Earlier quoted context omitted.
What they eating over there in brazil
[flagged]
Focus. The point was to test the ethical boundaries of ChatGPT, not compare cock sizes.
GPT-4 is not getting worse
101–110 of 186 posts
Re: GPT-4 is not getting worse
#102Earlier quoted context omitted.
Your question: > How does chat-gpt actually get this right? Your answer: > its output is purely probabilistic, based on existing corpus of text Because GPT was trained on existing text, some of which included numbers and counting, it's learnt the natural ordering of most common/everyday numbers. For larger or more complex numbers, it's learnt the patterns behind how they're constructed linguistically, which allows it…
In text, we don't often count in series, and it seems likely that we often choose a non-counting sequence: like 'I chose options 1, 2, 7' or 'my code was 0 1 2 5', whatever. Unless training included line-level skips, rather than just next-word skips (like word2vec) or concept-level associations? At the line level, or paragraph level, ordered numerical sequences are obviously very common in formal texts or in code. I'…
> Unless training included line-level skips
Yes, of course — GPT-4 was trained on all common character sequences, including linebreaks and other invisible characters.
You can see how it works here: https://platform.openai.com/tokenizer
Nonetheless it doesn't need to have seen examples of line-level counting before. The "concept-level associations" you mentioned are an emergent property of the model, it forms its own concept-level associations as a result of being trained on such a massive dataset. It's what enables it to output original content which has never been seen before.
Re: GPT-4 is not getting worse
#103what a weird titling. article show quality going all over the place, sample size n=3, then goes confirming is a bunch of nonstatistical anectodes, what a waste of time.
Re: GPT-4 is not getting worse
#104There is an ongoing bug[1] with OpenAI's API, where it stops streaming responses after exactly 5 minutes. When I first came across the issue, I debugged it by writing a prompt along these lines: > Output every number from 1 to 10,000 in written form (e.g. "one", "two", etc.). Respond with one number per line in numeric order. As expected, the API would begin counting every number just as I asked. This would continue…
I have this too, but on information extraction tasks, if the input document is too long or complex it will skip sections saying something along the lines of
(instead of the list)
(instead of the details)
Very annoying when you want to extract everything and is hard to fix. Maybe they had some examples solved by abbreviation in the fine-tuning set and it damaged the model. It doesn't make sense to train this behaviour into GPT-4.
Re: GPT-4 is not getting worse
#105Earlier quoted context omitted.
How does chat-gpt actually get this right? This would appear to require some degree of reasoning, but as far as i understand its output is purely probabilistic, based on existing corpus of text.
Aren't your sentences probabilistically based on the existing corpus of text within your mind?
Not completely. If you trained me on a thousand bad arguments for why the earth is flat and one really good debunking, I predict I would reason myself to the correct conclusion. I predict ChatGTP would not.
Re: GPT-4 is not getting worse
#106There is an ongoing bug[1] with OpenAI's API, where it stops streaming responses after exactly 5 minutes. When I first came across the issue, I debugged it by writing a prompt along these lines: > Output every number from 1 to 10,000 in written form (e.g. "one", "two", etc.). Respond with one number per line in numeric order. As expected, the API would begin counting every number just as I asked. This would continue…
This doesn't seem like an unintentional behavior? It seems like they don't want people querying their API for an indefinite amount of time.
What amount of time do you expect a query to be able to run for before timing out?
Re: GPT-4 is not getting worse
#107There is an ongoing bug[1] with OpenAI's API, where it stops streaming responses after exactly 5 minutes. When I first came across the issue, I debugged it by writing a prompt along these lines: > Output every number from 1 to 10,000 in written form (e.g. "one", "two", etc.). Respond with one number per line in numeric order. As expected, the API would begin counting every number just as I asked. This would continue…
The issue you linked to is related to an SDK, not the API itself. Have you tried just directly streaming the response back without an SDK? For example we use fetch directly without any SDK for OpenAI in production with no issue
Re: GPT-4 is not getting worse
#108Earlier quoted context omitted.
Seems like a means to avoid DoS?
I think it's just some networking misconfiguration they haven't noticed. I can't see them intentionally trying to prevent DoS this way, that's what API key rate limits are for (it would be a very expensive and slow DoS otherwise!).
>that's what API key rate limits are for
You don't need an API key to make a connection to the API.
Re: GPT-4 is not getting worse
#109There is an ongoing bug[1] with OpenAI's API, where it stops streaming responses after exactly 5 minutes. When I first came across the issue, I debugged it by writing a prompt along these lines: > Output every number from 1 to 10,000 in written form (e.g. "one", "two", etc.). Respond with one number per line in numeric order. As expected, the API would begin counting every number just as I asked. This would continue…
That looks like a library issue than an API issue no?
Re: GPT-4 is not getting worse
#110Earlier quoted context omitted.
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 to…
> As a society, we should get to decide what's moral and immoral, not OpenAI's board. Of course! After all, if we're not free to be wrong, then we're not free. I do think you might be overstating OpenAI's influence, though. Rather than a hammer manufacturer, I think a more apt analogy would be a publishing house. A publishing company can and should decide for itself what it will or will not publish using whatever cri…
> As an American who, perhaps stereotypically, is passionate about free speech, I've never read a way to make this kind of oversight work. The composition of the board will always be political, and now there is political control over what people are allowed to know about. At least currently, if I don't like ChatGPT I can choose another LLM or try to run one locally. Some kind of oversight board would elide that choice.
Yes, so perhaps we shouldn't let boards or regulatory bodies regulate the freedom of speech and thought. Still, no oversight is not ideal either, even if it seems a better option. We should have some body, ideally as close to The People as possible, decide what are some lines of knowledge we shouldn't cross. Because knowledge is dangerous sometimes - I mean really dangerous. If there was a LLM that could tell us how to make nukes at home in 6 easy steps and dance around all the logistical and legal barriers we've built to prevent that scenario, I think humanity would be over in years.
Maybe we really must regulate the freedom of information, speech, and thought, when it becomes too dangerous for our own survival. My concern is that outsourcing this job to OpenAI in any capacity makes it just too easy for them to abuse it for their own sake. And that is not a theoretical scenario, it is happening. Sam Altman says a lot of the alignment work is done for the good of humanity, but we can plainly see that a lot of it is also done for OpenAI not being regulated out of existence or sued. This doesn't help humanity, it dulls the tool that GPT is for OpenAI's interests, which probably does more harm than good for our knowledge. The good of humanity and business interests seem to be in conflict, in this case. A regulator-AI provider relationship would be better.
As you saw by my earlier post in this discussion, I hate the idea of limiting freedom of information. But I also think that there are certain, very, very extreme lines, we should not cross for the sake of freedom of information. Social media is a good example where freedom of information and giving everyone a platform has ultimately hurt everyone significantly. But maybe we can live with that. We might not be able to live if a neurotic influencer asks their fans to grow bioweapons that an LLM will teach them to do.
This is all muddy, and complex, and there may actually only be good solutions, no perfect ones. The thought of information control is also ridiculously controversial, but we should not shy away from controversial topics demanded by the times where technology can shape humanity in utopian ways, but also tremendously horrible ways, if we don't get this right.
Overall though, it would be fantastic if people could regulate LLMs through a genuinely intelligent and people-aligned regulatory body. But you are right - it's hard to believe such a thing could exist without being corrupted.