Deadcow is trying to introduce everyone in this next AI wave we are living, creating a decentralized AI cloud where everyone benefits.

What is DeadCow?

Let’s think about it “normal“ terms. Deadcow is like Airbnb’s for your phone where you lend your phone unused computation and in exchange you will get money without doing anything besides having the app in your phone.

Why DeadCow?

There is severals why’s.

1. For devs and companies: Cheap, fast, scalable and secure.

2. For light node operators: Make money out of your idle device.

Deadcow really focuses on the decentralization to make the future cloud economy not only secure for diverse institutions who manage sensible data such as healthcare, finance,etc. But also really focusing on making a whole new wave of people benefit from it.

Besides all said it also takes advantage of devices already existing which also helps out the environment.

How it works?

Decentralized layer where tokens will be distributed to the users who provide nodes where the inference to run.

There will also be used ipfs to store the models and assigned them to the different nodes depending the distance to provide the best inference quality possible and also avoiding have all the blockchain (all models) in all nodes making it not scalable at all.

Inference framework that will be used at the nodes where it will get assigned tasks depending on the location to provide a high speed, low latency service.

Why blockchain?

We believe that taking advantage of the advantages of the blockchain

• Security and privacy for both node providers and users.

• Decentralization by not having a central entity looking and your data and deciding something for you.

• Incentives for people to lend their devices and make it sustainable

Use cases

There is not limit, but at Deadcow we believe “small” or medium models will be widely use because you will be able to train, fine tune them and have actually a personal AI for your specific case which some times is hard to get for really specific situatios to achieve with Large Language Models.

Challenge and future goals

The main challenge of the first version of DeadCow is getting the first nodes to trust the network and then regarding technical challenges are managing the hardware limitations of the different devices by integrating the different parts of deadcow.

Challenges

• Network latency

• Different type of devices and software

• Assignment of nodes focusing on proximity for scalability

A lot more challenges too, but the ones mentioned are the most difficult ones i would say.

Future Goals

A lot of them, but let’s list the main ones

• Train models by letting the users and companies train and fine tune models

• Parallelazation of big models by letting small nodes such as phones to have parts of the LLM when running the inference.

• Add support for GPU and other specific devices made for AI such as ASICS to be able to run for bigger models and giving the end users other capabilities.

Stay in touch at deadcow.xyz for the launch.
Lautaro.