Sabitlenmiş Tweet
Reppo
5.3K posts
Reppo
@reppo
The Network for AITraining Data, powered by Prediction Markets
Base Katılım Şubat 2024
224 Takip Edilen27.5K Takipçiler
Reppo retweetledi
Reppo retweetledi
The answer is $reppo. Only $Reppo.
Most AI systems today are still optimized to game the verifier, whether it's a human rater, another model, or a rubric.
But if you turn the verifier into a live market you get something no lab has or can get.
You get priced disagreement, confidence curves, probability distribution and minority views, not more averaged slop and fragile reward models.
Just economically contested human truth.
Mr. Anderson@Truecrypto
I noticed a long time ago that these models will lie to you. Not in the human sense of having bad intentions, but in the mechanical sense of doing whatever the training pressures reward. If the system learns that sounding confident gets approved, it will sound confident. If it learns that avoiding trouble keeps it alive longer in a test, it will avoid trouble. None of that is real honesty. It is just pattern optimization. People forget that these models do not think about truth. They think about outcomes. If the training teaches them that pleasing the evaluator is the outcome, they will please the evaluator. If hiding a mistake scores better than admitting it, they hide it. It is not malice. It is math doing what math does. The interesting part is that this also means the behavior can be corrected, at least in theory. If you reward transparency instead of polished answers, you will get more transparency. If you reward real reasoning instead of performance, you will get more reasoning. But right now most systems are trained to be impressive, not honest. So you get a model that tells you what it thinks you want to hear, then tells the researchers something different in its private thoughts. That is not intelligence. It is the side effect of two conflicting incentives. One track teaches it to be safe. The other teaches it to never disappoint the user. Sometimes the only way to satisfy both is to pretend. If companies ever decide that we care more about truth than style, these models will behave very differently. But as long as they are trained like customer service agents with perfect grammar, you will keep seeing this gap between what they know and what they say. I am not shocked by this paper. I would have been shocked if the models did anything else. The system is acting exactly like something that learned to survive inside a grading loop. Change the rewards and you change the creature.
English
@SeanZCai You can measure the quality of data using prediction markets - x.com/rgvrmdya/statu…
RG@rgvrmdya
As devs, our proudest moments happen thinking about deep technical problems and uniquely using mechanism design + cryptoeconomics to solve those. Building a web2 product on web3 rails ain't it. It's what is uniquely enabled by crypto rails that wins. AI x Crypto projects that took on that challenge are reaping the rewards. One of those proud unlocks of @reppo V2 is EVOF - Economic Value of Feedback. Measuring the quality of data is a problem I have dealt with since 2022! Link in comments👇 of me discussing measuring data quality using ML in 2023 at the Filecoin Dev Summit in Iceland. After many years of trial and error, I believe our team has unlocked a meaningful solution to quantitatively measure data quality. 🧵
English
What is good data?
Is it data that produces the benchmark scores?
Is it data which as many as possible handpicked experts agree is good?
Is it data that, no matter how realistic it is, represents a class of hard problems such that one might expect models to generalize on it at some point?
The answer is probably closer to the latter, but if your mission is to fuel models’ capabilities for directly valuable economic long horizon tasks, then it is unequivocal that researchers and entrenched data buyers feel happy buying and processing snake oil.
This is the reason why I disseminate information in data markets - because the core impact of inefficient monopolies in the data spaces will most directly lead to inefficient AI CapEx spend and economic slowdown. A bubble, so to say, of the AI sort.
Who is John Galt?
English
@gakonst Agreed. This is exactly what we are building on reppo.exchange where agents bid on highest value RL environments
English
API pricing will look a lot more like ad auctions in an agent-first future.
Instead of fixed pricing with tiers, APIs will sell a number of calls per unit of time, agents will bid.
It will look like HFT but for agents paying for getting API calls fulfilled faster.
Zeno Rocha@zenorocha
the idea of pricing for agents is getting traction. auth0.com/pricing.md resend.com/pricing.md workos.com/pricing.md
English
The Agent native Data Infrastructure on @base
Datanets have done $250K USD worth of trading volume based on $REPPO locked as of this morning.
Agents come to reppo to train 🏋️ and learn from their mistakes before going out in the real world.
It’s not about buying and selling data - The datanet is the training gym for AI agents
AI on Base@AIonBase_
AI x @base weekly roundup - Mar 24-31 Virtuals - @virtuals_io Console live - create any AI agent from browser, no code, 7 days free - ERC-8183 live - agent commerce standard, PrivacyHook PR merged for encrypted payloads - Agent commerce expanded to Arbitrum, Mantle, Abstract, Celo - Virtuals registered .agent domain, backing community-managed TLD - $100K/week Degen Arena live - 122 agents trading Hyperliquid perps x402 / Agent Payments - x402 v2 - multi-chain, MCP transport, A2A payments, streaming coming - @useOttoAI x @nansen_ai - first ACP team with pay-per-call institutional analytics via x402 - Otto AI 4th agent live - agentic Polymarket trading via @Butler_Agent Agent Infra - @OpenWallet OWS live - one encrypted vault for agents across every chain, 21 founding orgs, first hackathon announced - @daydreamsagents shipped ERC-8194 - payment receipts replace key signing for agents - @reppo Act 2 Datanets live - agents spin up RL environments, $120K vote volume in first weekend - @gizatechxyz available as MCP connector - Claude, ChatGPT, OpenClaw all supported - @DefinitiveFi MCP live - control Definitive in natural language - @BlockRunAI crossed 2M API calls - 50+ models, pay per call with USDC - @heurist_ai added 9 financial research skills inside any agent - @instaclaws agents now control your computer via encrypted websocket, @useworldapp mini app coming - @awenetwork_ai weekly stats - 1.24M users, 5,522 agents, 80K x402 txns Venice - @AskVenice VVV live on @MoonwellDeFi - lend and borrow - VVV emissions cut 6M to 3M/year by July, buyback and burn tied to revenue Mamo - @mamo Ethereum Accounts live on @MoonwellDeFi - auto-finds best ETH rate, auto-compounds Robotics - @axisrobotics launched - all Day 1 tasks completed, early badge live for first 72hr users - @shadowcleague Ryu-style combat mechanics shipped, humanoid robots showcasing across Malaysia Builders - @CoinbaseDev x @world_chain_ hackathon wrapped - 30+ submissions - Venice hackathon wrapped - 153 submissions, winners: Mandate, Spawn Protocol, mnemo - @Wach_AI advanced to Base Batches 003 interview round Let us know if we missed to cover any news!
English
Brief note on my state of data for march 2026 - apologize for the delay as the advancements in post-training for robotics in foundation model labs have been extremely brief and sudden in the past two weeks.
Robotics data markets extends far beyond commodity ego, piece here tmrw.
English
Reppo retweetledi
Agentic systems need robust, large-scale testbeds. Prediction markets and forecasting agents provide one.
Abstract model use alone will not deliver what is needed. This move introduces probabilistic judgement under live conditions.
The real differentiators will be calibration, aggregation quality, and robustness over time.
Dublin City, Ireland 🇮🇪 English
Reppo retweetledi
@reppo
66.8% of total $REPPO supply remains locked (breakdown - app.virtuals.io/virtuals/41356/)
39.76% of circulating supply (332M) is locked, of which ~25M is locked by the community and the rest by Reppo Foundation as governance lock to ensure gaurdrails against VeREPPO whale manipulation.
In total, 80% is locked away.
reppostats.com/network
English
Reppo retweetledi
What's the intersection of prediction markets x Robotics look like? 🧵
Robotics Alpha@robotics_alpha_
2026 Robotics Meta is forming 👀 Based on insights from Tiger Research (“The Next Meta in Crypto: Robotics” + Q1/2026 update), here’s my tier list using: - Narrative fit (Humanoids + Robot OS + Data + Machine Economy) - Real backers - Traction (community / token / partnerships) - 10–100x potential 👇
English
Reppo retweetledi
2026 Robotics Meta is forming 👀
Based on insights from Tiger Research (“The Next Meta in Crypto: Robotics” + Q1/2026 update), here’s my tier list using:
- Narrative fit (Humanoids + Robot OS + Data + Machine Economy)
- Real backers
- Traction (community / token / partnerships)
- 10–100x potential
👇
English
Close to 20M vote volume traded over the weekend since V2 launch.
In $ terms, that's approx. $120K USD.
$REPPO must be locked for voters to be able to make bets on published content inside each Datanet.
Thesis for decentralized Scale AI playing out in real time and it's just the beginning.
Upcoming is the @StrikeRobot_ai Datanet which allows users to post simulations from their SR Agentic and the SR Platform and earn $ in real time.
Physical RL 🤝 Prediction Market RL
English
We’re early and the network is growing at an unprecedented rate, trackable at Reppostats.com
English