This is a topic that is starting to get a lot more attention.
Artificial intelligence, specifically generative AI, is on the minds of many. We cannot hit up any tech related page without seeing it. The developments are quickly being incorporated into applications we utilize on a daily basis. Unfortunately, so far, the results are mixed as Google showed.
There are a number of concerns regarding this technology. Leaving the "it will kill us" aside, the major problem is the massive undertaking required to produce these models. They require a great deal of compute which means a lot of funding is required. Naturally, this feed into Big Tech.
We also get the open versus closed debate. While most in crypto would likely fall on the side of open source, we see some issues there. The number of open source models is expanding yet there is a major challenge. They are being released by mega tech companies. When the likes of Meta are behind it, that could be an issue.
Smaller players are having difficulty gaining traction. This stems from the fact compute is required, something that is affecting everyone. Another growing problem is the need for data, something start ups are really confronted with.
Could cryptocurrency be the solution?
Image generated by Ideogram
Cryptocurrency As The Driver Behind Open Source AI
If you want to see open source AI, who do you want to develop it?
We can most likely agree that Meta opening up Llama is a good thing. However, what about competitors in the open source realm? We can appeal to the likes of Elon Musk who opened up Grok, at least the weights (I am not sure if he opened everything up).
However, that is not an enviable position either.
The key is going to be for the open source community to create their own LLMs along with other technologies. This could be aided by the fact that compute gets less expensive over time. For this reason, in a year, the same models can be done on smaller systems.
Of course, the bigger entities have the same, if not a larger, advantage.
Even still, regardless of the amount of compute required, there is still a cost. Here is where crypto can step in.
Mining Pools
Do you remember when there was a lot of fuss about Bitcoin mining and the energy it used? Why was that the case?
The reason behind it was the fact that crypto mining requires high powered, advanced computers. They are driven by GPUs, something that aided Nvidia's stock. Today, data centers are requiring enormous amounts of compute, something miners already have.
Could mining pools aid in this endeavor by providing compute for AI purposes?
The only difference is the data that these machines are running. Compute is compute. To the system it is secondary to the purpose of calculations. Computers process data. It is similar to your laptop. It does not care if you are saving a text document or watching a video. Sure, the latter requires more processing power and memory but it is still simply a process it is designed to handle.
There were already a number of deals signed by crypto mining companies for this purpose.
Funding Of AI Models
Another area where cryptocurrency could apply is in the area of funding these operations. Through the capital formations, there might be a solution to a major problem.
Here is an idea of the type of money that we are dealing with:
Any pre-training cycle for a 20 billion-plus parameter model could cost between ten to a hundred million dollars and involves a multi-month process with many failed attempts.
What we would be looking at is a circular nature of funding and revenues. If we look at this as infrastructure, we can see how a new incentivization mechanism is applied.
There is little doubt that Web 3.0 requires AI. Nothing in the digital realm (and likely the physical) is going to exist without it. Therefore, it is in the best interest of the entire industry to get into this.
Suppose there is a need for $25 million for the development of an open source model. This could be done through tokenization in the form of a DAO. Here we see the money raised with the token holders having a stake in it. This can come from numerous sources around the world.
With such a set up, the license associated with the model requires a portion of the revenues generated to be put back into the DAO. This means any company that is utilizing the model contributes to the payouts.
It is likely that many Web 3.0 initiatives set up using the model will be tokenized. We can envision a process where the tokens from those are flowing into this DAO.
The DAO could go a couple different ways with the money it receives. Obviously, there could be distribution to the token holders. It is probably this happens in part. The other is to keep funding other training models.
Here is how the open source realm expanded. We keep feeding in value accrued by development using the models, value that is captured through tokenization.
A Data Solution
The other piece of the equation is data.
Here is a recent situation that is only starting to get attention. In spite of 30 years of people typing away, there appears to be a data shortage coming up.
What this means is companies are getting very stingy with it. Naturally, this helps to maintain a closed system since only the larger players can afford to give the likes of Reddit $60 million. The democratization of data is still not occurring.
Some might share their data but I would say the majority are not. Closed models certainly are not sharing what they are using. Of course, OpenAi keeps getting itself into hot water over the accusation it is taking copyrighted material. With Microsoft's backing, it can afford to fight/delay these suits.
The same is not true for smaller entities.
Here again, blockchain can be a solution. Not only is it a good solution to counter the increasing control of centralized AI platforms by Big Tech, it can also be the feeder into the data used to train.
Open access to data is something that blockchain can provide. In fact, the first major problem that blockchain might offer is the "democratization of data". This should be a focus of all.
It no longer is merely whether an account gets closed or not. Posts being taken down is a negative of centralized social media entities but, in the grand scheme of things, isn't the end of the world.
What is crucial, as we advance deeper into this realm is the fact that data is not democratized. That is the first challenge of Web 3.0.
Couple that with creative funding mechanisms through crypto and we can see how all of this can present a viable alternative to major tech.
Posted Using InLeo Alpha