📲 Follow us on X | 📨 Past Editions 

Welcome to the second edition of the TAO Times, where we share weekly updates on all of the most important events occurring within the Bittensor ecosystem!

For validators, subnet owners, miners, token holders, & community members.

Here is what happened during the week of August 25th ↘️

Ecosystem

• The highly anticipated v7.3.1 update and mainnet upgrade that enables Ledger hardware support and child hotkeys has been delayed. Additional updates are expected to arrive on Sept. 3, 2024.

• dTAO has been live on the incentivized testnet for a week now, with competitors out to amass as much TAO as possible. Naturally, various games are being played with deregistering and re-registering subnets in an attempt to capture these emissions. Datura & Sai were the top earners this week.

Subnets

SN4 (Manifold): Josh from Manifold announced upcoming updates to the subnet, including enhanced DDOS protection, expanded validator statistics, and progress on a new verification mechanism that could enable multi-model functionality for the entire subnet.

SN8 (Taoshi): Discussion centered around refining SN8's incentive mechanism to address issues with inactive traders earning rewards from past high-leverage trades, with proposed solutions including a softmax function for top miners and a "capacitor" feature to better control incentive distribution. Additionally, there was debate about lowering registration fees to increase competition and accessibility, balanced against concerns that high fees encourage more cautious trading strategies.

SN9 (Macrocosmos): The team released its white paper, which demonstrates that decentralized, incentive-based AI model development can produce models competitive with state-of-the-art equivalents like GPT-2 and Falcon-7B. The paper outlines the subnet's architecture, design choices, and results, showing how their 700M and 7B parameter models outperform or match industry standards, while also presenting a roadmap for future improvements and scaling of the subnet.

SN11 (Dippy AI): Will be hosting an event at Google's Toronto office on Sept. 30, featuring Joseph Jacks, Mog Machine, and Google's Gemma team, to discuss open-source AI efforts and network with the AI community. You can RSVP here.

SN12 (Compute Horde): Discussion this week revolved around optimizing SN12's performance and fairness, with key topics including the improvement of the dancing mechanism to combat weight copying, adjustments to scoring systems to better reward honest miners, and addressing reliability issues in weight revealing.

SN13 (Macrocosmos): The team released v1.6.0, introducing significant updates to the Hugging Face upload process and dataset validation system for miners and validators, including improvements in data handling, encoding/decoding mechanisms, and preparation for future enhancements to the scoring system and data sources.

SN15 (De-Val): Officially launched on mainnet, offering a decentralized evaluation platform for LLM performance testing, with features including hallucination detection and comprehensive summarization assessment.

SN17 (404 Gen): Released its highlights of the week indicating that the team has made significant progress in 3D asset generation and storage, developed a Blender plugin, implemented its Discord bot for Text-to-3D asset creation, and improved prompt generation for industry-specific applications.

SN19 (19.ai): Has rebranded, launching a new website with advanced chat and image generation features, introducing a leaderboard, and sharing an interactive roadmap for future upgrades and API developments.

SN20 (BitAgent): The team released v0.3.6, enhancing WandB data logging, reducing miner sample size to improve performance, fixing tool data types, and providing a new analysis notebook for validators and miners.

SN28 (Foundry): Pushed its new incentive mechanism to mainnet. The new mechanism shifts from RMSE to a ∆ factor, introducing tiered directional accuracy rankings, and implementing exponential reward distribution based on miner rank.

SN30 (Bettensor): The team revealed in its Discord channel that they will be transitioning their subnet to incorporate more AI-driven predictions alongside human inputs, while addressing technical issues on the testnet and developing a new incentive mechanism to enhance fairness and competition among miners.

SN34 (BitMind): The team revealed that 40 new miners have already started mining on the subnet.

SN38 (Tatsu): Released its white paper, revealing its user-centric framework that verifies the authenticity of online identities. The subnet is designed to address the challenges of fragmented identity verification systems. The OTF provided constructive feedback on Taτsu's subnet registration, highlighting areas for improvement to align with their standards for AI research and capabilities.

SN40 (VectorChat): Released a new chunking framework that allows users to submit tasks via the API server, which are then processed by the validator and returned to the API. This will allow validators to begin monetizing their bandwidth.

SN42 (Masa): The subnet is officially live on mainnet, and is receiving ~ 0.41% of emissions during the time of writing. Masa’s dual token incentive structure will enable miners and validators to earn both TAO and MASA through their data contributions from platforms such as Twitter, Discord, YouTube, and others. That data will be used to train AI models for use cases such as trading signals, market sentiment analysis, personalized AI companions, and more.

Pushin’ τ

Ventura Labs published this color coordinated ecosystem map that classifies the subnets according to their use cases.

That is all for this week folks! Make sure to subscribe to receive next week’s update. 🤝

You read and share, we listen and improve. Send feedback to [email protected]

Disclaimer: This newsletter is provided for educational and informational purposes only and is not intended as legal, financial, or investment advice. The content is not to be construed as a recommendation to buy or sell any assets or to make any financial decisions. The reader should always conduct their own due diligence and consult with professional advisors for legal and financial advice specific to their situation.