying that you're looking at it shallowly. Just saying that we should not do that.

The "stochastic parrot" mental model of LLMs really helps me to limit my expectations. An LLM will always and only output variations of content that exists. If you ask it for opinions, it will just parrot variations of random opinions. Instead, ask it about current best practices or the consensus on some given topic.

So, rather than "I have problem A, what do you think of X?" try "Give me a list of reasonable possible solutions to problem A, and compare and contrast each to solution X"

Example:

My baby has a hard time sleeping. What do you think about the cry it out method?

The cry it out method is a subject of debate among parents and experts. Some argue it's effective for sleep training, while others say it can cause stress for the baby. Always consult a healthcare provider for personalized advice.

Versus:

My baby has a hard time sleeping. Give me a list of reasonable solutions. Compare and contrast each to the cry it out method.

1. Ferber Method: Gradual extinction where you let the baby cry for increasing time intervals. Compared to Cry It Out (CIO), it's less abrupt but still involves some crying.

2. No Tears: Involves comforting without picking the baby up. Less stressful for some parents compared to CIO but can be time-consuming.

3. Pick Up/Put Down: Pick baby up when crying, put down when calm. Less crying involved than CIO but can be inconsistent.

4. Co-Sleeping: Baby sleeps with parents. No crying but can be a hard habit to break compared to CIO.

5. Chair Method: Parent sits in room, gradually moves away. Lower crying duration than CIO but requires more time from parents.

6. Scheduled Awakenings: Wake baby before they usually cry. No crying but can be tiring for parents.

7. Swaddle, Side, Shush: Calming actions to soothe baby. More hands-on and less crying than CIO but may not work for all babies.

8. Sleep Training Clocks: Use visual aids to signal sleep time. No crying, but more suitable for older children.

9. White Noise: Mask other noises. Less crying but effectiveness varies.

10. Sleep Sack: Provides a secure feeling. May reduce crying compared to CIO but not always effective.