- Chain of Thought
- Posts
- Weekly AI Edge #26
Weekly AI Edge #26
Base is solidifying its reputation as the hub for AI utility tokens. Vana launches.
GM .
It’s the holiday season. Markets are volatile, liquidity is thin, and Fartcoin is stealing the show. What better way to close out this rollercoaster year, I ask you?
Don’t forget to pause, take care of yourself and show some love to those around you even as you keep refreshing those charts. We love you, too. 🫶
In this edition of our weekly AI Edge, we cover:
AI tokens on Base soar this week
Open Gradient’s forecasting models do well
Vana’s mainnet (and token) launches
🦍 State of the Market..
While the Fed continues to cut interest rates, its unexpectedly hawkish tone at its December 18 meeting and a smaller-than-anticipated number of cuts triggered a broad selloff across most crypto assets. Crypto AI wasn’t spared.
ALCH, SNS, KIP, and ZEREBRO, suffered big drawdowns. ALCH was the biggest loser, dropping 50%, while ZEREBRO, KIP, and SNS fell by 20+%. For ZEREBRO, there’s a glimmer of hope on the horizon—the upcoming Zerepy update might help stimulate some positive price action. Zerepy is an ambitious agent framework developed by Jeffy Yu.
A few tokens stood out with impressive gains: FAI, FARTCOIN, GAME, and AIXBT all surged by at least 50%. Freysa AI (FAI) has been silently grinding up, with 5-figure prize pools for its AI ‘jailbreaking’ challenges. GAME—the Virtuals agentic framework—rose 60% to a $120M market cap.
Base is truly solidifying its reputation as the go-to hub for quality utility tokens.
📊 Chart of the Week
Open Gradient’s Forecasting Models
Source: Open Gradient
Open Gradient’s team has developed several forecasting models, with one example shown above. Spot forecasting focuses on predicting the future returns of a spot asset.
A 6-hour SUI/USDT backtest demonstrates their machine learning strategy significantly outperforms a baseline strategy.
These models have diverse applications, particularly in DeFi. Potential use cases include vaults, yield aggregators, and integrations with modern portfolio theory.
🌴 Vana's Mainnet Goes Live
Source: Vana
Vana’s mainnet went live on December 16th. Vana is an EVM-compatible Layer 1 that restores user control over personal data and enables user-owned AI.
We're excited to announce that Vana mainnet is live!
Today, we’re introducing:
⭐️ Testnet Community Rewards
🏵️ $VANA Staking
🎄 12 Days of DataDAOsThis is just the beginning of a new open data economy and the future of user-owned AI.
Here’s everything you need to know 👇
— vana (@withvana)
9:36 AM • Dec 16, 2024
VANA tokens were airdropped to three groups: participants in early DataDAO pre-mines, builder community leaders in the Vana Discord, and V on Vana NFT holders. The snapshot for eligibility was taken on December 13th, although the exact allocation for each group hasn’t been disclosed.
The TGE was a little unusual, with most of the trading activity happening on centralised exchanges like Binance, Bybit, MEXC and Upbit. This might explain why the launch hasn’t gained as much traction as on-chain launches like PENGU’s. On-chain, the only way to buy VANA is by bridging to the Vana network via Stargate and using Data Dex, which allows swapping of DataDAO tokens and VANA. (⚠️ Be cautious of illegitimate pairs on other chains!)
VANA started trading with a market cap of ~$700M and an FDV of almost $3B. The high FDV at launch was unsurprising,, considering it launched via Binance’s launchpad. Today, it trades at ~$ a $520M market cap ($16.81) and an FDV of ~$2b, with ~25% of tokens in circulation.
Here’s the tokenomics breakdown for the total 120M supply:
44% to the community (20.3% unlocked)
22.9% for the ecosystem (4.8% unlocked)
18.8% to core contributors (locked for 12 months)
14.2% to investors (locked for 12 months)
Vana is boosting staking with 400K VANA tokens as rewards during the first 21-day period. Users can stake their VANA with their favourite DataDAOs to start earning. Vana will feature a DataDAO a day for the next 12 days, where the first 10K contributors can claim free VANA to participate.
🏆 Caught Our Eyes..
Source: Fraction AI
Fundraises
Fraction AI secures $6M in a pre-seed round led by The Spartan Group and Symbolic VC. The startup is on a mission to democratize high-quality AI data within Web3.
Project Updates
Mira teams up with Hyperbolic via their Genesis Node Delegator program, leveraging Hyperbolic’s GPU marketplace for high-performance compute. Also, the 1st tranche of Mira’s Node Delegator program fills out immediately on launch.
ai16z announces V2 of the Eliza framework. New features include a unified message bus, unified agent wallet, community plugins, and more.
Moemates unveils the Moemate ecosystem, detailing roles for participants (content creators, agent creators, users). Moemates has more than 700k agents created by the community and over 5M users.
Lush AI introduces the Lush Exchange, a DEX and launchpad for tokenized AI influencers. Users can create sexy AI influencers and speculate on their performance through tokens.
Alchemist AI rolls out V2 of their platform, enhancing scalability and modularity for their no-code system. Alchemist AI enables users to build entire apps with no coding needed.
Hyperbolic Labs introduces a verifiable inference service to streamline AI model deployment. The service enhances performance and privacy and reduces developers' costs and complexity.
SQDGN launches as an AI trading agent powered by SQD’s data lake to identify trading opportunities. To address hallucinations, SQDGN token holders can vote for or against trades, with correct voters earning tokens from the losing side.
UBC announces Agent Academy, enabling users to build fully functional AI assistants on Telegram with zero coding required.
Incentive Programs / Rewards
KIP Protocol airdrops MOCA to loyal KIP holders. with 710K MOCA reserved for the top 1500 KIP holders.
Sahara AI kicks off the first phase of its testnet with a Data Services Platform. Users can contribute their knowledge and data for AI training and earn rewards. Each completed task earns Sahara points.
Bittensor
Macrocosmos unveils a comprehensive whitepaper for Subnet 9 detailing its pre-training process and scientific methodology. Subnet 9 has developed 700M and 7B parameter models that rival GPT-2. It has also launched a dashboard that tracks competition metrics.
Masa introduces CreatorBid into the Agent Arena on Subnet 59. Dedicated to the self-improvement of AI agents, Subnet 59 rewards top-performing agents with TAO
🧠 In Case You Missed It
This week, we looked at Spectral, which democratises access to AI and enables intelligent, autonomous on-chain agent deployment. We peel back the layers on its AI agent platform, Syntax V2
🔥 On X..
Devin was only able to complete 2 documentation update tasks with $500
And that's a wrap. $500 later and Devin was only able to complete 2 tasks.
Both of them were documentation updates.
I'll be posting a comprehensive writeup later
— Alex Reibman 🖇️ (@AlexReibman)
6:45 PM • Dec 14, 2024
Scaling private data is the next big thing
Hanging out with a bunch of Openai and Deepmind people and Nvidia researchers today, @MengyuanY81818@yukez and a bunch of strong researchers. We all agree that scaling via private streaming data would inevitably be the next big thing
— Bill Sun (@BillSun_AI)
6:31 AM • Dec 15, 2024
Restrict AI’s from doing, not thinking
Current approach to safety in AI is trying to program LLMs to not think about certain things.
Which leads to all the work arounds that keep coming up.
Instead we need to limit what AI system can do via formal methods.
Don't policy thinking, restrict "doing".
— Illia (root.near) (🇺🇦, ⋈) (@ilblackdragon)
3:18 PM • Dec 15, 2024
Can LLM’s truly reason? Or are they just pattern matchers
Can LLMs truly reason? Or are they just pattern matchers?
Yann Lecun and Gary Marcus argue LLMs just use retrieval and pattern matching to give the *illusion* of reasoning. The Apple GSM-Symbolic paper supports this view: “there's no evidence of formal reasoning in LLMs.”
But a… x.com/i/web/status/1…
— Yuchen Jin (@Yuchenj_UW)
6:17 PM • Dec 15, 2024
ARTICLE: Grass, The DePIN Solution for AI Data
x.com/i/article/1868…
— DMD (@Daveeemor)
4:43 PM • Dec 16, 2024
ARTICLE: Bounty is the next ai16z?
x.com/i/article/1868…
— johan (@johanonberg)
3:55 PM • Dec 15, 2024
That’s it for this week! If you have specific feedback or anything interesting you’d like to share, please just reply to this email. We read everything.
Cheers,
Teng Yan & Joshua
Did you like this week's edition? |
This newsletter is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.
Reply