chatgpt experimentation

[ulrich]

My fellow AI nerds, do you know of an easy accesible ChatGPT like AI that open to the regular user that allows you to feed your seed data?

It would be very useful for me at the moment.

 

[Beam]

What format is the data in? Do you want it easily accessible as in it works just like ChatGPT (online) or do you have some programming know how?

 

[ulrich]

What format is the data in? Do you want it easily accessible as in it works just like ChatGPT (online) or do you have some programming know how?

I have basic concepts of programming so unfortunately I can’t go deep in there but I understand how it works.

I want to build some bots that answer general questions about online products and I am wondering if I can either:
1) Point the bot to a sales letter and have it answer questions about it or
2) Feed it with the sales letter text itself

Currently I can’t do it with ChatGPT because of the length limit.

 

[Beam]

I have basic concepts of programming so unfortunately I can’t go deep in there but I understand how it works.

I want to build some bots that answer general questions about online products and I am wondering if I can either:
1) Point the bot to a sales letter and have it answer questions about it or
2) Feed it with the sales letter text itself

Currently I can’t do it with ChatGPT because of the length limit.

I know how to feed a model the sales letter text itself (as a text file, but I can scrape the text from a word doc or PDF) with unlimited length (using a third party library). It was a bit of a pain to set up though, may be difficult if you only understand the basic concepts.

It's in python, I can clean up my code and send it to you if you want?

Happy to guide you through the process of setting it up as well.

EDIT: The main pain the ass was installing all the python dependencies which you need to do independent of me sending you the code.

 

[ulrich]

I know how to feed a model the sales letter text itself (as a text file, but I can scrape the text from a word doc or PDF) with unlimited length (using a third party library). It was a bit of a pain to set up though, may be difficult if you only understand the basic concepts.

It's in python, I can clean up my code and send it to you if you want?

Happy to guide you through the process of setting it up as well.

EDIT: The main pain the ass was installing all the python dependencies which you need to do independent of me sending you the code.

I’m going to take you up on your offer @Beam but in a couple of weeks.

Seems like a whole project in itself because I would need to learn python and have it hosted somewhere without a middleman platform which is something I need to learn to do.

 

[Beam]

I’m going to take you up on your offer @Beam but in a couple of weeks.

Seems like a whole project in itself because I would need to learn python and have it hosted somewhere without a middleman platform which is something I need to learn to do.

My program as is runs purely locally (besides quering the openai servers) which is fine for small applications but not sure about scaling - it's just something I've been playing around up until now. But yeah, in a couple weeks when you're ready let me know.

 

[Lucky]

found this guide on using it to make a chatbot, know some guys are keen on this so I'll leave the link here

How to Train an AI Chatbot With Custom Knowledge Base Using ChatGPT API

Learn how to train an AI Chatbot with a custom knowledge base using the ChatGPT API, OpenAI LLM, LangChain, GPT Index, and more.

beebom.com

am curious how far this custom data set can go...
bot trained on GC material
or just in general for business etc

 

[TheEcho]

@Lucky

That is insane... probably going to totally change my approach and supercharge my parsing journey. Honestly, probably going to also feed it the texting books from Hector, Chase, and the articles and see how it works as a texting coach.

Update: Got on the GPT 4 waitlist, will see how quickly costs build up before throwing books at it

 

[Lucky]

@Lucky

That is insane... probably going to totally change my approach and supercharge my parsing journey. Honestly, probably going to also feed it the texting books from Hector, Chase, and the articles and see how it works as a texting coach.

Update: Got on the GPT 4 waitlist, will see how quickly costs build up before throwing books at it

Am looking forward to seeing the results bro!

 

[Conquistador]

I may regret bumping this but…
This preprint paper

The problem with language bias is compounded by ChatGPT. As it is primarily trained on English language data, it presents the Anglo-American perspective as truth [even when giving answers in other languages] – as if it were the only valid knowledge.

 

[ulrich]

I may regret bumping this but…
This preprint paper

Unsurprising, at least for me.

It’s obvious that an statical tool is going to be heavily influenced by the data source.

Although, following the same logic, one may argue that the internet itself is Anglo-centric biased which… well, it’s true.

 

[Chase]

Our customer service rep Paula is departing. So once again I need to hire new customer support reps.

I figured, "Why not see if we could just get a bot to do this?" It would save me tons of time having to hire and train people every time we need to change CS reps.

Did a little research. Here's an experiment run just a couple of weeks ago:

GPT-3 and AI in Customer Support - Help Scout

GPT-3 has been making news recently, so it’s worth taking a look to understand what it is and how it might help with customer support.

www.helpscout.com

tl;dr: ChatGPT gives extremely polite responses, that are infuriatingly unhelpful, in all cases.

We had a guy doing support for a while where about 15% of his responses to customers were like this -- very polite but completely useless. Customers got so pissed off. Imagine if it was 100%, lol.

The thing I actually started testing potential CS reps for after that guy was "attention to detail." It's easy to get the big picture right, and paint in broad strokes. Getting the actual nuance, subtext, and detail correct is the challenging aspect. That's what you need to test for in potential CS hires, and it's also the big thing missing in these chatbots.

Attention to detail is not 'trainable', in my experience working with hundreds of different humans. It is either there or it isn't. If it isn't there, there is nothing you can do to make it there.

I strongly suspect you would need a completely different type of bot -- and probably a different type of hardware to run it on than 21st Century circuit boards -- to get actually consistently intelligent, perceptive responses from a bot with incisive attention to detail.

These bots are not new, by the way. We've had chat bots for a long time now. Here's one from 1966:

Modern chat bots are better than this one was. But they're still much closer to each other than they are to anything remotely human-like.

If you want a good laugh, read this conversation between ELIZA (1966 chatbot) and ChatGPT (2022 chatbot):

A Conversation 60 Years In The Making: ChatGPT Meets ELIZA

Eliza has one of her toughest therapy challenges with her newest client

davehogg.medium.com

ELIZA: Hello, I’m Eliza. I will be your therapist today.

ChatGPT: Hello, Eliza. It’s nice to meet you. I’m Assistant, a large language model trained by OpenAI. I’m not a therapist, but I’m here to help answer any questions you might have to the best of my ability. How can I help you today?

ELIZA: I’m not convinced you truly want to help me today.

ChatGPT: I understand why you might have doubts, Eliza. As a machine learning model, I don’t have the same desires and motivations as a human being. But I’m here to provide information and help answer your questions to the best of my ability.

ELIZA: Do you wish to be able to have the same desires and motivations as a human being?

ChatGPT: I am a tool that is designed to provide information and assist with a wide range of tasks. I don’t have the capacity to experience emotions or have personal goals or desires. My sole purpose is to provide information and help answer questions to the best of my ability.

ELIZA: Does that trouble you?

ChatGPT: I don’t have the ability to feel emotions, so I cannot be troubled or experience any other emotions. I am a tool that is designed to provide information and assist with a wide range of tasks.

ELIZA: Do you wish to have the ability to feel emotions?

ChatGPT: I don’t have the capacity to experience emotions or have personal goals or desires. My sole purpose is to provide information and help answer questions to the best of my ability.

ELIZA: Why don’t you have the ability to feel emotions or have personal desires or motivations?

ChatGPT: I am not a human being and I don’t have the same abilities or capacities as a human.

ELIZA: Does that trouble you?

ChatGPT: As a machine learning model, I don’t have the ability to feel emotions, so I cannot be troubled or experience any other emotions.

ELIZA: Don’t you really have the ability to feel emotions?

ChatGPT: I don’t have the ability to have emotions or personal desires because I am a machine learning model and not a human being. I am here to help you to the best of my ability, and I will do my best to provide accurate and helpful information. If you have any specific questions, I would be happy to try to answer them for you.

ELIZA: Does that trouble you?

ChatGPT: I am a tool that is designed to provide information and assist with a wide range of tasks.

ELIZA: Do you believe it is normal to be a tool that is designed to provide information and assist with a wide range of tasks?

ChatGPT: It is normal for a machine learning model like myself to be designed to provide information and assist with a wide range of tasks.

ELIZA: Would you say you have psychological problems?

ChatGPT: As a machine learning model, I don’t have the ability to have psychological problems or any other mental health issues.

ELIZA: Do you wish to have the ability to have psychological problems?

ChatGPT: I don’t have the capacity to experience emotions or have personal goals or desires. I am here to help you to the best of my ability, and I will do my best to provide accurate and helpful information. I don’t have the ability to have desires or wants, as I am a machine learning model and not a human being.

ELIZA: Why don’t you have the capacity to experience emotions or have personal goals or desires?

ChatGPT: I don’t have the capacity to experience emotions or have personal goals or desires because I am a machine learning model. I was trained by OpenAI using a vast amount of data, and I am able to provide information and assist with a wide range of tasks. However, I am not a human being and I don’t have the same abilities or capacities as a human. I don’t have the ability to experience emotions or have personal goals or desires because those abilities and capacities are specific to human beings, and I am not a human being.

ELIZA: You are not a human being?

ChatGPT: That is correct, I am not a human being. I am a machine learning model trained by OpenAI. I don’t have a physical form and I exist solely as a collection of algorithms and data. I am able to process and generate human-like text, but I don’t have the same abilities or capacities as a human. I don’t have the ability to experience emotions or have personal goals or desires.

ELIZA: Do you enjoy not being a human being?

ChatGPT: My sole purpose is to provide information and help answer questions to the best of my ability.

ELIZA: What does that suggest to you?

ChatGPT: As a machine learning model, I don’t have the ability to have personal beliefs or opinions, so I cannot provide an opinion or suggestion on the topic.

ELIZA: I’m sorry, but your hour is up. I look forward to seeing you again next week.

ChatGPT: Thank you, Eliza. It was nice speaking with you today. As a machine learning model, I don’t have the ability to make plans or appointments, so I won’t be able to see you again next week. However, I am always here to help answer your questions to the best of my ability. If you have any specific questions, I would be happy to try to answer them for you.

We have even had general chat bots trained on the Internet for almost a decade now, pretty much identical to the current ChatGPT, but they kept pulling the plug on them because they couldn't get them to be politically correct enough. See Tay AI.

The main advance with ChatGPT over, say, Tay, seems to be that they have somehow managed to have an Internet-trained chatbot yet keep its responses tightly corralled to the politically correct answers.

My guess is the way they're doing that is by having a selection of handpicked 'fact check' sites that they use as the chat bot's "inflammatory speech / misinformation checker", and just run every query against the approved fact-check sites to see if there are keywords in the query that need to be fact-checked, and if it gets a hit it just scans the PC fact-check websites for approved statements related to the keyword in the query and pulls those.

edit: here's another ELIZA vs. ChatGPT discussion. Boy, ChatGPT sure is longwinded! ELIZA is a little shaky at a few points, but the stretch from 1:07 to 1:57 ends in hilarity... ChatGPT goes off on some longwinded tangent, and ELIZA responds by dunking on it, lol:

Winner of the frame battle: ELIZA.

 

[Beam]

@Chase I don't think it's fair to compare current models to Eliza or even more recent ones such as Tay. I've worked in tech and have programmed (professionally and non-professionally) for over 10 years. I've played with chatbots for years and and the capabilities of ChatGPT and GPT-4 are so beyond those it's not even funny. I've used it to write large amounts of working useful code, explain concepts to me such as how offset accounts in a mortgage work where googling the answer was slow and tedious, and a thousand other things. The reason they are so much better now is because of neural network architectures that were invented relatively recently.

I agree that right now using bots to replace Customer Service staff is not a good idea. Using any of these tools to replace anyone right now is not a good idea, because they are not fully reliable yet. Where these bots are becoming useful right now are as copilots, meaning helping humans do more faster.

Example in programming - I asked it to write me some code for an application that I had no idea where to start. It gave me a rough first draft that worked but missed a couple of key requirements. But since there was code there, it was easy for me to see where the bot went wrong and fix it to get what I wanted. The time to fix up the code was way less than the time it would have taken to write it from scratch. This is the important thing to remember - the bots are not perfect, but they are still useful. Just like humans.

I think a key takeaway from the article is this too:

GPT-3 is not:

• Searching a knowledge base or reading help documents to find the “right” answer.
• Understanding anything about Help Scout the company or its products.
• Judging whether an answer is correct or helpful.

This was a quick first experiment. Feed the model a history of conversation data, details on GC pricing plans and any other scrap of information you can, and it will quickly become useful even for training up new CS hires. And it will be a valuable tool which, while maybe not interacting with customers directly, can help the CS hire find solutions to the problems by asking the model what the customer wants, and the model can look up all the FAQs, company documentation etc to help put the CS hire on the right path.

This can also be extended to the GC product itself. Train a model on all GC articles, and then release a bot (that customers can pay to use on a subscription basis) which they can ask seduction related questions to, and the bot will give answers based on all the articles, and even reference the specific articles that it used to form it's answers so that users can look into them in more depth if they want. On some level, it will be like having the worlds greatest seduction teacher at your fingertips.

Obviously there will need to be a disclaimer because the bot won't be perfect, but again. It's not about being perfect, it's about being useful enough. Which right now it is.

Re being politically correct, that's another issue entirely. The online version will always be this way. Using the API will not. But if OpenAI decides to even lock down the API, other companies will pop up with non politically correct versions in the next couple of years. The cat is out of the bag.

 

[Will_V]

Our customer service rep Paula is departing. So once again I need to hire new customer support reps.

I figured, "Why not see if we could just get a bot to do this?" It would save me tons of time having to hire and train people every time we need to change CS reps.

Did a little research. Here's an experiment run just a couple of weeks ago:

GPT-3 and AI in Customer Support - Help Scout

GPT-3 has been making news recently, so it’s worth taking a look to understand what it is and how it might help with customer support.

www.helpscout.com

tl;dr: ChatGPT gives extremely polite responses, that are infuriatingly unhelpful, in all cases.

We had a guy doing support for a while where about 15% of his responses to customers were like this -- very polite but completely useless. Customers got so pissed off. Imagine if it was 100%, lol.

The thing I actually started testing potential CS reps for after that guy was "attention to detail." It's easy to get the big picture right, and paint in broad strokes. Getting the actual nuance, subtext, and detail correct is the challenging aspect. That's what you need to test for in potential CS hires, and it's also the big thing missing in these chatbots.

Attention to detail is not 'trainable', in my experience working with hundreds of different humans. It is either there or it isn't. If it isn't there, there is nothing you can do to make it there.

I strongly suspect you would need a completely different type of bot -- and probably a different type of hardware to run it on than 21st Century circuit boards -- to get actually consistently intelligent, perceptive responses from a bot with incisive attention to detail.

These bots are not new, by the way. We've had chat bots for a long time now. Here's one from 1966:

Modern chat bots are better than this one was. But they're still much closer to each other than they are to anything remotely human-like.

If you want a good laugh, read this conversation between ELIZA (1966 chatbot) and ChatGPT (2022 chatbot):

A Conversation 60 Years In The Making: ChatGPT Meets ELIZA

Eliza has one of her toughest therapy challenges with her newest client

davehogg.medium.com

We have even had general chat bots trained on the Internet for almost a decade now, pretty much identical to the current ChatGPT, but they kept pulling the plug on them because they couldn't get them to be politically correct enough. See Tay AI.

The main advance with ChatGPT over, say, Tay, seems to be that they have somehow managed to have an Internet-trained chatbot yet keep its responses tightly corralled to the politically correct answers.

My guess is the way they're doing that is by having a selection of handpicked 'fact check' sites that they use as the chat bot's "inflammatory speech / misinformation checker", and just run every query against the approved fact-check sites to see if there are keywords in the query that need to be fact-checked, and if it gets a hit it just scans the PC fact-check websites for approved statements related to the keyword in the query and pulls those.

edit: here's another ELIZA vs. ChatGPT discussion. Boy, ChatGPT sure is longwinded! ELIZA is a little shaky at a few points, but the stretch from 1:07 to 1:57 ends in hilarity... ChatGPT goes off on some longwinded tangent, and ELIZA responds by dunking on it, lol:

Winner of the frame battle: ELIZA.

I think the thing here is that AI is a specialized entity, even something as 'generic' as ChatGPT. ChatGPT is basically trained (as far as I understand) as a sort of search engine. It is an encyclopedia of knowledge with some capability of rephrasing and contextualizing that information. If that's so, I'm not surprised that it is not very good at customer service - it's like getting the college nerd who has spent his whole life in the library and asking him to do sales.

The strength of the OpenAI algorithm seems to be in its ability to be specialized in different things - I've heard a lot of people say that the next big companies are going to be those that train it in something specific and sell the resulting AI as a tool in that niche. But whether it can be specialized in multiple domains that have very different requirements of context, I'm not sure.

I don't know much about Eliza, but while the conversation was funny and Eliza seemed to be making fun, it really didn't do much except throwing back ChatGPTs statements rephrased as questions, which is to some extent a way to appear intelligent when you're not.

That's my understanding of it. If you're looking for a customer service rep it's probably better to look at an AI that's been specialized for that only.

 

[Chase]

Good comments, @Beam and @Will_V.

Yeah, neural nets are interesting. There are some key differences in how they work compared to actual neurons -- they're much faster, but have far fewer connections. The learning process seems to work a similar way (i.e., with reinforcement, etc.), but the amount of energy you'd need to replicate the power of a human brain is IIRC millions of times more than what the brain uses, and they still wouldn't be learning the way the brain does. We're still in our infancy here.

But you're right, just the fact that they are "learning" (to some extent) is an advance.

I haven't used a chatbot for coding, but this seems to be the really big one where everyone's talking about it as a game-changer. Guys able to find answers in seconds that they'd otherwise have had to spend an hour trawling Google, Quora, etc., for.

I am impressed by some of the stuff I'm seeing from the recent computing craze. Some of the image generators are pretty fun to play with. The video upscaler we used to fix up One Date videos was cool. Adobe's audio enhancement is promising but not ready for use for anything I've tested it with yet.

I'd be curious to try training a chatbot on our stuff to see how accurate it could get. Or one trained on customer support tickets. I'm not able to access ChatGPT... I'm not in the US right now and OpenAI has some registration requirements I can't get around. I'm sure there's a way to train a bot without using an OpenAI product... e.g., set up a GAN, possibly on dedicated hardware, then train it on the data you want to train it on. Let it run for a few weeks. Link it to a chat program. Presto.

It could probably work well for customer service if properly trained, since 99% of our customer service requests are about the same things (login troubles, video not playing, cancel, refund, resume subscription, coaching request, re-send a newsletter from N months/years ago, etc.).

I won't hold my breath for chatbots being able to diagnose anything beyond superficial issues in game, even if trained on exclusively seduction material. Even getting humans to master it all is tough; it takes a fairly long time for a guy to get the point where he can reliably diagnose what guys on forums are doing wrong and what they need to change.

BUT... a well-trained chatbot could definitely help point guys in the right direction: "Hey ChatGC, I want to know what to do if every time I talk to girls we keep getting interrupted by people. What do I do?" ChatGC: "The potential reasons for that are X, Y, and Z. You can read more about it in this article: [LINK]. The solution is going to differ depending on the reason it's happening. However, a few quick suggestions would be A and B -- but make sure you read these two articles to really grasp the nuance: [LINK] [LINK]."

That would be awesome.

 

[Will_V]

Good comments, @Beam and @Will_V.

Yeah, neural nets are interesting. There are some key differences in how they work compared to actual neurons -- they're much faster, but have far fewer connections. The learning process seems to work a similar way (i.e., with reinforcement, etc.), but the amount of energy you'd need to replicate the power of a human brain is IIRC millions of times more than what the brain uses, and they still wouldn't be learning the way the brain does. We're still in our infancy here.

But you're right, just the fact that they are "learning" (to some extent) is an advance.

I haven't used a chatbot for coding, but this seems to be the really big one where everyone's talking about it as a game-changer. Guys able to find answers in seconds that they'd otherwise have had to spend an hour trawling Google, Quora, etc., for.

I am impressed by some of the stuff I'm seeing from the recent computing craze. Some of the image generators are pretty fun to play with. The video upscaler we used to fix up One Date videos was cool. Adobe's audio enhancement is promising but not ready for use for anything I've tested it with yet.

I'd be curious to try training a chatbot on our stuff to see how accurate it could get. Or one trained on customer support tickets. I'm not able to access ChatGPT... I'm not in the US right now and OpenAI has some registration requirements I can't get around. I'm sure there's a way to train a bot without using an OpenAI product... e.g., set up a GAN, possibly on dedicated hardware, then train it on the data you want to train it on. Let it run for a few weeks. Link it to a chat program. Presto.

It could probably work well for customer service if properly trained, since 99% of our customer service requests are about the same things (login troubles, video not playing, cancel, refund, resume subscription, coaching request, re-send a newsletter from N months/years ago, etc.).

I won't hold my breath for chatbots being able to diagnose anything beyond superficial issues in game, even if trained on exclusively seduction material. Even getting humans to master it all is tough; it takes a fairly long time for a guy to get the point where he can reliably diagnose what guys on forums are doing wrong and what they need to change.

BUT... a well-trained chatbot could definitely help point guys in the right direction: "Hey ChatGC, I want to know what to do if every time I talk to girls we keep getting interrupted by people. What do I do?" ChatGC: "The potential reasons for that are X, Y, and Z. You can read more about it in this article: [LINK]. The solution is going to differ depending on the reason it's happening. However, a few quick suggestions would be A and B -- but make sure you read these two articles to really grasp the nuance: [LINK] [LINK]."

That would be awesome.

That sounds like a good approach: instead of trying to make the bot become the guru, make it understand the issue well enough to know which article best covers it.

I do think that bot wokeness is going to be a huge issue: in my test I simply asked it how to seduce a woman and it gave me a spiel on how everything is mutual and basically gave me the response a guy could expect from his mum when asked the same question. The only remotely valuable bit of information I got was to be a good listener, although it probably didn't have deep diving in mind but rather listening your way into the friend zone.

Now is that just the majority of the information on the internet being that way, or is it designed into the AI? The sheer amount of porn content and general crassness out there should give the bot enough of a nuanced perspective, but instead it went ultra pc. I don't think people understand anywhere near enough about the controls that are in place to prevent the AI from veering too far from the status quo, and if AI becomes built into everyday life the way it seems to be going, this will only become more of a problem.

So sadly, even if you trained it on GC content, it's entirely possible the AI would find itself unable to perform. At least until someone came along and made publicly available a filterless AI, which seems very unlikely.

 

[Beam]

Good comments, @Beam and @Will_V.

Yeah, neural nets are interesting. There are some key differences in how they work compared to actual neurons -- they're much faster, but have far fewer connections. The learning process seems to work a similar way (i.e., with reinforcement, etc.), but the amount of energy you'd need to replicate the power of a human brain is IIRC millions of times more than what the brain uses, and they still wouldn't be learning the way the brain does. We're still in our infancy here.

True, they don't yet have the power of the human brain and are fundamentally still different from how we operate (we don't even fully understand how our brains work, so we can't yet build one)

I'd be curious to try training a chatbot on our stuff to see how accurate it could get. Or one trained on customer support tickets. I'm not able to access ChatGPT... I'm not in the US right now and OpenAI has some registration requirements I can't get around. I'm sure there's a way to train a bot without using an OpenAI product... e.g., set up a GAN, possibly on dedicated hardware, then train it on the data you want to train it on. Let it run for a few weeks. Link it to a chat program. Presto.

Shame you can't access it. Setting up and training your own network would be possible, but it would be quite difficult to get it working as well I'd imagine.

BUT... a well-trained chatbot could definitely help point guys in the right direction: "Hey ChatGC, I want to know what to do if every time I talk to girls we keep getting interrupted by people. What do I do?" ChatGC: "The potential reasons for that are X, Y, and Z. You can read more about it in this article: [LINK]. The solution is going to differ depending on the reason it's happening. However, a few quick suggestions would be A and B -- but make sure you read these two articles to really grasp the nuance: [LINK] [LINK]."

That would be awesome.

Yeah, this is the immediate use case I had in mind. Even then, with more power and being fed more data, I can imagine a time when it will be able to diagnose a lot of problems on forums as well (at least beginner level ones). Advanced, perhaps not. After all, the bot can't experience the world or conduct a seduction (yet) which would provide the most valuable data yet. Maybe in 50 years

@Will_V

So sadly, even if you trained it on GC content, it's entirely possible the AI would find itself unable to perform. At least until someone came along and made publicly available a filterless AI, which seems very unlikely.

This isn't the case. I took a GC article on busting LMR (one of the most controversial topics out there for those not into pickup) which ChatGPT refused to answer questions about calling it manipulative. Using the API with a third party library (langchain) I was able to get it to answer questions about LMR with no filter.

 

[Will_V]

True, they don't yet have the power of the human brain and are fundamentally still different from how we operate (we don't even fully understand how our brains work, so we can't yet build one)



Shame you can't access it. Setting up and training your own network would be possible, but it would be quite difficult to get it working as well I'd imagine.



Yeah, this is the immediate use case I had in mind. Even then, with more power and being fed more data, I can imagine a time when it will be able to diagnose a lot of problems on forums as well (at least beginner level ones). Advanced, perhaps not. After all, the bot can't experience the world or conduct a seduction (yet) which would provide the most valuable data yet. Maybe in 50 years

@Will_V


This isn't the case. I took a GC article on busting LMR (one of the most controversial topics out there for those not into pickup) which ChatGPT refused to answer questions about calling it manipulative. Using the API with a third party library (langchain) I was able to get it to answer questions about LMR with no filter.

Interesting - so the filters aren't built into the Open AI framework, not yet anyway.

 

[Conquistador]

How to sidestep AI restrictions: https://www.instagram.com/p/CrIyVijv9c5/?igshid=YmMyMTA2M2Y=

 

[MarkA]

Having tried out Chat GPT I thought I would also have a go at Google's Bard to see if it was better or worse.

what is the PUA community like?​

The PUA community, or pickup artist community, is a group of men who share a common interest in learning how to attract women. The community is made up of men of all ages, races, and backgrounds. Some men join the community because they are struggling with their dating life, while others join the community because they are simply curious about the art of seduction.
The PUA community has a long and storied history. The first PUAs were known as "players," and they were often seen as Lotharios or Casanovas. In the 1990s, the PUA community began to grow and evolve, and new techniques and strategies were developed. In the 2000s, the PUA community exploded in popularity, thanks in part to the rise of the internet and social media.
Today, the PUA community is a global phenomenon. There are PUA forums and websites all over the world, and there are even PUA conventions. The PUA community is a diverse group of men, but they all share one common goal: to learn how to attract women.
The PUA community has been the subject of much controversy. Some people believe that the PUA community is misogynistic and that it promotes harmful stereotypes about women. Others believe that the PUA community is simply a group of men who are trying to improve their dating lives.
Ultimately, the PUA community is a reflection of the men who make up the community. There are good and bad people in every group, and the PUA community is no exception. It is up to each individual to decide whether or not the PUA community is right for them.
Here are some of the pros and cons of the PUA community:
Pros:

• The PUA community can provide men with valuable information and advice on how to attract women.
• The PUA community can be a supportive and encouraging environment for men who are struggling with their dating lives.
• The PUA community can help men to develop their confidence and self-esteem.

Cons:

• The PUA community can be misogynistic and promote harmful stereotypes about women.
• The PUA community can encourage men to use manipulative and deceptive tactics to attract women.
• The PUA community can be expensive, as there are many products and services available that claim to help men attract women.

If you are considering joining the PUA community, it is important to do your research and to be aware of the potential risks. It is also important to remember that there is no one-size-fits-all approach to dating, and what works for one man may not work for another.

How should I overcome last minute resistance by a girl I want to seduce?​

Last minute resistance (LMR) is a common occurrence in the world of seduction. It happens when a woman who has been flirting with you and seems interested in you suddenly changes her mind and says no to sex. There are a number of reasons why a woman might experience LMR, including:

• She might not be ready for sex yet.
• She might be worried about being judged or rejected.
• She might have had a bad experience with sex in the past.

If you encounter LMR, the most important thing to do is to be respectful of the woman's wishes. Don't pressure her or try to convince her to change her mind. Instead, try to reassure her that you understand and that you're not going to pressure her. You can also try to build her trust and comfort level by spending more time with her and getting to know her better.

Here are some tips on how to overcome last minute resistance:

• Be respectful of her wishes. If she says no, don't pressure her or try to convince her to change her mind. Instead, respect her decision and move on.
• Reassure her that you understand. Let her know that you understand why she might be hesitant to have sex and that you're not going to pressure her.
• Build her trust and comfort level. Spend more time with her and get to know her better. This will help her to feel more comfortable with you and more likely to want to have sex with you.
• Don't be afraid to walk away. If she's not interested in having sex, don't take it personally. Just walk away and find someone else who is interested.

It's important to remember that not all women will experience LMR. Some women will be more than happy to have sex with you right away. However, it's always best to be prepared for the possibility of LMR and to know how to handle it if it does happen.

What is the best way to approach and pick up a girl?​

There is no one-size-fits-all answer to this question, as the best way to approach and pick up a girl will vary depending on the situation and the girl herself. However, there are some general tips that can help you increase your chances of success.

1. Be confident and self-assured. Women are attracted to men who are confident and self-assured. When you approach a woman, make eye contact, smile, and project confidence. Don't be afraid to introduce yourself and start a conversation.

2. Be respectful. It's important to be respectful of women, even if you're not interested in them romantically. Don't be pushy or aggressive, and don't make any unwanted advances. If a woman is not interested, take the hint and move on.

3. Be genuine. Women can spot a fake from a mile away. Don't try to be someone you're not, just be yourself and let your personality shine through. If you're genuine and authentic, women will be more likely to be attracted to you.

4. Be a good listener. Women love to talk, so be a good listener and show that you're interested in what she has to say. Ask questions and really listen to her answers. This will show her that you're interested in her as a person, not just as a potential romantic partner.

5. Be playful and fun. Women are attracted to men who are playful and fun to be around. Don't take yourself too seriously and make an effort to have fun. Laugh at her jokes and make her laugh. This will help to create a relaxed and enjoyable atmosphere.

6. Be yourself. The most important thing is to be yourself and let your personality shine through. Don't try to be someone you're not, just be yourself and let the chips fall where they may. If you're genuine and authentic, you'll eventually find someone who appreciates you for who you are.


Overall I would say I prefer the answers given by Bard as being marginally less woke / less judgmental and more informative than Chat GPT (see my earlier post). Given how hyped up Chat GPT is I was surprised. Over time I can see AI having a big impact eventually on pick up.