One of the most exciting [features](https://inleo.io/@leoglossary/leoglossary-feature) discussed for the [LEO](https://inleo.io/@leoglossary/leoglossary-leo) platform is Leo [AI](https://inleo.io/@leoglossary/leoglossary-artificial-intelligence).  This is something that has enormous potential.  


In this [article](https://inleo.io/@leoglossary/leoglossary-article) we will deal with some of the concepts pertaining to it along with some of the evolution that we could see over [time](https://inleo.io/@leoglossary/leoglossary-time).

 https://img.inleo.io/DQmbVEKX949DtsocMq5Q7bi9HKxrfoMVFHpPM76fgyiBtQH/image.png 
Image generated using Ideogram


# LeoAI: Making Leo A Focal Point


A lot of the discussion pertaining to AI is speculative due to the uncertainty of where things are heading.  We have a sense of the general direction yet how things completely evolve is uncertain.


The best way to think about this is as a Small Language Model (SLM) that is trained using a Large Language Model (LLM).  


It is the LLM that provides the general knowledge.  In other words, it was trained on all the [data](https://inleo.io/@leoglossary/leoglossary-data) from the [Internet](https://inleo.io/@leoglossary/leoglossary-internet).  For the sake of this article, we will simple refer to that as [ChatGPT](https://inleo.io/@leoglossary/leoglossary-chatgpt) although it is possible another model is used.


What is the potential of LLMs?  Here is where we have to do some forecasting.  


One of the beliefs is that LLMs are a thread to [Google](https://inleo.io/@leoglossary/leoglossary-google-company) search.  When looking for information, most turn to a [search engine](https://inleo.io/@leoglossary/leoglossary-search-engine).  The belief is LLMs such as ChatGPT could take some of this traffic away from there.  It is as easy to enter a prompt as a search.


Of course, there are still some issues with accuracy regarding the LLMs, something that will improve as more training is done.  


Here is where we see the first benefit to LeoAI.  


If we are indeed able to treat it like ChatGPT, we can do "search" without ever leaving InLeo.  We were told there will be a page built similar to how [front ends](https://inleo.io/@leoglossary/leoglossary-front-end) of LLMs such as ChatGPT and [Gemini](https://inleo.io/@leoglossary/leoglossary-gemini). 


This would have a host of benefits such as time on site, number of pages, and queries.  All of this adds to the usefulness of Leo.


## Clients


One of the things mentioned is the idea of [building](https://inleo.io/@leoglossary/leoglossary-building-structure) a client similar to copilot.  What this means is we can get the ability to access it when writing articles.


Grammerly is one [application](https://inleo.io/@leoglossary/leoglossary-application) that we can parallel.  The difference is the information provided will be a great deal more knowledgeable.


LeoAI is also a SLM.  This is being trained on Hive data.  Anything on the [blockchain](https://inleo.io/@leoglossary/blockchain) is being fed into the model which will provide it additional data as compared to the LLM.


The client will offer suggestions to the [writer](https://inleo.io/@leoglossary/leoglossary-writer) in an effort to improve the articles.  This will not offer full text writing in the sense of one could prompt a full [post](https://inleo.io/@leoglossary/leoglossary-post-hive).


LeoAI will also provide information based upon the activity and results pertaining to our [on-chain](https://inleo.io/@leoglossary/leoglossary-on-chain) activity.  For example, one could query which articles have the most views or which was most popular.  This could help with topics that ranked well as an example.


Basically anything that is on chain will be fed into the model.  This is accessible view LeoAI.


## Multimedia


For the time being, the focus is upon text.  


This is only one aspect to AI.  We are seeing the same generative [technology](https://inleo.io/@leoglossary/leoglossary-technology) applied to [images](https://inleo.io/@leoglossary/leoglossary-image) and [video](https://inleo.io/@leoglossary/leoglossary-video).  Here is where some major insight can be garnered regarding the potential.


Here is something we must stress:  The [day](https://inleo.io/@leoglossary/leoglossary-day) LeoAI goes live, that is the worst it will be.  Each generation of training only improves the model.  That means we are going to see continued improvement as people utilize it more often.


What about additional features?


Here is where things can change a great deal over the next couple [years](https://inleo.io/@leoglossary/leoglossary-year).  It is also where we are dealing with a bit less clarity.


One thing we can safely conclude is that image generation is another aspect that could be integrated.  We know this is starting to improve a great deal with products such as Ideogram.  Could that be incorporated into LeoAI?


Consider the potential to write an article and then generate an image using LeoAI.  This all could be done internally.


As time passes, the idea of text-to-video might become more likely.  Could this be added at a date in the future?


A lot of this is going to depend upon the [cost](https://inleo.io/@leoglossary/leoglossary-cost) of training these model and that comes down in the future.  I saw one who speculated that a model like Llama3 could be trained for around $10K in a couple years.  We will see how this progresses with the different technologies.


## Winner Take Most


This is the final point to consider.  


Technology is often a winner take most proposition.  Again, if we look at search, Google dominates.  While there are many other options, nothing has threatened its dominance.


What does LeoAI do in this regard?


While it is not going to rank on the level of the LLMs, could it become a leader on Hive?  


Remember, that much of what determines the capabilities is the amount of compute, data, and the amount of training.  It gets harder to catch up over time.


This is why getting it live is important.  We also have to look at adding more to the data.  This can be done on-chain.  There is another factor.  The prompts help to [train](https://inleo.io/@leoglossary/leoglossary-train-railroad) the system.  In other words, LeoAI improves due to the data generated by using LeoAI.


The idea is to get a feedback loop going that enhances each layer that LeoAI touches.  Here is where the multi-dimensionality of the platform enters.  The data from something such as [LeoDex](https://inleo.io/@leoglossary/leoglosary-leodex) can also be fed into the model, helping to further train it.


These are some of the possibilities that we could see unfolding.  Right now, we are speculating a great deal, providing estimates based upon general trends in AI.


We will see how it is refined to fit Leo.


____


[What Is Hive](https://inleo.io/@leoglossary/leoglossary-what-is-hive)


Posted Using [InLeo Alpha](https://inleo.io/@taskmaster4450le/the-potential-of-leoai-jm)