Hacker News Reader

Comment by phillipcarter | Hacker News Reader

I don't think it's any worse at all. I think what most people are expressing here is reaching the limits of the technology and realizing that it's not magic.

View on HN

No, OpenAI 100% pushed an update recently that is very noticeable where they basically "nerfed" the responses. I'm sure it was a business decision (either costing them too much to give away the kitchen sink for free or they want to turn around and charge more for the "really good, old" version of GPT-4 ) but you can visually see where the bot used to try to answer complex tasks, now it has an extra layer where it says "that is too complex to discuss here" (aka it is being trained to not engage by default, whereas that used to not be the case)

It's likely that this is the result of training it to avoid bullshitting. It gave confident garbage, and they're trying to stem the flow. This likely leads to less certain responses in complex tasks.

Here is the crux. When asking about non-code stuff, it would confidently lie and that is bad. When asking about code, who cares if the code doesn't work on the first go? You can keep asking to fix it by feeding error messages and it will get there or close enough.

It's obvious what is happening. ChatGPT is going the non-code route and Copilot will go the code route. Microsoft will be able to double charge users.

And you didn't enable plugins?

I guess you're entitled to an opinion. I've had the same (error-prone, often unhelpful) experience since gpt-4 was released. It's a little faster now which is nice.

>OpenAI 100% pushed an update recently that is very noticeable where they basically "nerfed" the responses

>they want to turn around and charge more for the "really good, old" version of GPT-

If this was the case it is provable by submitting queries to `gpt-4-0314` and comparing them to `gpt-4`.

To me, it doesn't feel like the gpt-4 in the playground (and in the API) is the same as the model used in chatgpt. It's hard to prove though since both are nondeterministic.

My perception is that the ChatGPT website is a particular tuning of `gpt-4` with a custom assistant prompt.

I think you're wrong. I've noticed no change.

Exactly this. The magic has worn off, and simultaneously, we're coming to terms with the reality that no model can exist in isolation. So changes made, like getting it to not create false rumors about people, affect the rest of the system.

I suspect it's the same people complaining how Google's gotten worse.

One time, I went to the hotel restaurant. I ordered something; it was amazing. Figured I'd do it again a few days later. It wasn't bad, but it wasn't as good. There wasn't any specific difference I could put my finger on other than that the novelty wore off.

No, GPT4 was originally a very kurtz asshole. It was hilarious and I shared it with everyone. Now it’s incredibly polite and explains things thoroughly. But it’s a little dumber.

Google has gotten worse. So what you're saying is people complaining about GPT-4 are correct?

Reply on news.ycombinator.com