There are two AI projects on HIVE. One of them has already been released and the other one is being trained with HIVE data that was likely not part of any other existing LLM projects. 
The Narrative vs Reality
If you have learned anything about how "AI" work behind the scenes, seeing a video like this should be somewhat hilarious. Technological infeasibility does not invalidate the narrative. The unfortunate reality is that most of the time, it is the underlying technology that take a backseat while the narratives run rampant. Eventually this leads to bubbles.
Bubbles can be extremely lucrative as long as the exit is made before the eventual crash. The crash is nearly impossible to time and the way down is usually a lot faster than the way up. Listen to what Yann LeCun, a computer scientist working in machine learning, computer vision, mobile robotics and computational neuroscience. See the level of hype he is making about LLMs.
Our goal should be to provide utility to more users and bring in more activity and exposure to HIVE. We do not have to reinvent the wheel! We don't even need to change the focus outside of what we are doing on HIVE. We can assimilate these "AI" features into the DAPPs we already have.
Top 10 AI Research Papers of 2024
The number of papers published doubled when compared to 2023. We have two more weeks left in 2024. The most surprising part is Meta coming out on top with full guidance to create a state of the art Large Language Model. Mark Zuckerberg is embracing open source at least when it comes to LLMs.
The Simplest AI Integration
The size of the audience that will use AI generated images using diffusion models shrinks when they have to pay $20 per month for the service. Not everyone can justify that expense base on their usage and income. The free image generations are not that great with their results and they could still face restrictions and waiting times to use these images. 
These AI images can be accesses straight from the publishing UI while users are writing their articles. Alternatively they can visit here and start using these 11 AI models by spending few cents per image generation. They could generate 100+ images using some of the best models and only spend a tiny fraction of what they would have spent as a subscriber. These are the available models:
Perfect Use Case for HBD
The only other comparable option is stablecoin payments of EOS, Tron and other DPoS blockchain that came after HIVE's first of its kind feeless model. If you are new to cryptosphere, you may not know that 
HBD Stats by 
Hive Dollar Monitor by 
INLEO is Training Simba
The name of the AI might change at release. 
Processing is a very expensive bottleneck. Not only LLMs are expensive to train, using a large model is also very expensive. I cannot use anything bigger than a 7B model locally. This is a tutorial from a previous Microsoft employee that show how to run these models locally:
If Ollama seems too daunting due to the Terminal, there are front ends to get the job done in an even more user friendly manner. LM Studio is one such option that is available for Linux, MacOS and Windows.
Once Simba (or whatever the final name will be) is launched, Premium subscriber will have an even easier time accessing an AI where it comes with a host of other additional features. This list does not even mention that there is an extra 1% APR for delegations for 
Embracing AI is Good Marketing
The best part is that we don't have to distract ourselves too much. Using $HBD micropayments for powerful AI models via a HIVE front end should not be very difficult.