AI writing messages on Bitcointalk.org

CoinEraser

legendary

Activity: 1988

Merit: 1768

Quote from: LoyceV on March 15, 2023, 09:09:04 AM

Quote from: CoinEraser on March 15, 2023, 08:14:26 AM

the changes from the previous ChatGPT 3.5 are very minor, according to Sam Altman, the CEO of OpenAI.

As long as the CEO is human, I'm not impressed.

Hopefully it stays that way. Cheesy

Quote from: Zilon on March 15, 2023, 12:52:17 PM

Sam Altman will keep bringing in newer versions within shorter intervals because information is never static and i doubt if there will ever be a flawless openAI. Deep learning/transfer learning is yet to out grow the phase of negative transfer learning on the models been feed on. A mirror change is what we keep getting -snip-

That's quite possible. I haven't been all that interested in this topic until now, but since it got here in the forum I've been keeping an eye on it.
I'm curious when the next version will be released and how long it will take. Sad

Quote from: fillippone on March 15, 2023, 07:06:17 PM

I am sure he's trying to downplay the relevance of this upgrade.

That could be. Since I don't use ChatGPT, it's hard for me to rate it.

Quote from: fillippone on March 15, 2023, 07:06:17 PM

There are people already building businesses on this new version.

Thank you for your hint. That was really quick. Shocked

fillippone

legendary

Activity: 2380

Merit: 17063

Fully fledged Merit Cycler - Golden Feather 22-23

Quote from: CoinEraser on March 15, 2023, 08:14:26 AM

Either way, the changes from the previous ChatGPT 3.5 are very minor, according to Sam Altman, the CEO of OpenAI.

I am sure he's trying to downplay the relevance of this upgrade.

There are people already building businesses on this new version.

Zilon

sr. member

Activity: 966

Merit: 421

Bitcoindata.science

Quote from: CoinEraser on March 15, 2023, 08:14:26 AM

....

Sam Altman will keep bringing in newer versions within shorter intervals because information is never static and i doubt if there will ever be a flawless openAI. Deep learning/transfer learning is yet to out grow the phase of negative transfer learning on the models been feed on. A mirror change is what we keep getting

If there is any industry the ChatGPT-4 will long struggle with is the blockchain because protocols keep changing as well as newer invetories and use case. Building models on everyday experience can have a long time positive impact but on a consistently improving network it might be a lifetime of discoveries

LoyceV

legendary

Activity: 3290

Merit: 16489

Thick-Skinned Gang Leader and Golden Feather 2021

Quote from: CoinEraser on March 15, 2023, 08:14:26 AM

the changes from the previous ChatGPT 3.5 are very minor, according to Sam Altman, the CEO of OpenAI.

As long as the CEO is human, I'm not impressed.

CoinEraser

legendary

Activity: 1988

Merit: 1768

The new ChatGPT-4 was released yesterday and is available through the GPT-4 Plus paid plan. ChatGPT is said to have been improved and can now also present images to the text. Nevertheless, the improvements are probably not as good as hoped (which I think is a good thing). Either way, the changes from the previous ChatGPT 3.5 are very minor, according to Sam Altman, the CEO of OpenAI.

Source: https://twitter.com/sama/status/1635687853324902401

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: Volgastallion on February 08, 2023, 05:45:21 PM

Sorry to re bump this thread but i see today some people... how can i say they dont think or for them doesnt matter the health of the forum.

I say this because in one post of the beginers sections, i clearly detected a IA account and im not an specialist in this, but the sad part its after some more usser found and claim this (and really wise users) some other keeps coming and talking with this IA user.... come on man at least read the thread.

With GPT4 i dont know what to expect because can be a lot of hype or this can go beyond us imagination.

BPIP should include an feature to identify IA's. Maybe modify the extension to put a badge under their name that says "Suspected bot account".

Quote from: fillippone on March 13, 2023, 07:28:29 PM

We will soon discover it as GatGPT4 is expected to be launched next week, apparently:

Yeah but it will take months before the services that make use of GPT4 are generally available to the public, including those form OpenAI itself.

fillippone

legendary

Activity: 2380

Merit: 17063

Fully fledged Merit Cycler - Golden Feather 22-23

Quote from: Volgastallion on February 08, 2023, 05:45:21 PM

Sorry to re bump this thread but i see today some people... how can i say they dont think or for them doesnt matter the health of the forum.

I say this because in one post of the beginers sections, i clearly detected a IA account and im not an specialist in this, but the sad part its after some more usser found and claim this (and really wise users) some other keeps coming and talking with this IA user.... come on man at least read the thread.

With GPT4 i dont know what to expect because can be a lot of hype or this can go beyond us imagination.

We will soon discover it as GatGPT4 is expected to be launched next week, apparently:

Quote

Microsoft Germany's Chief Technology Officer, Andreas Braun, said they will introduce GPT-4 next week. Braun gave the information at an event in Germany last week. "We will introduce GPT-4 next week, there we will have multi-modal models that will offer completely different possibilities – for example, videos," Braun said in an interview with Heise. He also added that LLM is a game-changer because it lets machines understand concepts statistically that were previously read and understood by humans only.

GPT-4 release date and more: Everything we know

I guess that even if the improvement will be outside the language models, the interpretation of input will be orders of magnitude better than j it currently is.
We will soon be replaced by _{yet another} bitcoin omniscient AI.

Volgastallion

sr. member

Activity: 630

Merit: 314

CONTEST ORGANIZER

Sorry to re bump this thread but i see today some people... how can i say they dont think or for them doesnt matter the health of the forum.

I say this because in one post of the beginers sections, i clearly detected a IA account and im not an specialist in this, but the sad part its after some more usser found and claim this (and really wise users) some other keeps coming and talking with this IA user.... come on man at least read the thread.

With GPT4 i dont know what to expect because can be a lot of hype or this can go beyond us imagination.

franckuestein

legendary

Activity: 1960

Merit: 1130

Truth will out!

Quote from: fillippone on July 19, 2020, 03:57:11 AM

-snip-

I think it is something needs to be addressed.

+1.

If this is a forum where legitimate users are supposed to express their opinion, why should we be reading what someone has requested from an AI and then copied and pasted here?

What I like about Bitcointalk is reading people's opinions. Not what the AI returns to them in order to increase their post count, merit or whatever. Don't be fooled, you already know the main reason why someone would consider doing it...

An interesting read from the meta section of stackoverflow where they managed the same problem and is full of opinions on this topic: https://meta.stackoverflow.com/questions/421831/temporary-policy-chatgpt-is-banned

Welsh

staff

Activity: 3304

Merit: 4115

Quote from: NotATether on January 13, 2023, 06:33:30 AM

I'm pretty sure GPT4 will not be free and will be available only through a subscription.

Meanwhile I am still waiting for OpenAI to get back to me regarding my spot on the ChatGPT professional waitlist.

Which will limit the problems that everyone's having with the technology. The thing is, something like this when used appropriately can help in several fields. It's the abuse that comes with that, which is unfortunately taking away from the limelight of genuine usages. For example, it's pretty incredible to find documentation on certain snippets of code, much faster than using traditional search engines.

The subscription will be one roadblock to misuse, but also the tools coming out to verify it. ChatGPT has already talked about implemented some sort of identifiable system, which is largely the case now it just needs to be fine tuned, and verifiable via their system rather than third parties. Honestly, I don't think we'll have a huge problem going forward once these things come into place.

Lucasgabd

legendary

Activity: 2506

Merit: 1113

There's no need to be upset

wow, GPT4 will be crazy to see

for those interested in checking more information about GPTZero verification tool there's a good viral thread on twitter
here:
https://twitter.com/edward_the6/status/1610067688449007618?s=27&t=dP7KaJjQT-vSn1y2jz_32A

fillippone

legendary

Activity: 2380

Merit: 17063

Fully fledged Merit Cycler - Golden Feather 22-23

Quote from: WatChe on January 09, 2023, 03:29:49 AM

There are countermeasure now coming and the first one which I saw is GPTZero, an app that detects AI-written essays. Many students are using ChatGPT for doing there college assignments and this is negatively impacting there ability to think on there own. Once you pass your school/college then one can use such tools in professional life.

Image source

I saw on twitter a lot of examples where the AI can produce text that are perfectly detected as human. This is done is several ways: adding smileys after every word, adding slight typos or double commas.
Of course it’s a mice and rat race in these kind of techniques.

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: fillippone on January 12, 2023, 01:46:58 AM

Chat GPT3 is only the beginning: OpenAI is working hard to improve their product:

~

I'm pretty sure GPT4 will not be free and will be available only through a subscription.

Meanwhile I am still waiting for OpenAI to get back to me regarding my spot on the ChatGPT professional waitlist.

fillippone

legendary

Activity: 2380

Merit: 17063

Fully fledged Merit Cycler - Golden Feather 22-23

Chat GPT3 is only the beginning: OpenAI is working hard to improve their product:

Quote

GPT-4 is going to launch soon.
And it will make ChatGPT look like a toy...

→ GPT-3 has 175 billion parameters
→ GPT-4 has 100 trillion parameters

I think we're gonna see something absolutely mindblowing this time!

It's the most powerful, cutting-edge technology *in the world*, available through a Low-Code solution!

If you're not already planning on starting an AI-based SaaS or thinking about how to build AI into your current solution...

From LinkedIn (not providing link because of opec).

WatChe

hero member

Activity: 1120

Merit: 571

20BET - Premium Casino & Sportsbook

There are countermeasure now coming and the first one which I saw is GPTZero, an app that detects AI-written essays. Many students are using ChatGPT for doing there college assignments and this is negatively impacting there ability to think on there own. Once you pass your school/college then one can use such tools in professional life.

Image source

Lucasgabd

legendary

Activity: 2506

Merit: 1113

There's no need to be upset

Quote from: Ultegra134 on January 05, 2023, 01:27:08 PM

Quote from: DdmrDdmr on January 05, 2023, 12:05:37 PM

Quote from: krishnaverma on January 05, 2023, 11:02:40 AM

There are tools to detect this. You can find them by simple Google search. <…>

There are, but at least the ones I’ve tried out ,that allow you to perform a free trial online, have seemingly very poor results detecting GPT-3 based Chat GPT text, and that’s just feeding a direct output as input to the detector, with no spinning or customization at all in between to even try to throw them off their tracks. There are a few additional ones I would have been prone to try, but they required subscription and payment (a no go at this stage).

Were the outputs to eventually have an inbuilt digital watermark, there could be an easier basis to detect AI generated type text, but even so, each company will have used their own method to generate the (potential) watermark, and I’m pretty sure that adding a step or two in the end-user’s pipeline would largely degrade the watermark’s persistence and utility (though that requires some minimal effort on the user’s side, which is not a done deal).

Indeed, not all services are reliable and free at the same time. Even if that were the case, we simply couldn't go around to each forum post and test if it was written by AI. There are some hints that may indicate that a post was written by AI, but my main concern is that services such as ChatGPT may even create duplicate content if the same parameters are used, and it may cause post bursts and spam issues since it is capable of producing a wide variety of posts within a short period of time. AI may not produce low-quality posts, but it'll definitely create non-engaging discussions, which in some cases, may be worse than plagiarism.

ChatGPT won't be free forever, they're spending more than 100k usd per day on API calls
at some point they'll probably start to charge for it which will change the dynamics a bit

but still important to discuss it here since the price to use it will probably still be lower than what signature campaigns pay here in the forum so people would still have an incentive for arbitrage. (creating low effort posts using AI)

Obito

sr. member

Activity: 1274

Merit: 293

Phew... I'm safe coz GPT-3 isn't available in my country although I could probably do it with VPN, I'm too lazy to do it anyways. Not to mention I feel bad that my writing isn't genuinely coming from me. Good thing that we can use tools to detect and hopefully we got enough manpower to be able to catch all of them.

royalfestus

hero member

Activity: 2464

Merit: 519