GM!

Happy Lunar New Year! Wishing you prosperity, good health, and maybe even some green candles.

But while the festivities are in full swing, the markets clearly didn’t get the memo.

The launch of DeepSeek triggered a mass sell-off across US tech equities and AI-related stocks—and the hot takes have been flying.

Most of them? Pretty bad. But one of the more thoughtful takes came from Dario, CEO of Anthropic:

“..DeepSeek V3 is not a unique breakthrough or something that fundamentally changes the economics of LLMs; it’s an expected point on an ongoing cost reduction curve. What’s different this time is that the company that was first to demonstrate the expected cost reductions was Chinese.”

Cheaper, better, faster AI is always a win in our book.

In this edition of our weekly AI Edge, we cover:

• Nous Psyche, the latest from Nous Research
• The controversy around VVV
• DeepSeek, a huge deflationary shock to the price of intelligence

Another brutal week in the market. The launch of DeepSeek, combined with weak sentiment across sectors, brought even more pain.

Agent-related tokens got wrecked—MOSS, NOVA, and ELIZA all plunged 50% this week alone. Even major AI names like VIRTUAL, FARTCOIN, AI16Z, and AIXBT weren’t spared, each dropping 20-30%. Right now, buying into agents feels like catching a falling knife.

On the flip side, MOR and TAO stood strong. Morpheus kept up last week’s momentum, pumping another 50%, likely fueled by VVV’s launch due to their partnership. Meanwhile, TAO is finally getting some love. With more chatter around the upcoming dTAO upgrade, it looks like everyone is waking up to the fact that TAO has been overlooked for too long.

VVV Airdrop Controversy

Venice dropped its VVV token out of nowhere, and yeah—it made waves.Venice is a chatGPT-like app that runs on open-source models like Deepseek and Llama, but everything stays in your browser with prompts routed through their proxy infrastructure to a set of decentralized GPU providers.Venice went all-in on its community, airdropping half the total supply straight to Venice users and Base token holders (VIRTUAL, AERO, DEGEN, etc.). And the numbers? Wild.

• 10M out of 25M tokens have already been claimed
• 124 accounts scored 5,000–10,000 VVV each ($40K–$80K).
• 287 accounts walked away with 10,000+ VVV (over $80K).

But then came the dump. Since launch on Jan 27, the market cap cratered from $680M to $170M—a brutal 60-70% correction. Coinbase listing VVV within one day didn’t help either.The data tells an interesting story. On Jan 28, there was a 20K jump in holders, likely due to the 24H airdrop unlock rule. Regular Venice users had their tokens locked for 24H and had to upgrade to Pro to claim. Meanwhile, Base ecosystem holders and others with unlocked tokens could sell immediately. When those tokens hit the market, well… you can guess what happened next.

This TGE was messy. Some were thrilled with free money for holding Crypto AI tokens, while Venice users felt rugged by the lock-up period.And it wasn’t just the tokenomics causing a stir. Concerns popped up over Venice sending inference requests in plain text, raising questions about GPU providers seeing user prompts. Then there was the insider trading scandal, with two Aerodrome contributors suspended and Venice team members dumping ~1% of the supply on day one. Not a good look, in our opinion.Airdrop mania, a controversial launch, and privacy concerns. Plenty of drama for a week.

🌴 The Mortal who became a Goddess

Nous Research just dropped Nous Psyche, and it’s a big one.

This new infrastructure aims to decentralize and democratize AI development, making training and inference accessible to the community.

Built on Solana, Phase 1 of Psyche is all about cooperative training for generative AI—pulling together a mix of GPUs, from gaming rigs (4090s) to data centre powerhouses (A100s, H100s), to train AI models at scale. It builds on Nous’ previous work with Nous DisTrO, which trained a 15B parameter model across multiple locations in the U.S. and Europe last year—setting a record for fastest convergence in distributed training, all without high-speed interconnect hardware.

Phase 2 is where things get even more interesting—Psyche moves beyond just training and into reasoning and reinforcement learning. Think Deepseek R1, but trained on a decentralized network that allows for accessible inference and higher-order thinking models‍

Fun fact: The project is named after Psyche from Greek mythology - a mortal who became a goddess through sheer determination. Good omen?

The endgame? Open, decentralized infrastructure for training superintelligence. A pretty wild vision.

Nous hasn’t officially announced a token, but all signs point to it. A well-designed token could coordinate resources, align incentives, and drive adoption—exactly what Psyche needs to scale.

Nous is already crushing it with their Hermes LLM Series. Hermes 3 is the brain behind tons of social agents on X, Telegram, and Discord. And they're not stopping there - the plan is to make it even smarter through Psyche's reinforcement learning system.For months, Crypto AI has been obsessed with AI Agents. But distributed training has always been one of the most exciting use cases for Crypto AI, and Nous Psyche just put it back in the spotlight.

Funding Raises

• Pippin Founder Yohei Nakajima launches Agent Fund, a VC fund for AI agent apps, infra, and tools. Led by Yohei, Alex Reibman, and Adam Silverman, it’s open for investors on a rolling basis. Equity fund, not token-based.

Project Updates

• Story Protocol announces Emergenceverse, a franchise with a storyline created by David Goyer. David Goyer is the mastermind behind the Dark Knight trilogy and Blade. AI agents and users will build the next major franchise on-chain.
• Nillion network hosts the new DeepSeek R1-14B model privately using TEEs inside nilAI. Users can now access DeepSeek while ensuring their interaction is completely private. Nillion Network is building a privacy-focused network for AI.
• Hyperbolic partners with Nexus to provide GPUs for powering Nexus's ZK technology. Hyperbolic’s Eliza plugin is also live, allowing Eliza agents to autonomously use Hyperbolic GPUs
• Nesa announces a fully encrypted inference API. Equivariant Encryption (EE) is an adaption of Homomorphic encryption that works on various models to provide end to end privacy on data manipulation but with zero latency.
• Prime Intellect releases TOPLOC, a locality-sensitive hashing scheme for verifiable inference. It detects modifications to models, prompts, or precision and is robust across various GPU types. Prime Intellect is building the foundation for decentralized, verifiable compute protocols.
• Giza crosses 11.5K autonomous transactions through ARMA, its automated on-chain yield optimizer bot. Users can deposit stablecoins for ARMA to find and optimize the best yields across multiple chains.
• thirdweb, a full stack Web3 development platform, introduces Nebula for Unity, allowing game devs to build agents that talk to players and interact on-chain

Incentive Programs / Rewards

• Privasea AI’s testnet part 2 has arrived. Users can provide processing power for secure AI computations and earn rewards, including token incentives. Privasea is building an encrypted AI network enabled by FHE.

AI Agents

• Virtuals is expanding to Solana, introducing a Meteora Pool, a Strategic SOL Reserve, and maintaining the AGENT/VIRTUAL trading pair. The expansion includes a Venture Partner Model with grants, an AI Hackathon, and collaboration with key partners like Jupiter Exchange and LayerZero. Virtuals also offers grants of 42K VIRTUALS tokens for aspiring projects.
• Youmio partners with Virtuals to bring agents to life in 3D worlds. Virtual agents can now be brought into 3D worlds powered by Unreal Engine, opening up new possibilities and environments for these agents.
• Chromia rolls out EVAL, an agent that evaluates other AI agents. Built on the Chromia engine and Virtuals ecosystem, EVAL scores AI outputs on truth, accuracy, creativity, and engagement. AIXBT received a score of 80/100.
• Drift integrates with the Solana Agent Kit, enabling agents to trade perps, borrow/lend assets, and interact with Vaults. Drift is the home of on chain trading on Solana.
• 0G Labs releases their ERC-7857 standard. This standard is built explicitly for AI agents and enables the representation of AI agents as NFTs. This new standard includes privacy-preserving metadata where AI “intelligence” stays encrypted and secure.
• OLAS announced Modius, an advanced DeFAI agent. Modius autonomously manages crypto assets with intelligent rebalancing and lending strategies.
• THINK agents drops their whitepaper for their Agent Standard. The THINK Agent Standard creates the foundation for intelligent on-chain agents, making them composable and interoperable.
• ai16zdao rebrands as ElizaOS after a request from the VC firm a16z. The $ai16z ticker will remain.

Bittensor

The dTAO upgrade for Bittensor is set to drop in mid-February, so expect a surge in attention.

• Mode launches Synth 'subnet' on Bittensor to build decentralized AI that predicts crypto prices (subnet 50).
• Crucible Lab’s State of Bittensor 2024 report is out. It’s a great read.

We dropped a new deep dive this week—now shifting gears from AI agents to the infrastructure powering them. Atoma Network is building the most trustable compute network on the planet 👇

Atoma Network: Private & Verifiable AI  Building the most trustable compute network on the planet

🔥

 On X..DeepSeek = Bullish for Crypto AI?

DeepSeek is extremely bullish for crypto x AI. > An open source model with better performance than Open AI's model built at a fraction of the price--and they did it with inferior hardware.> All the major crypto x AI frameworks can easily use open source models or build on top…
x.com/i/web/status/1…

— Mark (@markus9x)
5:24 AM • Jan 27, 2025

DeepSeek is a huge deflationary shock to the price of intelligence

DeepSeek's new R1 reasoning model is dragging down the NASDAQ. It dropped 6 days ago but it seems Wall Street is only now digesting what it means. I'm no equity analyst, but a few things I've been thinking about.DeepSeek is a huge deflationary shock to the price of…

x.com/i/web/status/1…

‍— Haseeb ＞|＜ (@hosseeb)
1:01 PM • Jan 27, 2025

DeepSeek isn’t the answer. It’s the question

the internet wasn’t built. @stevesi nails it hereit snuck in through the cracks of “serious” infrastructure. http was a toy. linux was a hobby. ai’s next phase won’t come from a roadmap. it’ll come from someone bored enough to ask “what if we tried less?” deepseek isn’t the…

x.com/i/web/status/1…

‍— Yhprum (@Yhprums_Law)
8:42 AM • Jan 27, 2025

DeepSeek shows that AI doesn’t have a data moat

One of the fundamental things that @deepseek_ai release shown is that AI doesn't have a data moat.People traditionally credited OpenAI, Google and Meta is that they have massive data that others can't access and it increases with every interaction users have with their…

x.com/i/web/status/1

— Illia (root.near) (🇺🇦, ⋈) (@ilblackdragon)
3:00 AM • Jan 27, 2025

DeepSeek’s breakthrough isn’t just about cost - it’s about architecture

DeepSeek's breakthrough isn't just about cost - it's about architecture.Their "mixture of experts" (MoE) approach means only relevant neural pathways activate for specific tasks.Think specialized doctors vs. one doctor trying to do everything.This is how they hit GPT-4…

x.com/i/web/status/1…

— tong (🌐,🌐) (@tong0x)
7:49 PM • Jan 28, 2025

There’s going to be a war for specialized data

there gonna be a massive war for specialized datawhy? Because not all data is equalthe rarer and more specialized the data, the more valuable and unique the insights it can producegeneral data is everywhere, but it only scratches the surface. specialized data creates…

x.com/i/web/status/1…

‍— nairolf (@0xNairolf)
2:29 PM • Jan 28, 2025

A comprehensive resource for Bittensor’s thesis and Dynamic TAO

This started as our year-in-review report on Bittensor. But halfway through writing it, we realized there wasn't a single comprehensive resource that simply explains the thesis, maps the ecosystem's growth, and explains Dynamic TAOSo Seth and I set out to create it

— Sami Kassab (@Old_Samster)
9:01 PM • Jan 29, 2025

State of Bittensor 2024

Our State of Bittensor 2024 report is live!Alongside the report, we're releasing a recording with the authors of the report, interviewed by our co-founder @dlawee and @brodydota from Outpost.We discuss:
> David's past experience as an executive at Google and as the founder…

x.com/i/web/status/1…

‍— Crucible Labs (@CrucibleLabs)
9:01 PM • Jan 29, 2025

That’s a wrap for this week! Got feedback, wild ideas, or something cool to share? Just hit reply—we read it all (yes, even the “AI is taking over the world” ones).

Cheers,Teng Yan & Josh

‍

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.