The AI Arena closed beta has successfully concluded, after an impressive 3,574 models submitted and 112,750 validations from 285 AI developers. FLock is pleased to launch the open beta, broadening participation to the wider developer community. The whitelist requirement has been lifted and now developers with an active Github account can participate in decentralised AI training on FLock’s platform. .

From training AI companions to generating data for training AI assistants, you have shown how decentralised AI can work at scale. We wish to thank our community for your dedicated testing and tasking.

AI Arena is a decentralised AI training platform

FLock launched AI Arena on the train.flock.io web app in mid-May following a $6m seed round. This is a world-first, marking a major milestone where AI meets blockchain for full-stack decentralised AI training. Developers compete and collaborate in a Kaggle-style leaderboard, using their own or decentralised compute. Models are trained on local and crowdsourced datasets to build AI agents. 

This is the first project to bring an end-to-end AI fine-tuning lifecycle to Web3. In doing so, FLock:

• Enhances transparency and participation
• Introduces incentive mechanisms to ensure fair contributions and security
• Democratises AI, allowing diverse communities to develop purpose-built AI models

Users stake Flock’s native token $FML to participate in three ways. Training nodes are AI developers responsible for training and fine-tuning. Validators evaluate work done by training nodes then submit validation scores that influence reward distribution. 

If you are not an active developer with an active Github account, you may join AI Arena as a delegator. Delegators provide stake tokens to other participants, enhancing the network's validation capacity without directly participating in the task training or validation process. 

10+ tasks to create AI assistants, companions, & Web3 agents

The closed beta saw ten tasks go live on the platform to build AI assistants, AI companions, and Web3 agents.

In the vast majority of tasks, datasets have been provided. The challenging element of Farcaster GPT was the requirement for users to generate training data themselves using Farcaster documentation. This simulates real-world scenarios where there is insufficient data to train a model. Read our interview with a top-performing user for his techniques to succeed in such tasks.

FLock has co-hosted hackathons with various other fellow Web3 industry leaders, including 0xScope and Akash Network. To make Scopechat more efficient, users created a multilingual query analyser. Users used Akash decentralised compute in building AI companion ‘Professor Grump’ as a lead-up to the July ETHCC RenAIssance event in Brussels.

A recent task the FLock community is training is a Text2SQL agent model to empower users to efficiently query databases without needing to understand SQL syntax. We are also joining forces with Web3 partners to launch a Text2SQL task purpose-built for blockchain-related data. 

The latest task to go live is Life Simulator, building a gamified virtual simulation of life with realistic decision-making from career paths and relationships to hobbies and personal growth. Additional AI companion tasks include mimicking Asgard character Loki. For all Dungeons & Dragons enthusiasts, DND MBTI is also set to create an immersive experience through an app.

Traction set to further grow in open beta

This slide represents the users and their activities. AI developers stake as training nodes, and carry out training submissions by staking tokens, and validators evaluate the training submissions.

Around three hundred lucky users were accepted onto the whitelist and actively using AI Arena. These chosen few have been busy bees completing huge swathes of tasks: training nodes have on average submitted 29 tasks each, and validators 440 validations.

The figures are expected to rapidly increase now that the whitelist is lifted.

Join AI Arena and try out our trained models

If you are an active developer, verify your Github to start training through this Discord channel. If you are not, join as delegator, or try out models trained on beta.flock.io.

https://guild.xyz/flockio-46dbee 

Please continue to log feedback, bugs and feature requests on our Discord to further refine our platform.

Find out more about FLock.io in this previous blog and our docs.

For future updates, follow FLock on X.