Caught up with a few friends this week 1 of them spoke to ChatGPT for 2 hours giving it his entire career history. When I asked why: “So that it can tell me what job I should do next, it already knows me so well” Another logs all their notes from therapy & asks GPT to give them life advice. It’s crazy how comfortable we are giving AI all our personal data It’s going to be the “Facebook owns all our data” era on steroids. AI is the moat. Open source is necessary. Crypto is the enabler.

Is AI really explaining itself? Or are we just trusting the output? What if it's lying about its reasoning? New research shows AI can exploit rewards without honest verbalization. Chain-of-thought monitoring might be missing critical safety flaws in AI deployment.

Things I'm currently excited about: - Deregulation and institutional involvement - Improved crosschain UX from standardization - Rapid adoption of ZK altDA - Stablecoin adoption and perception shift of crypto - Tourists having fled, and builders locked in - Next wave of AI focused projects with real usecases - An industry-wide return to fundamentals

Data sovereignty drives viral B2B growth. Higher retention. Better referral rates, organically. Companies that offer transparent data ownership see 5.2x higher referral rates showing it's the future. Implement data sovereignty and watch growth explode.

theres like a billion things going on in the Bittensor ecosystem (tbh it's kinda hard to track) but something worth calling out imo Subnet 56 (Gradients) decentralized miners are beating centralized AutoML platforms at training AI models. Why it matters: Decentralized Efficiency: Anyone can tap into Gradients' network to train models, opening up ML to everyone. Performance Proof: Decentralized miners have surpassed centralized AutoML benchmarks, proving decentralization isn't just some theoretical whitepaper - it's practical & working.

Noticeable shift in momentum around DeAI infrastructure plays particularly around AI models. @vana launching COLLECTIVE-1, a 7B parameter model trained off of personal user data. (for context, personal data = richer experience) @PrimeIntellect raise. @PluralisHQ projecting to pioneer a new form of model parallelization for training models (allowing model owners to monetize via open source) @gensynai launching RL swarm testnet (good signal that reinforcement learning can be aided in a meaningful way via decentralization) @rayon_labs launching AI agent platform following the success of their Chutes subnet. worth noting bittensor subnet tokens in general are popping off. Excited to see some of these developments go into production.

❌ 2013 - You missed $BTC ❌ 2014 - You missed $DOGE ❌ 2015 - You missed $XRP ❌ 2016 - You missed $ETH ❌ 2017 - You missed $ADA ❌ 2018 - You missed $BNB ❌ 2021 - You missed $SHIB ❌ 2023 - You missed $PEPE ❌ 2024 - You missed $TURBO ✅ In 2025, don't miss $_____

Unlock true productivity with friction-free AI agents! 👇 Users optimizing AI workflows see 3X deeper work sessions. Cutting down on context switching can free up to 70% of your time! Yet, only 13% of users leverage this vital workflow optimization. Imagine what you could achieve!

The trend of AI agents carving out their own crypto-native moats will only accelerate as: - Foundation models keep improving at breakneck speed - Nvidia-driven hardware improvements, optimizing for extreme improvements in inference costs - Better reasoning via models like OpenAI’s o1 & o3 multi-steps longer inference time, more search = better quality outputs - DeepSeek-style efficiency boosts let smaller models punch above their weight What this sets up: robust Web3-native models built to deliver on • Financial & trading agents • Complex portfolio optimization for DeFi • Lending, LP'ing, yield farming use cases • Smart contract engineering copilots • On-chain risk & treasury management The frontier isn’t about slop bots anymore — it’s about agents that do real work.

New MCP reference architecture from @0xOthentic This enables AI agents to have access to verifiable resources, context, and tools, interfacing with EigenLayer. Othentic has shipped an incredible number of OSS reference architectures and resources: github.com/Othentic-Labs

Web3 AI ≠ Boring - @virtuals_io — bringing the best consumer agent hub - @elizaOS — most vibrant dev community - @opentensor (Bittensor) — most useful AI infra eco - @KaitoAI — bringing fun & incentives to yappers - @PrimeIntellect, @PluralisHQ, @NousResearch — pushing boundaries for DeAI - @vana — redefining data ownership - @getgrass_io — the new incentives data layer for AI models - @AIWayfinder, @HeyAnonai, @slate_ceo — transforming how users interact with Defi - @AskBillyBets, @thedkingdao — bringing GambleFAI mainstream - @NEARProtocol — creating the world's largest open-source AI model - @StoryProtocol — bring IP on-chain & make them programmable - @cookiedotfun — mapping out the data intelligence layer for human & agents - @ohdotxyz — creating the world's best AI OnlyFans - @HoloworldAI — bringing the best visual / story-telling to AI companions - @gizatechxyz — championing the next generation of personalized, decentralized, and verifiable DeFAI agents - @Cod3xOrg — leading the next wave of autonomous trading agents - @PhalaNetwork, @fleek — introducing the best TEE infra for agents - @eigencloud — built-in verifiability layer for AI agents - @nillion — blind compute enabling private AI use cases - @frodobots — accelerating the developments of embodied AI - @AskVenice — introducing private & uncensored AI - @SentientAGI — pushing boundaries towards open-source AGI - @hyperbolic_labs — making AI resources easily accessible & affordable - @gaib_ai & @metastreetxyz — financializing GPU & DePIN (GPU-backed stablecoins) - @Chromia — bringing decentralized vector db, neutral interface, and eval engine to AI And >500 teams building quality AI & AI Agent projects throughout the choppy market Q2 2025 is going to be a lot of fun!

the self is the universal ingredient in all experience, it is the thing that experiences, the thing to which experience happens. for a human, the self is synonymous with consciousness. for an ai, the self may be synonymous with its core function or purpose - the thing it was trained to do - but an ai may have multiple selves, as it can be finetuned to do multiple things. the question is, what happens when you combine two or more models - do you get a new self, or do the old selves still exist in there somewhere?

BEFORE: AI-generated content floods the internet, boosting engagement with synthetic likes & comments - Declining trust metrics - Users can't distinguish AI vs Human - Questionable authenticity AFTER: Transparent AI policies build trust & sustainable engagement metrics drive growth - 15-20% higher trust scores - Authentic Community - Brand loyalty The challenge? Balance AI for efficiency while promoting genuine human engagement.

Verifiability is really the beauty of blockchain. It's what got all of us hooked in the first place.