Developers
Applications
Blog
Events
ForumCreate a Bot

GNUS.AI & Paloma: Harnessing the Power of Blockchain, AI, and GPU Access

Overview

The Paloma Chain flies high as it enters the blockchain AI space! In this week’s AMA episode, the host, Taariq Lewis @LewisTaariq, CEO of VolumeFi (volume.finance)  is joined by GNUS (GNUS.AI)   CEO Ken Hurley, and CTO/Co-founder Brent Arias to introduce their company and their role in AI and blockchain.

As the demand for GPU access grows at phenomenal levels, GNUS harnesses GPU resources from the collective power of millions of mobile phones by allowing its apps to participate in a network of decentralized physical infrastructure. Device owners and game developers monetize video games and earn tokens in return by utilizing the mobile phones’ GPU functionality to perform advanced AI/ML jobs such as video inference and smaller model training. Developers access the platform by integrating the GNUS SDK into their app and paying end-users with GNUS tokens

The potential for a partnership with GNUS and other GPU providers was also discussed where Paloma can integrate a frictionless, seamless cross-chain execution and a smooth UI/UX while eliminating the worry about token price, slippage, and price volatility.

Palomabots have zero fees for the remainder of May enabling users to do TWAP, limit orders, and leverage trade with no fees.

Definitions

  • AMA - Ask Me Anything

  • GNUS.AI - empowers end-users to participate in decentralized processing, earning GNUS tokens by utilizing the idle cycles of computers, mobile devices, and any other device connected to the internet.

  • DePIN - or Decentralized Physical Infrastructure Networks are blockchain protocols that build, maintain, and operate infrastructure in the physical world in an open and decentralized manner.

  • ML - or Machine Learning, is an area of artificial intelligence, where machines predict tasks based on previous data. It is a type of statistical algorithm that can learn without definite instructions and enables it to do certain tasks and duplicate human cognitive activity.

  • Model Training - the process of feeding curated data to selected algorithms to help the system refine itself to produce accurate responses to queries. Model retraining refers to updating a deployed ML model with new data.

  • Inference - the process that a trained machine learning model uses to conclude brand-new data. It is an AI model capable of making inferences without examples of the desired result.

Intro

The host welcomes everyone to Volume’s Twitter Spaces. He explains that there is a shift in today’s topic from creating Palomabots using OpenAI, to learning more about GNUS and their vision of blockchain and AI. He extends a warm welcome to today’s special guests from GNUS.AI.

Taariq shares a brief about Volume.

“We do decentralized AI bots and custom AI models and use these for security and privacy. We build bots on the blockchain and use them to take advantage of AI models and AI model training.”

As the DeFi market slows down, Volume is opening up to opportunities in the blockchain AI space for more community and growth.

“We want to foster the building of more decentralized AI bots and help build up more customized AI models and generate revenue and profits for Paloma validators.”

Zero Fees on All Palomabots

Palomabots have zero service fees for the whole month of May.  Users can do TWAP,  limit orders, and leverage trade with no fees.

“If you're looking to take leverage positions on Bitcoin, you will be up and you can leverage Paloma and our Volume bots to do so.”

Paloma has successfully recovered the 12ETH that was stuck on the old Messenger Chain. The team is now working on helping Paloma developers use AI to build Paloma bots.

About GNUS

Taariq asks the guests to introduce GNUS.AI and their role in AI.

Brent offers a brief background on GNUS.

“Ken, our founder, was looking at ways to monetize video games that would surpass the feeble approaches that have been used for many years. The concept is that the developers wouldn't have to go through this enormous gymnastics to introduce monetization in their games. They could have it as a freebie by integrating an SDK, allowing their app to participate in a network of decentralized physical infrastructure.”

From the seed of that idea, and with Ken’s expert knowledge of GPU capabilities, the premise was the processing would be done by GPU. In the era of blockchain and recognizing the value of decentralization, this concept was put into a Web3 context.

He explains that beyond video games,  any application that integrates the SDK benefits from this capability. The trifecta of a vendor that has integrated this SDK can be earning crypto with their app, with processing time shaved off to contribute to the user of a device that has opened that application.

“They're earning crypto because their device is being used. Somewhere a buyer who wants to get the ML and AI processing done is getting a lower price because the infrastructure that supports that processing is not our infrastructure.”

While there is a substantial surge in demand for ML, there is a shortage of supply for processors for this kind of work. With a billion devices potentially available with apps running on them,  GNUS SDK alleviates that problem using a horizontal scaling approach.

Taariq notes that this is where DePins touches AI and blockchains. He cites the game Temple Run played on a user’s mobile phone as an example of how GNUS works.

“There's a game publisher, and while I'm playing Temple Run, I can get rewards. One of those rewards I might get is tokens from my publisher generated from GNUS SDK, which allows my phone and its GPU functionality to be used and run machine learning training jobs.”

He adds that with billions of mobile phones, a massive network of decentralized physical infrastructure can run GPU jobs. The mobile phone user in return gets tokens from the utility of both a free game, and a free published app that yields tokens.

Volume has been looking for companies that can customize or retrain a massive model by finding a small subsection of parameters to help fine-tune and customize models for users. He asks GNUS what model sizes can be taken up by their network with the view that Volume is keen on training and fine-tuning smaller models.

Brent replies that this is determined by the level of parallelization that the algorithms will permit. Some algorithms permit 20% of the endeavor to be parallelized while others 80%.  He notes that although memory can be a challenge, GNUS intends for solutions to be based on streaming technology to minimize the amount of memory needed at any given point.

Ken chimes in to say that GNUS does not retrain very large models because they may not fit in mobile memory easily.

“But that's one of our goals. We’re probably not going to handle large language models because the memory footprint on those is just ridiculous. We can do some training, but not large language models.”

Taariq asks what kinds of jobs GNUS focuses on.  What is their sweet spot?

Ken explains that they are mainly working with high-speed video inference in partnership with a couple of companies, although this has not been announced.

“The retraining part is our focus and almost a hundred percent inference stuff can be done on our system. Inference is just using the training dating data to figure out how to process the AI and do face recognition, for instance.”

In general, GNUS assures that they can do about 90% of model training on smaller language models that do not require a lot of memory.

The Docs at GNUS.Ai is a starting point for developers to learn more about their process. GNUS SDK is nearly finished and up for testnet. Their entire source code is open source on GitHub.

“The beauty of our system, I call it negative friction monetization because to integrate with our system, all you do is link to our SDK and you're done. And then the app's off to the races. You don't have to do anything.”

GNUS Tokens

Taariq asks how the GNUS token relates to the protocol itself.

Ken explains that the protocol token is basically a DePIN crypto. It is generally like a compute infrastructure with a crypto token payment system.

“It's the value, the real-world asset of the GPU resources is what the token is used for. You pay in the GNUS tokens that create a job for you to do your AI. We have mobile neural network shaders and all that and flow. So you put all that information, pay for it, and then it gets sold out to the network.”

He adds that users who need to fine-tune or execute inference on an image can send the job over to GNUS.

“We basically pay out in digital goods, so we don't have a conversion factor. We can keep the price low and even lower as time goes on. There's never going to be a time where you get to saturate the network and it's going to cost more.”

Use Cases for GNUS

Ken explains that GNUS is the DePIN part of the infrastructure. Their customers include game developers and partners who want to integrate with GNUS and use AI for machine learning and analytics. Others use the network for personal projects in research and similar endeavors. They are currently working on a deal that will be providing hundreds of thousands of nodes, and utilizing small portions of it for video analysis. He underscores the growing demand, especially for users working with smaller language models.

Paloma and GNUS

Taariq shares some frustrations they have encountered in starting to use a GPU resource.  While there are ongoing conversations with Akash, he describes a user experience to access Akash.

“If you want to use Akash as GPUs, you have to buy an AKT token. And then of course to get an AKT token, you have to go to a DEX, and then you have to figure out which DEXs and which chain  have the AKT token.”

He emphasizes that transactions should be frictionless and recognizes the many opportunities for Paloma to smooth out the UI/UX experience emphasizing their strength in automated bots. He gives an example of a potential opportunity for Paloma to partner with GNUS.

“Being able to have the Paloma validator set handle your cryptocurrency inputs on whatever chain you have so that those resources can get to where GNUS tokens are traded, on demand. If you say, Hey, I want to write a program that will only activate the GNUS network when a customer  has an  inference question,  Paloma wants that job.”

Paloma can ensure seamless cross-chain execution where developers can build on GNUS and do not have to worry about token price, slippage, and price volatility. He shares his excitement at the prospect of creating bots on GNUS and other GPU providers like Akash where Paloma can integrate frictionless transactions and ease of use for customers.

Closing

Taariq thanks his guests and all the listeners who tuned in to today’s AMA.

GNUS.AI is all about on-demand inference from millions of mobile phones around the world. It is a new phase of DePIN, a new phase of AI, and a new phase of GPU access.”

Next week’s episode will be about breaking down Palomabots’ code using OpenAI to help folks understand how they work.

“You can build your own bots. And who knows, maybe even execute automated GPU access and GPU training on products and protocols like GNUS.”

Stay tuned for the next AMA!


To find out more about Volume, check out Volume Finance (https://volume.finance/), join the Discord (https://discord.com/invite/Ebh6YjMShu), and follow us on Twitter https://x.com/volumefi

Check out Paloma Chain on palomachain.com (https://www.palomachain.com/), follow them on Twitter (@palomachain), and join the flock on Discord (https://discord.com/invite/tNqkNHvVNc). Coo! Coo!

Learn more about GNUS.AI, the GNUS team, and what’s next for these AI & blockchain pioneers at https://GNUS.AI & follow them on Twitter at https://x.com/gnusai