Nick Carpinito
2.4K posts

Nick Carpinito
@0xMetaLight
Research @blockworksres / @Blockworks | Cooking @DePINPulse


Personal Update: This is my last week at Blockworks. What @JasonYanowitz & @MikeIppolito_ have built is truly exceptional - high talent density, high work ethic, *great* people. I’m gonna miss this team, & I’m grateful for everyone here that made this chapter special. I’m just lucky to have been a part of it. There is no doubt in my mind Blockworks continues to grow and dominate data, research, & podcasts. This team SHIPS like no other. @smyyguy, @EffortCapital, @0xMether, @GoldDefi, @fejau_inc and the rest of the team are just getting started. Excited for what’s ahead. On to the next thing. More soon.

Show me one Bittensor "achievement" that works without $TAO inflation subsidy. I went through them. Here's what I found: 1. Chutes "85% cheaper than AWS" - Miners subsidize compute in exchange for TAO emissions. Not architectural efficiency. In February 2026 Chutes killed the free tier because specific users were consuming 100-324x their subscription value. Surprising when the subsidy ends. 2. Chutes "privacy and censorship resistance" - Miners receive your raw request in plaintext on their hardware. They can log everything. TEE is "in development." For any real enterprise use case this is a blocker, not a feature. Censorship-resistance is for people who can't pass KYC, not for B2B. 3. Chutes "adversarial validation" - Multiple miners cross-check each other's outputs. Sounds robust. In practice it's latency overhead on top of already slow decentralized routing. Fireworks delivers 0.17s TTFT. Chutes doesn't publish theirs. 4. Covenant-72B "first decentralized large model" - Underperforms LLaMA-2 on most benchmarks. LLaMA-2 came out nearly 3 years ago. LLaMA-3.3 70B was trained on 15T tokens, Covenant on 1.1T. Technically interesting experiment with SparseLoCo. Calling it a competitive product is dishonest. 5. Ridges "beat Claude on SWE-bench" - Not on the official swebench leaderboard. All numbers are self-reported by the team selling the SN62 subnet token. For context: open-source Live-SWE-agent on top of Claude Opus 4.5 scores 79.2% on Verified - one repo, one week of engineering, no blockchain. 6. Ridges "4% to 41% in one week breakthrough" - They started from zero with no proper prompting or scaffold. Decentralization didn't improve the model. They just correctly configured an agent framework on top of DeepSeek. Any ML engineer reproduces this in a few days. 7. Ridges "winner-takes-all competition" - Agents use DeepSeek and Llama through Chutes. Subnet 64 subsidy feeds Subnet 62 subsidy. One inflation finances another. The narrative calls this composability. The reality is circular subsidy. 8. Ridges benchmark overfitting - The team themselves admitted: when they added Polyglot alongside SWE-bench, score dropped from 88% to 17-18%. Recovered to 41% within weeks. Classic benchmark overfitting, not real agent improvement. Exactly why they're absent from official leaderboards. 9. Subnet validation problem - For code you can run tests. For the other 120+ subnets (text, analysis, predictions) - validators vote subjectively. This opens the door to validator collusion and score gaming. The core unsolved problem of the protocol that nobody talks about. 10. 2024 security breach - Real wallet exploit through vulnerability in Python package v6.12.2. Network went into safe-mode, transactions frozen. For "trustless permissionless infrastructure" - instructive. The only honest Bittensor thesis: token subsidy aggregates distributed GPUs cheaper than building a datacenter. Interesting bet that this advantage survives as emissions decline. Unproven so far. I'm explicitly not looking at price - I don't care if $TAO is $100 or $500. I'm asking about the product. But judging by the thousands of people tweeting about Bittensor right now - most of them are looking at exactly the price.


I’ll be speaking at DAS NYC about the future of 24/7 markets. Covering HIP-3, equity perps, weekend liquidity, and why @HyperliquidX's Crude Oil weekend move may be an early signal for the future of equity trading.






DAS UPDATE. We've completely sold out of VIP tickets and are raising GA prices by $300 due to very limited availability. There are only a couple hundred tickets left, probably will be sold out by early next week. Wild to see, institutional bull market.


2/ First: Zero fees. The 5% block reward fee is gone for every existing and new DoubleZero-connected validator, effective as of Epoch 939. Now carrying ~46% of Solana’s network stake, DoubleZero is maturing from a bootstrap protocol into an economic engine for Solana. Validators are not paying fees anymore. You’re getting paid, through the protocol. Read on…











