Growing ChatGPT use on the site.

[barcoMeCasa]

Originally Posted by grantmc

I'm eager to hear your thoughts and experiences on this topic. Let's share our stories and insights, fostering a community that appreciates the diversity of vessels that grace our beautiful seas.

Fair winds and smooth sailing,
[Your Forum Username]

Very obvious that's not a human post on here. Not opinionated enough and no sarcasm against anyone with an alternative viewpoint. Nowhere near enough bias. Far too positive and cheerful.


Meanwhile drifting a little, https://deepmind.google/discover/blo...-for-geometry/

We are living during times of great change - was it Huxley said you will never see the enormity of revolutions at the time, only in reflection afterwards. Or something like that.

People aren't wearing enough hats.

[wholybee]

Originally Posted by Iron E

It's not that simple, To design an API to, say, have AI appear drunk or angry, I would think that you would need an understanding of Style Transfer Knowledge, NLP's, syntax, vocab nuances, emotional language, and an understanding of at least another dozen English language and programming skills to write the API. Not just, "Act Drunk" It would be easier to have AI generate text, then C&P into a "Drunk Text" generator.

I also wouldn't think that person would need CF to fine-tune sailing-specific responses, and they surely don't need to post in the forum to get data... all sounds very tin foil hat area to me.

Have you used the ChatGPT API? When I was learning it, that was actually one of the first things I tried. "Be Drunk." I also did "Be a teenager", "Be angry", "Be a surfer.", "occasionally give wrong answers." Simple scripts written in Javascript to send and receive HTTP JSON, with those directives in the system messages.

I have then subsequently used that "custom" AI in an android application. It's still GPT4, but tuned to work exactly how I want it too. It is as simple as I describe.

The purpose of using a forum like this, and using it for fine tuning, is to get a better response than just "Be a cruiser responding on an Internet sailing forum." The AI will use the forum to learn the syntax, vocab nuances, emotional language etc. that you speak of. How to use the format strings and how to quote another message. It will also learn some more technical knowledge specific to sailing, LFP batteries, anchoring, what anchor is best, the pros and cons of all the arguments we have, etc. From all that, it could then write a response that is basically indistinguishable from a human on the forum, quoting other messages in the thread, agreeing or disagreeing with others in the conversation, etc. That part is not trivial, but entirely possible, it's done on Facebook, and I think that is what the goal is here.

The Intent of those AI features is so a company can create an AI help assistant, and create fine tuning data (and also use function calling, another feature the API supports) to learn about the companies products, what troubleshooting steps to take, etc. And for the AI to know if it should speak to the user like they are a novice that needs walked though each step, or someone with lots of experience.

Which comes back to my earlier guess. That someone/company is creating an AI product for the cruising community, using CF as data to fine tune/build it.

https://platform.openai.com/docs/api...e/introduction
https://platform.openai.com/docs/guides/fine-tuning

[CrispyCringle]

Who is a AI bot on the forum? I dont doubt it. But I dont see what folks are referring to, including the OPs original post with link.

Would the AIs all please stand up? Thank you.

[wholybee]

Originally Posted by CrispyCringle

Who is a AI bot on the forum? I dont doubt it. But I dont see what folks are referring to, including the OPs original post with link.

Would the AIs all please stand up? Thank you.

Lol, good luck getting them to stand up. The post is 100%, undoubtedly an AI created post. To check it, I posed the question about wind generators to ChatGPT, and was given almost the exact answer, varying by only a few words, and adding another brand not in the post. Which is expected as ChatGPT will not give the exact same answer twice.

Also noting that it was the first post by a user. As if an AI signed up, then opened a discussion, and posted a response. Now image if someone has that figured out, how to have an AI automatically log into this site and open a discussion and make a post, what if they have it make 100 posts a day?

[Wotname]

Originally Posted by wholybee

......
Also noting that it was the first post by a user. As if an AI signed up, then opened a discussion, and posted a response. Now image if someone has that figured out, how to have an AI automatically log into this site and open a discussion and make a post, what if they have it make 100 posts a day?

Also interesting that it is a first post by member who signed up in 2007.

[wholybee]

Originally Posted by Wotname

Also interesting that it is a first post by member who signed up in 2007.

I did not notice that. Very interesting.

[Rohan]

Hmmmm I feel like that scene in the 1982 movie "The Thing" where an alien is killing the antarctic station members one by one and then impersonating them, and one guy says "I know I'm human, but I don't know about any of you"

(or something to that effect)

[Iron E]

Originally Posted by wholybee

Have you used the ChatGPT API? When I was learning it, that was actually one of the first things I tried. "Be Drunk." I also did "Be a teenager", "Be angry", "Be a surfer.", "occasionally give wrong answers." Simple scripts written in Javascript to send and receive HTTP JSON, with those directives in the system messages.

I have used and have subscribed to a few programs that connect to ChatGPT through GPT's API's. I pay on a tiered plan depending on how much I use. Surfer SEO, Koala and a few more. These are software programs designed to interface with ChatGPT through ChatGPT's API.

I'm not really sure what you're referring to when you say you use ChatGPT API. APIs are just an interface that connects GPT to a program. It sounds like you use a software tool that allows you to type in instructions that give you a specific GPT output.


Like I said, I used a few tools that do this. I'm really not impressed with the quality of the output. Mostly just expensive paraphrasing tools with embedded prompts.
My original reply that started this conversation was to a forum member who wanted GPT to do something very particular and I think that my answer of using "Custom Instructions" is the right way to begin.

I did not notice that. Very interesting.

It's an important piece of information I addressed in my 2nd post. AI wasn't around in 2007, but spam was,
You are correct; comment CAPTCHA for a new user's first 25 comments would help. Time limits on new members like CF have in the download section would also help. And user inactivity bans would also help.

[wholybee]

Originally Posted by Iron E

I have used and have subscribed to a few programs that connect to ChatGPT through GPT's API's. I pay on a tiered plan depending on how much I use. Surfer SEO, Koala and a few more. These are software programs designed to interface with ChatGPT through ChatGPT's API.

I'm not really sure what you're referring to when you say you use ChatGPT API. APIs are just an interface that connects GPT to a program. It sounds like you use a software tool that allows you to type in instructions that give you a specific GPT output.


Like I said, I used a few tools that do this. I'm really not impressed with the quality of the output. Mostly just expensive paraphrasing tools with embedded prompts.
My original reply that started this conversation was to a forum member who wanted GPT to do something very particular and I think that my answer of using "Custom Instructions" is the right way to begin.



It's an important piece of information I addressed in my 2nd post. AI wasn't around in 2007, but spam was,
You are correct; comment CAPTCHA for a new user's first 25 comments would help. Time limits on new members like CF have in the download section would also help. And user inactivity bans would also help.

What I mean be using the API is actually using it directly, not using a program that uses it. The "playground" is an interface openai has for testing the API via a web interface, where you can tweak all of the settings and write system messages and directives about how you want it to act. As a way to test different scenarios before you implement them in code.

I write software as a hobby and have written some programs that use the API directly. The API works exactly as I describe. It isn't a "tool" in is integral to how ChatGPT works. When you use ChatGPT from the normal user facing interface, there are already many system directives that set how it behaves. Mostly protections against abuse, explicit content, or to prevent it from going off the rails. The API opens that up and lets a developer customize those directives how they want/need. You can set a system message to "Act Drunk" and it will. For something simple like that, that is all you need to do. You can add more complicated instructions or fine tuning if you need more complicated behavior. All of that is documented on the openai links I provided above.

The "playground" is here (you might need a paid developer account to use the playground, I'm not sure, as I have a paid dev account):
https://platform.openai.com/playground?mode=chat

The account from 2007 is interesting. I don't know what it means. One possibility (actually a common problem on Facebook) is that older inactive accounts are hacked in lieu of making a new account. It's a way to stay more under the radar, and instantly have an account with many friends. But that is only a guess, who knows in reality why an account created in 2007 now has an AI posting from it. Maybe it is as simple as a guy started the account in 2007, and now is involved in AI development and thus is using the account for that.

[SV_FlyingTigress]

Originally Posted by jeffandlori

All gen 1 tech looks like a wonky toy. We're at the Atari pong level of this, what people should be considering is what the later versions will be like. The other thing that people in general are not understanding is how quickly those later versions will appear.

^^^ This is utterly correct. I am in AI security for a living. The security implications of AI, particularly the more recent GenAI/LLM's is monstrous.

[SV_FlyingTigress]

Just for laughs, one of my kids had a LLM create a pirate talk version of the bible. Looks like someone took the idea and printed it as the "Pirate Bible" (no not me; nobody I know) and is selling it on Amazon. I thought chapter 3 was particularly funny with the pirate version of the serpent interaction with Adam/Eve

[skipperpete]

Whatever we might think about large language models, they’re not going away anytime soon , here in Australia Chatgpt has just been approved for use in schools by students to assist them with essays and search, the reasoning is that its similar to allowing students to take scientific calculators into mathematics classes and exams.
My iphone has a Chatgpt keyboard in settings that can proofread my emails, translate my emails into any language and even change the “tone” of a message.
i dont u se itt tho

[requiem]

Originally Posted by leecea

IMO the first problem with AI or any prolific source of poor content is that the signal to noise ratio gets so bad that you can barely find the useful stuff. I experienced it years ago with a rapidly growing car forum and I think it's already happened to the overall set of web content.

I suspect the growth of content marketing has contributed to this by driving demand for such content. Unfortunately the AI-created content is particularly untrustworthy (not that intern-created content is much better) and it's poisoning search results.

[s/v Jedi]

AI’s benefit from the forum by absorbing all it’s content to gain knowledge.

The forum benefits from AI because it increases revenue, with this thread being the ultimate form of that

I wonder if an AI can be entertaining itself?

[GordMay]

Originally Posted by s/v Jedi

... I wonder if an AI can be entertaining itself?

What Happens if an AI Gets Bored?
In theory, it could become self-destructive - or even sadistic.
More ➥ https://www.scientificamerican.com/a...ai-gets-bored/