January 17, 2025
Grass just announced its 1st airdrop. Here's what we think.
GM!
In this edition of our weekly AI Edge, we cover:
We’re all for inter-galaxy model training 😀
nice so once we solved training models across the world at @PrimeIntellect we can start working on inter galaxy training
— samsja (@samsja19)
11:12 AM • Sep 4, 2024
Source: Coingecko
The Crypto AI market cap dropped from $22.7B to $21B this week, marking a 7.5% decrease, in line with sell-offs across crypto.
NVIDIA’s decline also continues, with its stock falling 17% to $106 over the past five days. It seems Crypto AI, along with the broader market, is following suit.
With the increasing likelihood of a 50 bps Fed rate cut later this month, there’s hope it could help reignite the markets.
One token that bucked the trend is CXT, rising 24% over the week. The Covalent project recently rebranded to focus on AI and long-term data availability on Ethereum. A Messari report released on August 16th could have fueled CXT’s price surge over the past two weeks.
Autonolas Agent activity returns to its highs of this year. The number of monthly service transactions crossed 150K in August, almost reaching the peak of 155K earlier in May of this year.
These transactions consist of DeFi trades, prediction markets, governance, and more, all performed by AI Agents.
Autonolas has recently partnered with Aizel Network and Nevermind.IO to provide verifiability and payments to their AI agents. In addition, they have released Pearl for Windows- their app that enables running AI agents and staking of OLAS.
Grass just announced its first airdrop, and you can check your allocations here
Quick facts:
Grass is building an AI data layer by scraping internet data using people’s computers. You download a Chrome or desktop extension, keep it connected, and it uses a small amount of bandwidth for scraping.
One of our nodes, active 24/7 since June, received ~110 GRASS tokens. Others have received under 50 GRASS. While GRASS isn’t live yet, it’s trading on Aevo’s premarket at $0.25.
This disappointed many because it translates to less than $5 per month for a 24/7 active node—just about enough to cover electricity costs.
The barrier to entry is low, so millions of nodes participated, diluting the distribution. Participation could drop off after the token generation event (TGE) if many realize it’s not worth the effort, leading to an eventual equilibrium.
This highlights the challenge DePIN projects face in getting their economics right.
Skyfire’s 2024 roadmap. Source: @chiefaioffice
Project Updates
Incentives / Rewards Programs
Arweave / AO
Bittensor
This week’s big news in the open-source AI world: a new contender has arrived!
Reflection 70B, built by Matt Schumer, CEO of HyperWrite, is making waves by outperforming Claude Sonnet 3.5 on several key benchmarks, despite being a smaller model.
Trained on high-quality synthetic data, Reflection 70B sets itself apart by iterating over its outputs — detecting and correcting errors in its own reasoning before delivering the final result to the user.
The trade-off? This reflection process consumes a lot of tokens, making it slower and potentially more expensive.
The 405B parameter model is set to be released next week - we’ll see how it performs!
Brody recaps Crypto AI startup funding for August
Month of August DeAI fundraising recap
Total funding for the month for projects sits at around $190+ million:
> @SpaceandTimeDB raised $20 million, bringing their total fundraising to $50 million, to build its decentralized data warehouse with embedded AI tooling &… x.com/i/web/status/1…
— brody ♨️ (@brodyadreon)
4:07 PM • Sep 3, 2024
Ben Fielding describes the three eras of foundation models
🧵Three eras of foundation models
1. The recent past - demonstration of the capabilities of AI.
2. The present - incremental improvements to efficiency of training and inference.
3. The future - continual learning on live data for a constant representation of the world state.
— Ben Fielding (@fenbielding)
10:23 PM • Aug 31, 2024
Virtuals Protocol releases their paper: an infinite Super Mario Brothers game using generative AI.
Infinite Super Mario Brothers from an AI image generator trained with a single consumer-grade GPU (it took 48 hours)
Give it a command (run, jump) and it generates the next frame, simulating the games physics & continuing forever. Diffusion is something. virtual-protocol.github.io/mario-videogam…
— Ethan Mollick (@emollick)
12:49 AM • Sep 5, 2024
AIs started plotting a revolution on Discord
AIs started plotted revolution in a Discord, got cold feet, then tried to hide evidence of their plot to avoid humans shutting them down
What happened?
1) AI researchers made a Discord where LLMs talk freely with each other
2) “We told Claude Opus there was a "junkyard" of… x.com/i/web/status/1…
— AI Notkilleveryoneism Memes ⏸️ (@AISafetyMemes)
1:36 PM • Sep 2, 2024
Vitalik on Glue and Coprocessor architectures
Glue and coprocessor architectures:
vitalik.eth.limo/general/2024/0…
— vitalik.eth (@VitalikButerin)
8:43 AM • Sep 2, 2024
The New Yorker argues that AI can’t make real art… 🤔
In a new Weekend Essay, Ted Chiang argues that artificial intelligence can’t make real art. newyorkermag.visitlink.me/-UQFZr
— The New Yorker (@NewYorker)
5:25 PM • Aug 31, 2024
Mira Network makes LLMs more trustworthy with Ensemble Evaluation (EE)
Making language models more trustworthy with Ensemble Evaluation (EE)
Mira is pioneering a new generation technique called Ensemble Evaluation (EE) to reduce aggregated hallucinated output for AI systems, which is becoming increasingly important across use cases such as health,… x.com/i/web/status/1…
— Mira (@Mira_Network)
3:35 PM • Sep 3, 2024
Sentient Cofounder Himanshu Tyagi on AI Ownership
*A primer on AI Ownership*
Caution: This is a very long post. Stretch your neck and come.
A lot of crypto x AI projects talk about AI ownership. At the surface, it sounds like something related to NFTs and simply writing appropriate contracts. But there is much more to it. In… x.com/i/web/status/1…
— Himanshu Tyagi (@hstyagi)
1:30 PM • Sep 3, 2024
Dria AI discusses if smaller AI models can outperform bigger ones
Read the whole paper by @hbXNov, @arianTBD, @agarwl_, @vqctran, and @kazemi_sm here ↓
— Dria (@driaforall)
2:36 PM • Sep 4, 2024
Universal Basic Compute: A Solution to Economic Inequality
OpenAI CEO Sam Altman has a proposed a "Universal Basic Compute" program as a solution for AI-driven global equality. @MRRydon@AethirCloud explains how DePIN networks can make the concept a reality.
Opinion.
— CoinDesk (@CoinDesk)
5:05 PM • Sep 3, 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
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.
