Topic: bitcointalk forum vs OpenAI - ChatGPT - page 8. (Read 3190 times)

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

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:

In the limitations section, the blog indicates that:

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/)

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.

Hitler was a poet indeed.

After Google AlphaStar defeated StarCraft pros, maybe it will turn on players of some Terminator franchise video game next.


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.

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?

@dkbit98
I don't know if the model is being trained by input or feedback.
Hopefully feedback and openai isn't dumb enough to introduce it in a training environment, crowds bias testing it an learning from constant barrage about people asking it to make racist scrips and poetry about Hitler.
If it's training now. It'll only be more biased as it'll have tons and tons of data on bias.
It seems to keep a conversational cache , and discards. (Thankfully- for now)
I think the feedback system is to can/compartmentalize bias/race/math/whatever issues and address them before launching a model that does learn from user input.
Releasing it free to all for a min was a good idea. After it's not free to use , people won't pay to make it say racist/stupid stuff. And it'll be trained to deal with those issues approiately. My 2 sats.

Now all it needs is one of those Darpa robots, and we're fucked.

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.

We could assume that it can’t save the info itself, but a subsequent step in the pipeline can, which might not technically be considered part of the AI solution i.e. a step in the chatbot’s code can log the information, whilst the AI itself might consider itself independent from the chatbot interface (or the left hand might not know what the right hand is doing).
An interesting question could be what degree of self-awareness it thinks it has, and whether it considers itself a chatbot, an api, an algorithm, or some kind of superior entity.


Edit: ChatGPT can say one thing and the opposite: I too asked it if it kept logs on the questions, and got a "No yada-yada-yada". Requerying provided a "Yes yada-yada-yada"...



Crap, the first version of your post was promising, until you called out its bluff…

Well, this turned out to be simpler than I thought.



j/k

It likely just remembers the previous question I asked.

Edit: not only that, but it blatantly lied when I asked the same question using one of examples from the OP.



I have changed my opinion. I think ChatGPT will fit in here just fine and will be in DT1 and a merit source within weeks.

Well I asked her or him whatever and it says no she’s not saving answers or maintaining any database.



Maybe I’m overreacting but I’m obsessed with it… it reminds me of a movie “Her (2003)”.

...

I actually looked around for that today, but I didn’t find a hint along those lines on ChatGPT, and as you say, this is just one out of likely many AIs of the kind to come. It would be a real shaker though if such logs, which I’m pretty sure exist, should eventually be made available and/or leaked (massive leak though) …

There are some (alleged) AI content detection sites around. I tried a couple of the freebie ones earlier on, and they were useless, giving the same score of human generated content to 5 or 6 posts I chose from this thread, as to content derived from @Shahzadafzal’s ChatGPT snapshots (extracting the text with Google Images to feed the text). There are some sites of the kind that you need to pay for to test them out (pass), which doesn’t seem like the best idea on their behalf, were they to be of any good.

I wonder if the ChatGPT output is logged (it probably is) and if they could implement an API to verify if a certain text was generated by it. That'd be a great help and not just for Bitcointalk. Although even if that happens there would be other less-cooperative AI sites eventually so it's a losing battle either way. Spammers gonna spam if they can make money with it.

Stackoverflow has added an anti-GPT policy to their site:
See: https://stackoverflow.com/help/gpt-policy

Now perhaps it’s easier to detect the usage of GPT in the more technical type threads, than in those that are less technical in nature, and it may be the latter where it becomes more of a challenge to detect this type of usage in a post.

Conceptually, any post using GPT to create content on Bitcointalk is plagiarizing, as the poster is not creating the content himself, and is in fact trying to pass on someone else’s content as his (albeit that someone else being an AI). According to OpenAI’s Sharing and Publication Policy, using the API, and I have to assume that that extends to the results of their chatbot, requires one to explicitly indicate that the content was AI-generated. Though this latter point is not technically of our concern, it seems like a reasonable request to place, in a similar fashion to links on posts that are largely/verbose based on other sources.

Detection of GPT usage is not going to be easy for the most, and likely, over time, people can pick-up on patterns such as the usage of near perfect English, consistency in its usage throughout all posting history, and/or alternating with changes in style (human/AI), lack of real interaction from a less than academic point of view, certain types of formal constructions, and so forth. This is obviously is not exclusive to GPT, nor sufficient to deem someone a GTP-plagiarizer with certainty, and maybe likenesses is the closer one can get shorter of a confession.

Now all this is, if it becomes an extended practice, is going to be a drag, whereby people will be able to create bag loads of posts with cero effort and thought, and although likely matching quality-wise a large base of posts that we encounter per se, it may easily become a new spam-fest source of neutral content.

One can raise its intellectual capacity so much via using AI, he may even get a chance to be included in chipmixer's sig camp lol.

Imagine a retard who can't even speak English properly posting like a Crypto Professor.

He can even use AI while talking to forum people lol. That's fucked up.

I think we will see so many experts posting soon.   Imagine everybody is posting like nullius in every one of his/her posts. Wohoo

Even if we introduce such a rule on the forum, who will check whether the text belongs to a person or an AI? Where to find that line between text written by a human and text written by a machine? If you know such an algorithm, please share it. I, for one, have no idea how to do it. And also, can you imagine how many topics will be created in Meta about the complaints against the moderators, that they deleted someone's post, because they thought that this post was written by an AI

Say what? It's more like the perfect plagiarism by word spinning tool.

Now the big question is, is it against the forum rules if you post AI's response as your own. It is not ethical but who is going to prove that is a plagiarized text since the AI can produce perfectly unique articles?

That's shitposters' dream became true.

All they need to do is typing the title of the topic and copy/paste the AI's reply.

ez pz lemon squeezy

Dalle works well if you phrase right
"3D Render synthwave robot working in a cyberpunk data center" something like that , you can even use GPT to write dalle prompts.

The writing style of gpt becomes apparent the more you interact with it. I've read slander articles about gpt I'm pretty sure GPT wrote.
Phrases like,
First
Furthermore
In conclusion, and a total lack of personality and generally lacks bias unless it's a forced paradigm.
It writes proper but you can tell there is no heart and soul in it.

We have been using it for a lot of stuff , I wrote R2/Iso compliant Nist 800-88 wiper for downstream in like an hour just debugging now. 

I like the workflow GPT provides.
I've been working on a startup for a while (my 3rd one) this has streamlined the process , it helps being married to someone who does instructional design.

I've been calling GPT NZT for nerds (from movie limitless)
GPT will probably be able to make NZT if trained on chemistry/pharma industries imagine typing in symptoms and getting a full chemical calculation of a formula that can cure you. , 2023 is gonna be a crazy year