Support NeoGAF

[Bluecondor]

This article from Reuters claims that a company (Databricks) is releasing an open-source chatbot that will be comparable to what we see in ChatGPT, but can be trained on much smaller datasets and will require far less computing power:

https://www.reuters.com/technology/...atbot-cheaper-chatgpt-alternative-2023-03-24/

Databricks CEO Ali Ghodsi said the release was aimed at demonstrating a viable alternative to training a kind of AI model called a large language model with enormous resources and computing power.

"The future will be that everyone has their own model, and they can actually train it, and they can make it better," Ghodsi said. "And that way, they also don't have to give away their data to someone else."

"My belief is that in the end, you will make these models smaller, smaller and smaller, and they will be open-sourced," Ghodsi said. "Everyone will have them."

--------------------------------------

So, if I am following this. Let's say you wanted to make chatbots based on the characters from the Harry Potter books. This AI model could be trained on the Harry Potter books/movies as the dataset. Then, chatbots on Harry Potter characters could respond based on this AI model that is trained on the Potter book dataset.

For those of you who know way more about this than me, am I correct in my Harry Potter example?

If so, this just makes sense. The next Harry Potter video game could (possibly depending on computing requirements) make it so that you could have more natural conversations with Potter characters on a wider range of topics. But, at the same time, you won't be able to get the characters to talk about non-canon topics, offensive topics, etc. (LOL, not that people won't try to get the characters to say everything under the sun and record it).

Anyway, this sounds really promising. Those of you with AI and/or game development, please set my expectations straight.

 

[Miyazaki’s Slave]

PLM (private or personalized) language models have a wide range of application.

Depending on the compute requirements and underlying systems (of the language model and future software) you could use your PLM in a game for highly customized NPC interactions like you are describing.

Will most likely be tied to things like your google/apple/Instagram/twitter SSO are today. They become a part of your online identity/user account.

 

[A.Romero]

Those are tall promises. Not even Amazon is putting out a LLM to compete with ChatGPT and they have all the resources they might need.

 

[K2D]

Any bets on what will happen first?..:

NPC's with machine learning dialog, or

dialog distributed ingame by live streamed AI chat..?

 

[Bluecondor]

PLM (private or personalized) language models have a wide range of application.

Depending on the compute requirements and underlying systems (of the language model and future software) you could use your PLM in a game for highly customized NPC interactions like you are describing.

Will most likely be tied to things like your google/apple/Instagram/twitter SSO are today. They become a part of your online identity/user account.

That sounds cool. So, essentially our internet/social media postings will shape our PLMs then? If my PLM is in a game, my character would use phrases and ideas from my postings. That will be really engaging.

 

[Bluecondor]

Those are tall promises. Not even Amazon is putting out a LLM to compete with ChatGPT and they have all the resources they might need.

Ya, you're right. Man, I always fall for these bold promises with AI. The Amazon comparison is so logical.

 

[Bluecondor]

Any bets on what will happen first?..:

NPC's with machine learning dialog, or

dialog distributed ingame by live streamed AI chat..?

So, I don't know if I am following everything, but is the AI model described in the article I linked something that could be used to generate machine learning dialogue within a game, whereas the dialogue distributed in-game by livestreamed AI chat something that would come from large LLMs on platforms like Playstation +, Xbox Live and Steam?

If this is what you are suggesting, then you raise an interesting possibility. The company platforms that currently make it possible for us to download games and play games online could evolve to hold these large LLMs that game developers could then utilize in games played on the platform. This would also make it easy to control the type of content in the game (i.e. offensive content, etc.).

I think you've convinced me that company platforms having large LLMs could be something that we see sooner, rather than later.

 

[Sakura]

This article from Reuters claims that a company (Databricks) is releasing an open-source chatbot that will be comparable to what we see in ChatGPT, but can be trained on much smaller datasets and will require far less computing power:

https://www.reuters.com/technology/...atbot-cheaper-chatgpt-alternative-2023-03-24/

Databricks CEO Ali Ghodsi said the release was aimed at demonstrating a viable alternative to training a kind of AI model called a large language model with enormous resources and computing power.

"The future will be that everyone has their own model, and they can actually train it, and they can make it better," Ghodsi said. "And that way, they also don't have to give away their data to someone else."

"My belief is that in the end, you will make these models smaller, smaller and smaller, and they will be open-sourced," Ghodsi said. "Everyone will have them."

--------------------------------------

So, if I am following this. Let's say you wanted to make chatbots based on the characters from the Harry Potter books. This AI model could be trained on the Harry Potter books/movies as the dataset. Then, chatbots on Harry Potter characters could respond based on this AI model that is trained on the Potter book dataset.

For those of you who know way more about this than me, am I correct in my Harry Potter example?

If so, this just makes sense. The next Harry Potter video game could (possibly depending on computing requirements) make it so that you could have more natural conversations with Potter characters on a wider range of topics. But, at the same time, you won't be able to get the characters to talk about non-canon topics, offensive topics, etc. (LOL, not that people won't try to get the characters to say everything under the sun and record it).

Anyway, this sounds really promising. Those of you with AI and/or game development, please set my expectations straight.

Technically this is doable, but running the models on a local machine, while also having to run the game would require pretty high end hardware. Not to mention it would be very hard (at least currently) to make sure they don't say anything non-canon.
I think it would certainly have applications in some games in the future though.

 

[A.Romero]

Ya, you're right. Man, I always fall for these bold promises with AI. The Amazon comparison is so logical.

It's pretty easy to get carried away with the hype (it happens to me often too) but I'm in this line of business and just a few days before I had a chat about it with a product manager from AWS and it seems they don't have plans to compete here or at least not yet.