secret hack to get good fast

[krockerman]

1.

Salesjobs (best option to get good is door to door sales):

Sales and Game have so much in common, so you will learn to get good fast at many of the fundemental skills, learn the principels first hand and get good at the technical stuff. 

2.

AI simmulations

I am working on AI simmulations, it might not be the most accurate but if you are unable to get enogh refrences you can still use simmulations to get good at various areas of game.

 

[NoSelfSelf]

Only thing is that sales give you infrastructure already on how you should move.While one must create it for himself(usually dont know how since one doesnt know himself)but when one is busy looking to fuck,need approval,thinks she's sexy then one moves on the infrastructure of women and that always either makes you lose yourself or getting short end of the stick.

 

[krockerman]

1 hour ago, NoSelfSelf said:

Only thing is that sales give you infrastructure already on how you should move.While one must create it for himself(usually dont know how since one doesnt know himself)but when one is busy looking to fuck,need approval,thinks she's sexy then one moves on the infrastructure of women and that always either makes you lose yourself or getting short end of the stick.

 

Ofcrouse it's a diffrent skillset and you are gonna have to adjust your approch but the fundementals of closing, hadnling social pressure, calibration, esclation (in the context of sales) mindset, value offering, rapport, rejection mangement etc goes along way. I have dated multiple girls at once. Now I have moved on but unless for sales I would not have been able to.

Edited November 7 by krockerman

[NoSelfSelf]

35 minutes ago, krockerman said:

Ofcrouse it's a diffrent skillset and you are gonna have to adjust your approch but the fundementals of closing, hadnling social pressure, calibration, esclation (in the context of sales) mindset, value offering, rapport, rejection mangement etc goes along way. I have dated multiple girls at once. Now I have moved on but unless for sales I would not have been able to.

If it gets you this then 🫡

[Schizophonia]

4 hours ago, NoSelfSelf said:

If it gets you this then 🫡

ahah

[NoSelfSelf]

16 minutes ago, Schizophonia said:

ahah

Its made as a joke but its actually game.

[Optimal_Prime]

good at what, pick up?

[Raze]

20 hours ago, krockerman said:

AI simmulations

How does this help?

[krockerman]

6 hours ago, Raze said:

How does this help?

Why do pilots use flight simmulators? how does it help? 

[Lucasxp64]

On 07/11/2024 at 4:27 AM, krockerman said:

2.

AI simmulations

I am working on AI simmulations, it might not be the most accurate but if you are unable to get enogh refrences you can still use simmulations to get good at various areas of game.

 

What kind of prompts you are using? Those common LLMs out there are too aligned with the purpose of being a "Helpful assistant" speech that it's hard for me to get it answering like an actual woman. But I believe if we had a lot of high quality conversation examples it could get better. They don't offer enough resistance and roleplay capabilities are lobotomized, they tend to be bad at narratives with conflict. They sound like some badly written roleplay.

The best results I had were with some uncensored models that were weakly aligned, they were closer to an auto-complete model than a "helpful assistant" alignment, Like "Smol LM 2 1.7b Instruct". https://huggingface.co/HuggingFaceTB/SmolLM-1.7B-Instruct Btw, use on the SYSTEM prompt how it should align, the user prompt is weaker at this.

But the 360M parameter model funny enough gave me better roreplay because it's less aligned... Maybe because it's dumber it works more like an auto-complete which tends to bias wildly how it answers based on the input text, which is good to give answers outside of that "helpful assistant" lobotomized crap. But it's too dumb to be useful, I don't recommend.

I'm still looking into other models that are larger. I could also set up in Kaggle (30h guarantee gpu uptime per week for free, an alternative to Google Colab) and use their Nvidia Tesla P100 (16GB) or T4 (16GB x2 SLI), but it's better if it can run on a CPU because I don't have a lot of GPU allowance. 

Edited November 8 by Lucasxp64

[krockerman]

11 hours ago, Lucasxp64 said:

What kind of prompts you are using? Those common LLMs out there are too aligned with the purpose of being a "Helpful assistant" speech that it's hard for me to get it answering like an actual woman. But I believe if we had a lot of high quality conversation examples it could get better. They don't offer enough resistance and roleplay capabilities are lobotomized, they tend to be bad at narratives with conflict. They sound like some badly written roleplay.

The best results I had were with some uncensored models that were weakly aligned, they were closer to an auto-complete model than a "helpful assistant" alignment, Like "Smol LM 2 1.7b Instruct". https://huggingface.co/HuggingFaceTB/SmolLM-1.7B-Instruct Btw, use on the SYSTEM prompt how it should align, the user prompt is weaker at this.

But the 360M parameter model funny enough gave me better roreplay because it's less aligned... Maybe because it's dumber it works more like an auto-complete which tends to bias wildly how it answers based on the input text, which is good to give answers outside of that "helpful assistant" lobotomized crap. But it's too dumb to be useful, I don't recommend.

I'm still looking into other models that are larger. I could also set up in Kaggle (30h guarantee gpu uptime per week for free, an alternative to Google Colab) and use their Nvidia Tesla P100 (16GB) or T4 (16GB x2 SLI), but it's better if it can run on a CPU because I don't have a lot of GPU allowance. 

I have been working on some prototypes. It involves alot of extra promts tho