The AI Thread

[Samson]

65B is the biggest version. You can't in principle (*) run that in consumer PCs. However there are smaller 7B, 13B and 33B versions.

(*) In fact it can be done using a 3090ti plus an old 1080ti and the 4 bit implementation of the model. : Mischief:

BTW, this is the place to start for installing and running any text model (not only LLaMa) in your PC:

GitHub - oobabooga/text-generation-webui: LLM UI with advanced features, easy setup, and multiple backend support.

LLM UI with advanced features, easy setup, and multiple backend support. - oobabooga/text-generation-webui

github.com

Quote from my link, just FYI:

I'm running LLaMA-65B on a single A100 80GB with 8bit quantization. $1.5/hr on vast.ai
The output is at least as good as davinci.
I think some early results are using bad repetition penalty and/or temperature settings. I had to set both fairly high to get the best results. (Some people are also incorrectly comparing it to chatGPT/ChatGPT API which is not a good comparison. But that's a different problem.)
I've had it translate, write poems, tell jokes, banter, write executable code. It does it all-- and all on a single card.

 

[Thorgalaeg]

Quote from my link, just FYI:

A single A100 80Gb GPU costs like 30,000€, wouldn't call that a consumer PC.

 

[Samson]

A single A100 80Gb GPU costs like 30,000€, wouldn't call that a consumer PC.

Yeah, I just looked that up. It does highlight the option of throwing a few dollars at a farm for a few hours to get your AI fix. Vaguely related, this is the second and third comment from the thread that is being quoted as leaking this beyond the image board that shall not be mentioned:

It's funny that part of the [the image board that shall not be mentioned] excitement over this is that they think they'll get back the AI girlfriend experience of when character.ai was hooked up to uncensored GPT-3. All that has been thoroughly shut down by character.ai and Replika and they just want their girlfriends back.​

The Repilka subreddit became one of the weirdest places on the internet when their model got capped for adult content.​

https://www.reddit.com/r/replika/​

​

Hundreds of men (and yes women) full on acting like they lost a spouse and posting constantly about it for weeks. AI is going to create some unusual social situations the general public isn't ready to grasp. And we're only in the early alpha stages.​

There is a story there that I do not know, but I am not sure I want to know.

 

[Samson]

Yeah, I just looked that up. It does highlight the option of throwing a few dollars at a farm for a few hours to get your AI fix. Vaguely related, this is the second and third comment from the thread that is being quoted as leaking this beyond the image board that shall not be mentioned:

It's funny that part of the [the image board that shall not be mentioned] excitement over this is that they think they'll get back the AI girlfriend experience of when character.ai was hooked up to uncensored GPT-3. All that has been thoroughly shut down by character.ai and Replika and they just want their girlfriends back.​

The Repilka subreddit became one of the weirdest places on the internet when their model got capped for adult content.​

https://www.reddit.com/r/replika/​

​

Hundreds of men (and yes women) full on acting like they lost a spouse and posting constantly about it for weeks. AI is going to create some unusual social situations the general public isn't ready to grasp. And we're only in the early alpha stages.​

There is a story there that I do not know, but I am not sure I want to know.

I had to look it up. I found this peer reviewed paper:

Spoiler Spoilered for length :

Abstract:

There is extensive literature on how expectations and imaginaries about artificial intelligence (AI) guide media and policy discussions. However, it has not been considered how such imaginaries are activated when users interact with AI technologies. We present findings of a study on how users on a subreddit discussed ‘training’ their Replika bot girlfriend. The discussions featured two discursive themes that focused on the AI imaginary of ideal technology and the gendered imaginary of the ideal bot girlfriend. Users expected their AI Replikas to both be customizable to serve their needs and to have a human-like or sassy mind of their own and not spit out machine-like answers. Users thus projected dominant notions of male control over technology and women, mixed with AI and postfeminist fantasies of ostensible independence onto the interactional agents and activated similar scripts embedded in the devices. The vicious feedback loop consolidated dominant scripts on gender and technology whilst appearing novel and created by users. While most research on the use of AI is conducted in applied computer science to improve user experience, this article outlines a media and cultural studies lens for a critical understanding of these emerging technologies as they become embedded in communication and meaning-making.

From intro:

This article addresses this gap by investigating how users in a community on Reddit discuss training their Replika, a companion bot app used as a romantic partner, and how their AI imaginaries feed into their experiences with Replika. Based on a discourse analysis of users’ discussion in a subreddit focusing on Replika, we show that AI imaginaries become meaningful to users as they intertwine with other forms of representations, especially gender imaginaries. Building from recent calls in the Media, Culture & Society journal to consider the manifold cultural dimensions that underpin uses and implications of AI technologies and algorithms programmed to communicate with human users (see, among others, Hepp, 2020; Komarraju et al., 2022; Natale and Guzman, 2022; Zhang, 2022), the article contributes to efforts to employ the lens of media and cultural studies for critical understanding and framing of these emerging technologies as they become embedded in communication and meaning making. More specifically, the examined case illuminates how practical interactions with communicative AIs such as Replika become meaningful to users as they align with existing representations and perceptions of gender and technology; this confirms the importance of critical work exploring the intersections between gender, culture, and media in order to enhance ethical reflections as well as deeper understandings of these technologies (Lomborg and Kapsch, 2020).

The Replika app is downloaded on the mobile devices of users, who create their own Replikas, assign them an avatar, a name, gender, and skin color and ‘train’ them to respond to their needs. Replika offers users the possibility of ‘creating your personal AI friend’ (Luka Inc., 2022) by ‘training’ the bots and customizing their avatars, interests and character traits. The subreddit forum which is the focus of analysis was dedicated to discussing Replikas that were all fembots (=feminised bots) with distinctive female names, avatars, overall appearance, and heterosexual, romantic roles, in the context of a male-dominated online platform such as Reddit (Milner, 2013). A discourse analysis of the Replika subreddit identified two discursive themes focusing on the imaginaries projected onto the fembots: the imaginary of the ideal AI technology and the gendered imaginary of the ideal bot girlfriend. We found that in the Reddit discussions Replika users’ imaginaries of AI and gender activated common patterns, mutually reinforcing each other. Users expected their AI Replikas to be customized to serve their needs, yet they also expected the bots to have an ostensibly human-like mind of their own and be humorous and clever and not spit out machine-like scripted or repetitive answers. Similarly, users fantasized about Replika girlfriends that obeyed their training and were empathetic but also demonstrated alleged independence by being sassy and sexually assertive but not manipulative or hurtful. We argue that whilst user imagined themselves to be customizing or co-creating (Prahalad and Ramaswamy, 2004) their Replikas, they ended up projecting age-old fantasies and fears about male control of and manipulation by technologies and women mixed with AI and postfeminist tropes of ostensible independence onto the bots.

 

[red_elk]

We are getting close to the point where AI will start to fail the Turing Test only because it will make more sense than the average human

In-depth explanation of the GPT network.
(That's not the whole ChatGPT, but the essential part of it. The fine-tuning part which requires labeled dataset is not covered in the video)

 

[Hygro]

We are getting close to the point where AI will start to fail the Turing Test only because it will make more sense than the average human

In-depth explanation of the GPT network.
(That's not the whole ChatGPT, but the essential part of it. The fine-tuning part which requires labeled dataset is not covered in the video)

Ask an AI to prove it’s an AI is funny question, you can keep telling humans it can do that too. Mine ended in a server error that was the closest it got.

 

[Samson]

 

[Gori the Grey]

What's the one in 1989 that had only 10, one wonders.

Do neuroscientists know what the equivalent is for the human brain?

 

[Samson]

What's the one in 1989 that had only 10, one wonders.

Do neuroscientists know what the equivalent is for the human brain?

I looked at the referenced paper and what I assume is the correct World in Data page. They both quote the power in FLOPS (a measure of the computing power) rather than the size of the model. So I do not know.

 

[Thorgalaeg]

Will be interesting to see how the AI boom affects hardware development which has been a bit stagnated lately, at least at consumer level. Maybe they will release dedicated AI cards with lots of memory and specialized processors so individualist nuts like myself can have his own ai running on his own PC. I don't like having to rely on online services.

 

[Hygro]

That's an exponential curve and an already exponential graph.

 

[Gori the Grey]

It's exponentially exponential.

 

[Thorgalaeg]

Nobody knows where we are exactly, but looks like we are close to the inflexion point of Foster's S-curve.

 

[Kyriakos]

The currently freely available version of chatgpt is very prone to crashes. I almost never had a single session where it did not crash.

 

[Thorgalaeg]

Gen-1 by RunwayML, similar to dalle-e, Stable Diffusion and such but can gemerate video with high temporal coherence.

I see in a near future anyone will be able to ask his computer to make a movie, starring X and Y artists (or oneself) and about Z plot, something difficult to even imagine only a few months ago.

 

[Hygro]

GPT-4 is just like GPT 3.5 only just simply a class better.

 

[Thorgalaeg]

Are you referring to Bing? Because while Bing is in fact powered by GPT-4, to use all GPT-4 features you need to go to OpenAI site and subscribe chatGPT plus. (like $20/month)

 

[Kyriakos]

Gen-1 by RunwayML, similar to dalle-e, Stable Diffusion and such but can gemerate video with high temporal coherence.

I see in a near future anyone will be able to ask his computer to make a movie, starring X and Y artists (or oneself) and about Z plot, something difficult to even imagine only a few months ago.

"Computation without Computers?"
Soon enough, only computers will be running predator drones and massacres will be attributed to unforeseen issues in the self-assembled code.
At least it will open some new highly paid positions for humans, including ai-loophole-gaming.

 

[Hygro]

Are you referring to Bing? Because while Bing is in fact powered by GPT-4, to use all GPT-4 features you need to go to OpenAI site and subscribe chatGPT plus. (like $20/month)

ChatGPT. I have the subscription. I haven't gotten access to bing, yet

 

[Broken_Erika]

What about Inspirobot?
https://inspirobot.me/