bitcointalk forum vs OpenAI - ChatGPT

[LoyceV]

Try writing a research paper on Richard Ramirez.
Then cite it. It's an interesting rabbit hole.  You'll learn that thousands of scholarly papers have been wrote about him.
And it's virtually impossible to say anything about him that hasn't been said before without a paper turning up %60+ plagiarized words. So in a way arnt we all doing the same thing?

We're not all doing the same thing, only a few (thousand) people waste their time on this guy (I've never heard of him). If there are thousands of research papers already, people should stop writing more of the same! That's kinda a thing with scientific papers in general: many of them refer to other papers, instead of doing actual research and adding something new.

poetry about Hitler

Hitler was a poet indeed.

[1714629989]

1714629989

[1714629989]

1714629989

[DdmrDdmr]

<...>
Seriously, this stuff is free only because people who are using it are feeding it information, training it and they are making it look good better than it actually is.<...>

In a blog on openAI’s website, they state that  the initial model followed a supervised learning approach, with human people training the model, claiming that:

ChatGPT is fine-tuned from a model in the GPT-3.5 series, which finished training in early 2022.

In the limitations section, the blog indicates that:

ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Fixing this issue is challenging, as: (1) during RL training, there’s currently no source of truth;<…>

From the above, I infer that it is not performing real life language related training, though the model will certainly be fine-tuned iteratively over time. I would however assume that it is gathering data to improve. Likely not on-the-fly, based on the demands and answers it provides (I cannot conceive it doesn’t log them for this purpose).

By the way, insofar, the results from ChatGPT seem to be way better than those generated by Meta’s Galactica (aimed at providing scientific results), which recently pulled the plug (at least for now) after just three days of spewing incorrect information (see also https://galactica.org/)

[fillippone]

It will be interesting to see the quality of AI posting relative to shitposters after manual filtering is done. Or maybe people will just give it more merits than ChartBuddy.

Maybe that has already been done.
I have probably merited that bot in my Merit giving thread.

Only thing is keeping me away from this certainty is that at the moment the AI seems at full capacity and refusing new interactions:




For sake of records, I have been discussing this in another thread:

AI writing messages on Bitcointalk.org

[libert19]

If it's get too widespread, it will lose it's meaning of genuine looking writing, and then there will be tools to counter it.

Anyway imo, no matter how good ai may get, it can never replicate human emotions. At least I don't think so.

[Hispo]

...
Anyway imo, no matter how good ai may get, it can never replicate human emotions. At least I don't think so.

You are right, Ai can't replicate human emotions but I am afraid that when we all participate on places of the internet like this forum (using only written words) we depend much on the same words, the context and emoticons to guess the emotions others try to express, and unfortunately and artificial intelligence could eventually master both context and emoticons not long in the future.

Take this example:

"I completely support Bitcoin  "
"I completely support Bitcoin!!!!!!!  "
"I completely support Bitcoin...  "
"I completely support Bitcoin ~"

[shahzadafzal]

I am still little confuse about this AI bot.  When you say that this bot will combine the answers with different sources and will come up with the best possible answer, does it mean it will copy the exact same lines from different sources and produce plagiarised content ?

There are other intelligent bots available in the market (Good ones are paid), which will give you plagiarism free articles for your sites etc. So i wonder if the OpenAI - ChatGPT free software is comparable to those paid ones  

That would be the wrong interpretation to say "bot will combine the answers", what I mean is it interprets and spits it out in its own wording. In other words it has learnt based on all the content that has been fed and now it can "understand" and answer based on his knowledge. As far as plagiarism is concerned it can give proper citation like I gave one in this example.

Regarding any other intelligent bots to be honest I haven't seen anyone, specially this much sophisticated. There might be few those are limited to chat only but ChatGPT is on another level. You ask it to write program in for example it write program in JavaScript to animate your photos. Other day I used to it towrite a RegEx code to check a specific Card Number pattern. So I don't think there are many free AI bots like ChatGPT.

Also like many other pointed out when it's free you are the product, definitely this "free" is part of their PR stunt + more training material for their AI.

Maybe even a moderator and could replace theymos in future!
Seriously, this stuff is free only because people who are using it are feeding it information, training it and they are making it look good better than it actually is.
Some people even suggested that all the g-captchas we are clicking all the time is also doing that, teaching some dumb program to repeat stuff based on our actions.

That is quit possible, if you know about google reCAPTCHA and how "How we all helped(unknowingly) Google to digitize books".  It's interesting to know how they did it https://www.youtube.com/watch?v=PQ-xzwj_p_4

Based on this we clearly don't know what this g-captchas is working on or if we are helping or training their dumb bots yet again.

Here's an interesting experiment:

- open a copper member bot account here, put a "NO MERITS" signature so nobody merits it
- the bot will randomly pick a topic from the topics with activity in the last 24 hours (filtering local boards out)
- The bot will take the last three posts from the thread, or the posts that were made over the last 24 hours, whichever is shorter
- After filtering out the BBCode and emoticons, it will feed this into chatGPT (this may be a problem because it doesn't have an API).
- Manually screen the output so that you're not autoposting a flame or gibberish
- Bot posts reply
- Repeat once per day

It will be interesting to see the quality of AI posting relative to shitposters after manual filtering is done. Or maybe people will just give it more merits than ChartBuddy.

I am 100% sure at this point that the quality of ChatGPT posts will be much much better then any shitposters here on bitcointalk.

Also AI alone won't be of any use without human intervention. In the end it should be us who will decide, this ChatGPT is certainly going to help students and researchers but they have to make sure they read and understand the ChatGPT output first before they present or submit it on their behalf.

But yeah for shitposters they don't care to read, they like copy paste

(this may be a problem because it doesn't have an API).

It definitely have APIs but not for public yet, on it's wikipedia page it says

Microsoft announced on September 22, 2020, that it had licensed "exclusive" use of GPT-3; others can still use the public API to receive output, but only Microsoft has access to GPT-3's underlying model.[7]

You are right, Ai can't replicate human emotions but I am afraid that when we all participate on places of the internet like this forum (using only written words) we depend much on the same words, the context and emoticons to guess the emotions others try to express, and unfortunately and artificial intelligence could eventually master both context and emoticons not long in the future.

Take this example:

"I completely support Bitcoin  "
"I completely support Bitcoin!!!!!!!  "
"I completely support Bitcoin...  "
"I completely support Bitcoin ~"

Trust me it can...

[drwhobox]

If it's get too widespread, it will lose it's meaning of genuine looking writing, and then there will be tools to counter it.

Anyway imo, no matter how good ai may get, it can never replicate human emotions. At least I don't think so.

Yes, it will not replicate human emotions, but it will do the job for humans. But I don't think this AI will replace anything yet.
It is still in development, but I am afraid this AI will learn from humans.

[LoyceV]

And here's the first one (as far as I've seen):

I admit I first chatted with ChatGPT about this before coming here
It said: ~irrelevant~
Is this true?

That website spits out millions of texts per day, if people start asking forums worldwide if it's true, that's a tremendous amount of spam already.  Go bug whoever created that "AI", ask them if it's correct.

[shahzadafzal]

That explanation though!!!

[WhyFhy]

That explanation though!!!

Hilarious! I avoid asking it anything that requires logic as its has none being a MLLM.
Sure, feels legit sometimes though.
The math it does is loosely based on memory,
ask it what 123x456 is under primed constraints,
It will say something like 56,888 or 56,088 or 58,066 instead of 56,088
The answer is usually visually close or just totally inaccurate.
It may get it right first try here and there but it points out quickly that its logic is memory based and not actually processed.

Here is an example of the logic without numbers.

Q.how can i determine my sisters age based on my date of birth.

A.To determine your sister's age based on your date of birth, you will need to know the exact date of your sister's birth. Once you have this information, you can use a calendar or a calculator to find the difference in the number of years between the two dates.

For example, if your date of birth is January 1, 2000 and your sister's date of birth is January 1, 2005, then your sister is 5 years old.

If you do not know your sister's exact date of birth, it may be difficult to determine her age accurately. It is always important to respect people's privacy and to ask for their permission before sharing personal information about them.

Q. What year is this example based on?
A. The example I provided was based on a hypothetical scenario and does not reflect the current year. The dates and ages mentioned in the example are purely for illustration purposes and are not intended to represent any real person or event.

To determine your sister's age based on your date of birth, you will need to know the exact date of your sister's birth and the current date. Once you have this information, you can use a calendar or a calculator to find the difference in the number of years between the two dates. This will give you your sister's age as of the current date.

Like what? Terrible
One of my friends said this and after poking around he's right.

"neural networks are inherently bullshit
it's why humans are so dumb"

[NotATether]

Hilarious! I avoid asking it anything that requires logic as its has none being a MLLM.
Sure, feels legit sometimes though.

Actually, I have successfully gotten ChatGPT to write code for me like a password hashing algorithm with salt, and generate instructions for making an OpenSSL certificate.

I even managed to get it to document my code.

So I had this 500-line file with about 30 C++ functions in them, and it spat out Doxygen for every single one!

The only limitation is that it will stop printing stuff after a character limit so I either command it something like "finish what you were writing" or, if that just makes it repeat stuff again, send my functions to it in pieces.

What would normally take me the entire day and a load of coffee only took me 1 hour with an AI. I plan to document about 100 more (medium length) functions across 10+ files with it as well.

Seriously, if they're going to sell this thing for $200/year, it's definitely worth it for me (as long as they remove the character limit because it's a time sink).

[WhyFhy]

Hilarious! I avoid asking it anything that requires logic as its has none being a MLLM.
Sure, feels legit sometimes though.

Actually, I have successfully gotten ChatGPT to write code for me like a password hashing algorithm with salt, and generate instructions for making an OpenSSL certificate.

I even managed to get it to document my code.

So I had this 500-line file with about 30 C++ functions in them, and it spat out Doxygen for every single one!

The only limitation is that it will stop printing stuff after a character limit so I either command it something like "finish what you were writing" or, if that just makes it repeat stuff again, send my functions to it in pieces.

What would normally take me the entire day and a load of coffee only took me 1 hour with an AI. I plan to document about 100 more (medium length) functions across 10+ files with it as well.

Seriously, if they're going to sell this thing for $200/year, it's definitely worth it for me (as long as they remove the character limit because it's a time sink).

I mean they are getting my money too, just not yet I kind of expect it to be cut off tomorrow.
The MLLM with doxygen is indeed amazing

Code:

        if wipe_method == "quick":
            subprocess.run(["shred", "-n", "1", "-v", d])
        elif wipe_method == "7-pass":
            subprocess.run(["shred", "-n", "7", "-v", d])
        elif wipe_method == "nist 800-88":
            subprocess.run(["shred", "-n", "1", "-x", "-v", d])
--------------------------------------------------------------
        if wipe_method == "quick-Nist-800-80-1-pass-clean":
            subprocess.run(["shred", "-n", "1", "-z", "-v", "-p", "-w", d])
        elif wipe_method == "Nist-800-88-3-pass":
            subprocess.run(["shred", "-n", "3", "-z", "-v", "-p", "-w", d])
        elif wipe_method == "Nist-800-88-7-pass":
            subprocess.run(["shred", "-n", "7", "-z", "-v", "-p", "-w", d])
        elif wipe_method == "Nist-800-88-35-pass-Nuke":
            subprocess.run(["shred", "-n", "35", "-z", "-v", "-p", "-w", d])

This is incomplete but gives you an idea of much you can elaborate on it.
by simply asking "what else could be added"
To me it ungated a lot of information, sure beats playing wack a mole on google Keywords/AdWords.
Now, I get a solution, I check it and implement it , if it doesnt work i poke why , and adjust from there instead of toiling in obscurity on google.
If I can't find out why through gpt, odds are its given me enough value to figure out the rest.
even this stoichiometric approach is much faster than anything I've ever tried in the past. making it the most valuable employee in the arsenal in my opinion.
It's a workflow consultant without even trying to be.
Just throwing it out there https://github.com/doxygen/doxygen is free, however the way GPT packs and wraps it up can't really be beaten.

[Mr.right85]

That explanation though!!!

So the AI is not that smart after all. It remains a garbage in and garbage out. The algorithms formaking these calculations weren't just fed to it. Although, he first instance of coming up with comprehensions seems uncommon but yet on point, it still remains that, this AI with the help of Google, seeks out the answers frm that which has already been written online and that includes the forum since, its online as well.

Non of that doesn't add some originality to the output produced by the AI.

Perhaps, it ought to have been smart enough to give us age 35 for our sister, AI.

[Synchronice]

Even this mf knows that Craig Wright isn't a real Satoshi

[FatFork]

That explanation though!!!

It's worth noting that even humans can struggle with math at times.

This is what I got after a few consecutive tries:


It even made me laugh with the note at the end.

[UserU]

Even this mf knows that Craig Wright isn't a real Satoshi

CW be like; damn Satoshi must be somewhat hiding behind this mf

[n0nce]

I wanted to try this, as well, so I 'fowarded' this question to ChatGPT. I think if I was OP, I'd be satisfied with the answer.

Both Cuckoo filters and Bloom filters have a time complexity of O(1) for most operations, which means that their running time is independent of the input size. This makes them both very fast data structures.
[...]
Overall, both Cuckoo filters and Bloom filters are very fast data structures with a time complexity of O(1) for most operations. However, Cuckoo filters are slightly faster than Bloom filters for lookups, and they also support deletion of items, which Bloom filters do not.

what do you think? what is your opinion?

[Ultegra134]

Alright, a few days ago I claimed in another thread that signature campaign managers must be on the lookout for users who might be using such services, but it might not be an immediate threat to the forum since most users who'd use such a service would perhaps have a hard time entering and maintaining their status in a signature campaign. However, after fiddling with ChatGPT a bit more, I'm starting to change my mind. As the OP already demonstrated, ChatGPT is capable of creating posts out of thin air. On top of that, when I messed around from my phone a few days ago, it would recreate the same answer, but now that I'm using my laptop, it isn't; was my phone, the website's bug to blame? No idea. I've literally generated three to four posts within a few minutes. The posts are completely irrelevant, but I'm guessing that with a few parameters, you could adjust them for numerous threads.

What if more and more users resort to their usage and result into huge non-enganging discussions where everyone is using AI to post? If such people are unable to enter signature campaigns, then it'll probably discourage them enough to not resort to such services.

I'll also quote a few of the generated responses I got.

[n0nce]

~

To be honest, I don't feel like it's very good at creating interesting, new topics. However, it can answer questions pretty well. Although sometimes it gets details wrong, which are often the most important part, especially in the technical section.

So, I guess it could be a threat, if people start using it to reply to questions without fact-checking the reply, and the content quality on the technical board starts to drop and get cluttered with misinformation. On the other hand, the merit system and simply users' reputation should regulate this sufficiently (e.g. red tags for misinformation or getting put on ignore).

I guess if users are knowledgeable about a topic, but lazy to write a reply, it could be handy to use ChatGPT to generate a reply and quickly read and fact-check it.

[bittraffic]

~

To be honest, I don't feel like it's very good at creating interesting, new topics. However, it can answer questions pretty well. Although sometimes it gets details wrong, which are often the most important part, especially in the technical section.

So, I guess it could be a threat, if people start using it to reply to questions without fact-checking the reply, and the content quality on the technical board starts to drop and get cluttered with misinformation. On the other hand, the merit system and simply users' reputation should regulate this sufficiently (e.g. red tags for misinformation or getting put on ignore).

I guess if users are knowledgeable about a topic, but lazy to write a reply, it could be handy to use ChatGPT to generate a reply and quickly read and fact-check it.

That's what they might just do though. The AI seems to have answered differently when I asked the same question asked by users here. When I asked it twice, it answered the same starting with.

As I mentioned before, ...

I just tried OpenAI yesterday. It was very impressive still. It could write an article to answer just one question as long as you specify it in your question.