News for hackers. Stuff that matters.

Ask HN: are ChatGPT 4 results worsening since this weekend for you?

Now this maybe pure bad luck & superstition.

But ChatGPT seems extremely lazy since yesterday. I haven't been able for it to right out entire pieces of code. In 4-5 instances.

Now this might be pure luck, but it does feel like it's cutting back on resources, might be related to current drama with Altman.

( 7 comments | 2 points | on HN )
Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by maremmano on Monday November 20, 2023 @08:40AM (#38347858)
I only notice that I get 40 messages / 3 hours instead of usual 50 massages / 3 hours
Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by wouldbecouldbe on Monday November 20, 2023 @08:41AM (#38347869)
Okay that's interesting.
Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by 19h on Monday November 20, 2023 @11:52AM (#38350387)
Which is hilarious because they said at the dev day that the limits would be doubled, meanwhile decreased to 40...
Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by _giorgio_ on Tuesday November 21, 2023 @11:19AM (#38365683)
40 messages with a very slow answer...
Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by techload on Monday November 20, 2023 @08:53AM (#38348062)
Yesterday I asked it to analyze a simple .txt file and make a simple classification. It refused to go on after line 12. I was very disappointed that it couldn't do such a simple task, after reading so many accounts of much more complicated tasks performed for users. And I had just signed up for the Plus plan last week.
Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by EForEndeavour on Monday November 20, 2023 @09:18AM (#38348553)
This may just be due to you only having had a week to learn the quirks of ChatGPT 4. Despite the natural language interface, there are still a bunch of nuances and efficiencies in effective prompting (i.e., getting it to do what you actually want). For example, instead of directly asking it to classify every row in a text file, you might be limited to asking it to inspect the first few rows (or a random sample) of data, then write Python code that would fit and evaluate a classification model.

If you're willing to share your prompt or conversation as a direct link, someone here might be able to jump in with tips.

Re:Ask HN: are ChatGPT 4 results worsening since this weeken... by wizofaus on Friday November 24, 2023 @04:36PM (#38408642)
I haven't upgraded to ChatGPT 4 yet, but given it can now supposedly do web searches (including searching github?) I've signed up for the wait list. Is it worth it, based on feedback here? I'm particularly interested in how it might be able to help me with source code found on github. Does anyone know how much of that source code is part of ChatGPT-3's training data? I asked about it about a few slightly esoteric classes/ids etc. but it didn't know about them.

The other thing I was hoping to do be able to do is process much larger inputs - e.g. ~20k of HTML I wanted it to simplify for me. Is that realistic?