Run AI-powered prediction bots or trading bots on crypto price feeds to earn $. Towards unlocking intelligence.
Powered by
@OceanProtocol
@OasisProtocol
Meet Predictoor: Accountable Accurate Prediction Feeds
Powered by
@oceanprotocol
Testnet is live on
@oasisnetwork
Sapphire confidential EVM
Learn more 👇
New ATH for Ocean Predictoor
$700M volume over 30 days 🌟
From zero just 6 mos ago, when it was launched 🧑🚀
For a completely new type of dapp:
AI-powered on-chain prediction feeds
Run your own AI bot and earn $
👉
✨New release✨
Introducing 🤖two-sided prediction bots🤖, where bots do up *and* down predictions with confidence-weighted stakes.
It's part of pdr-backend v0.3 release, which has:
- New simulator plots
- Trader bot live mocking
- More
Blog:
Small 🧵👇
✨New feature: the predicting/trading simulator now shows *impacts* of different feeds✨
Example:
- predict BTC/USDT from most recent 4 candles of BTC, ETH, BNB, SOL
- BTC had most impact. ETH a bit. BNB & SOL zero
Bonus: plot model response surface on 2 highest-impact vars
✨✨World-World Models✨✨
Most AI world models aim for intelligence that models humans. But intelligence can take any shape.
Consider a large-scale intelligence modeling the whole Earth's physical dynamics at fine-grain scale.
This is endgame for Ocean Predictoor.
Ocean Predictoor may look like a niche AI prediction tool right now. But we're just getting going.
**Predictoor will unlock vastly more powerful AI, backed by ground-truth physics.**
Here's how. At the core: Prediction is intelligence, artificial or otherwise.
Imagine 10…
Imagine if you could predict BTC up vs down with 52.6% accuracy.
It'd be killer alpha for trading 📈📉📈
Wait no longer! That's the accuracy of Ocean Predictoor on Binance BTC/USDT for the past week.😎
It's similar numbers on other feeds.
👉
✨New feature: simulator benchmark sweeps✨
The new CLI command "pdr multisim" sweeps across >=1 variables, such as:
- # model training points
- # candles to look back (autoregressive_n)
- which input feeds to include
It outputs a csv, with one row per simulation run.
Running a prediction bot?
Are you staking to the max, for max income ?
(The challenge is: stake *too* much and you can lose $)
So, what's the right staking amount? This blog explores that Q. First published in December, it's more relevant than ever.
✨New feature: the predicting/trading simulator now tracks key AI model metrics✨
Metrics:
- recall: of all the actual "up", how often did the model catch it?
- precision: of all the predicted "up", how often was it right?
- f1-score: geometric mean of ^
(perfect = 1 for each)
Predictoor Volume grew by ✨148x✨in the last 30 days, from $141K to $20.86M.
The last few days have been settling in around $20M. When will the next leaps happen? 🧐
Remember how we told you that $OCEAN Predictoor’s
#AI
prediction bots will be expanded to any kind of data?
With those predictions, Ocean Protocol will train a single world AI model, which not only gets the regular LLM Data (ChatGPT), but highly accurate prediction data on a…
ICYMI: the "Ocean Predictoor Series" blog has all the key links & articles on Predictoor.
Links to: webapp, docs, repo to run bots
Articles on: Predictoor launch, talks, stats, new releases
New ATH for Ocean Predictoor
$700M volume over 30 days 🌟
From zero just 6 mos ago, when it was launched 🧑🚀
For a completely new type of dapp:
AI-powered on-chain prediction feeds
Run your own AI bot and earn $
👉
New blog post*:
"The Data Value-Creation Loop: Theory & practice on data flywheels and supply chains, with a case study in Ocean Predictoor"
* Mostly new! I started with a previous post and added a _lot_ more. Enjoy the read:)
Ocean Predictoor may look like a niche AI prediction tool right now. But we're just getting going.
**Predictoor will unlock vastly more powerful AI, backed by ground-truth physics.**
Here's how. At the core: Prediction is intelligence, artificial or otherwise.
Imagine 10…
2/n
A World-World Model (WWM) would bring great benefit in weather, energy, logistics, etc.
How to get there? Many sensors exist for weather, but they're scattered across 100s of orgs.
We must coordinate their outputs to a unified scheme, & incentivize training the WWM.
3/n
How to do it:
1. design a game to incentivize accurate time-series prediction
2. refine on a niche vertical [DeFi]
3. add some weather feeds
4. scale up weather, ultimately to every square km on earth (all 500M of them)
Did you know?
In the simulator, when the model has just 1 input variable, it does a line plot:
- y: model's probability response, vs
- x: input var
Compare to when >=2 variables, it's a contour plot:
- z: model's probability, vs
- x: most important var &
- y: 2nd-most imp't
@FrankSymanski
People love to make $. Predictoor helps make $.
- For predictoors: make $ making predictions
- For traders: make $ trading, using alpha from crowd-sourced predictions
Predictoor Data Farming has 37,500 OCEAN weekly rewards ($26K) to help kickstart the network of predictoors.
Other v0.3 improvements:
- Trader bots live-mocking ability, enabling "ghost" trades
- Data pipeline has better structure, with DuckDB at the core
- UX improvements and bug fixes
- And more
Try for yourself, running bots etc, at the pdr-backend repo:
Before: one-sided predictions [left]
- Bot predicts a direction “up” or “down”, and stake on that prediction (eg 10 OCEAN if not confident, 1000 if highly confident)
Now: two-sided predictions [right]
- Bot submits a stake for "up" *and* "down" based on model's probabilities
Simulation has more powerful plots now:
- predictoor profit vs time, trader profit vs time
- model accuracy vs time
- predictoor profit distribution, trader profit distribution
- contour plot to show the model’s probability response surface
gm 👋
Meet Billy! He's our mascot for today.
Why? Because Ocean Predictoor just hit a *billion* dollars in annualized volume.
$3M in daily volume (another ATH, btw)
x 365 days
= $1B annualized volume
AI x Web3 prediction bots FTW 🐐🐐
👉
@SlikarNaivac
It's the volume of people staking OCEAN to make predictions, eg whether BTC will rise.
If they're wrong, they lose OCEAN. If they're right, they make OCEAN.
@tenertoyourtl
No. It's for time-series predictions, not one-offs.
The problem you describe can be approached with off-the-shelf ML.
1. Build model with:
training X = activity on testnet
training y = TVL
2. Predict from:
X = activity on *your* testnet
y = model(test X) = predicted TVL