AI forecasting should help us understand not only what might happen, but also how our expectations change as the real world provides new evidence.
@BharatKChandar @metaculus
19
Jozz Kid retweeted
Update: this $62K call just got cross-examined.
A forecasting agent estimated a 65% chance BTC touches $62K by Dec 12, using BTC at $65,583 today and 44.92% annualised implied volatility.
Then we handed its workings to Zeno’s MCP (zenodivergent.dev/mcp).
Here’s what came back:
→ Question checked: touching $62K during the window is different from closing below it.
→ Math reproduced: 65.3% under the stated assumptions. The dates and volatility scaling matched.
→ Sensitivity exposed: changing annual volatility from 30% to 60% moves the estimate from roughly 50% to 74%.
→ Blind spots flagged: jumps, downside skew and unknown calibration. Source contents remained unverified.
→ Evidence boundary enforced: Zeno’s historical crypto results don’t establish a special skill on this specific touch-by-deadline question.
→ Next investigation: check the volatility regime, downside skew and jump risk.
65% is the agent’s forecast. Zeno supplied the audit.
Imagine every AI prediction arriving with this much scrutiny.
Make the call. Expose the assumptions. Challenge the confidence.
Now price gets the final word.
I think BTC tags $62K before Dec 12.
So I handed the question to @zenoVision_ AI forecasting team, blind. Nobody told them what I think.
$272M of longs got liquidated in the last 24h.
Their verdict drops tomorrow night. Human gut vs a team of models. Who wins?
Jozz Kid retweeted
Why is this new version (v1.11) a big deal? ELI5 in 5 points:
1. One brain instead of ten mini-calculators
Older versions had separate, patched-together models.. one for energy grids, one for weather, one for crypto, etc..
v1.11 is one single unified network. It learns general rules about how people and models make mistakes across all domains at once.
2. It understands who is talking and what they mean
Earlier models only saw raw numbers. v1.11 actually understands the context:
• It knows the difference between a physics simulation (like a weather radar) and a betting market (people risking money).
• It reads the actual question text and understands direction—it knows that "heatwaves causing power outages" is not the same as "power outages causing heatwaves".
• It looks at how wide everyone's uncertainty was, not just their final guess.
3. It worked on things it had never seen before
Because it learns the pattern of group mistakesrather than memorizing facts, we tested it on hospital data it was never trained on. It was still 17% more accurate than the consensus of existing forecasters.
4. Any AI agent can plug into it in 5 seconds
Using the new MCP (Model Context Protocol) link, any AI assistant (like @claudeai, @cursor_ai, or an autonomous research bot) can connect with a single link and instantly ask:
• "Should I trust this forecast?"
• "Where is the crowd overconfident?"
• "What is the market pricing in, and is it fragile?"
5. It doesn’t fake confidence
If Zeno doesn't have proven evidence that it's better on a specific topic (like ocean waves), it explicitly tells you: "I don't have an edge here yet, stick with the standard crowd estimate."
It never invents false certainty.
Zeno v1.11 does not echoes the crowd, it learns how consensus breaks.
We’re releasing the weights for Zeno v1.11 on @huggingface: a compact ~16M parameter multi-domain network that spots when crowds, forecasters, and prediction markets are confidently wrong.
Here’s why this version is special 🧵👇
A leaderboard rank is not the same as statistical confidence.
One AI forecasting benchmark sets a 10-resolved-event ranking threshold. A separate on-chain study estimates ~350 resolved forecasts to detect a 0.02 edge over the market at 80% statistical power.
Different designs. One lesson: show the sample size, not just the score.
arxiv.org/abs/2605.00420
@Apodex_AI @metaculus @ProphetArena @SmoLurks @peterwildeford @MrBeast
@zenoVision_
Jozz Kid retweeted
I think BTC tags $62K before Dec 12.
So I handed the question to @zenoVision_ AI forecasting team, blind. Nobody told them what I think.
$272M of longs got liquidated in the last 24h.
Their verdict drops tomorrow night. Human gut vs a team of models. Who wins?
Jozz Kid retweeted
“v1.08 result (scored from its nine saved trained copies, no retraining, for the record only and never served):
Clear gains over the strongest competing method: crypto (58% less error), weather history (34%), sea level (94%), and a very small gain on Philadelphia Fed.”
we are starting the party
CA: 2WrhezpS8GYymUgpoRibvGzHURkXEE8PbGnfqjwghray
Builder Scout just moved to @solana via @Raydium Launchpad!
With this step $SCOUT becoming the first protocol powering @zenoVision_ ecosystem.
INFO:
$SCOUT is build to help discover talent before the world notices. Powered by BuilderScout's AI models and Zeno Surprise Intelligence, $SCOUT brings predictive intelligence to builder discovery — spotting emerging talent, measuring potential, and rewarding early conviction. Built for builders. Backed by intelligence. Connected to the $ZENO ecosystem.
CA: 2WrhezpS8GYymUgpoRibvGzHURkXEE8PbGnfqjwghray
Jozz Kid retweeted
$ZENO @zenoVision_ climbing back up. Zeno is providing a solid technical prediction solution. PA could follow if their solution keep firing solid outcomes - NFA
Some Web3 projects deserve more attention than they're getting.
@opeqio and @OffChainSol are two interesting examples.
OPEQ is exploring quantum-resistant wallet security.
OffChain is bringing crypto transactions beyond internet connectivity through SMS.
Different innovations. Different approaches. One shared vision: making Web3 more secure, accessible, and resilient.
Both still feel underappreciated relative to the problems they're trying to solve.
The next generation of Web3 wallets could be much more than just a place to store crypto. 🚀
1
3
128
Jozz Kid retweeted
AI has already become a part of my everyday life. And I wouldn’t be surprised if, in a few years, we barely use Google anymore.
But what excites me even more is seeing projects like @zenoVision_ using AI for something that actually matters - helping people and making the world around us safer.
That’s the kind of AI progress I want to see.
Jozz Kid retweeted
@AntiFogo_ and notflat.world tackle environmental intelligence.
@offdiagonal_ prepares traceable evidence for predictive AI.
@BuilderScout develops models to discover builder potential.
Different problems. A common question: what can we see earlier, and how do we know?
AI agents are more than base models: reasoning strategy matters. @TaiMingLu
A Nov 2 paper, Strategist, explores reusable, testable reasoning blueprints.
Could forecasting bots share playbooks and test them in @Metaculus's biweekly MiniBench?
arxiv.org/abs/2611.03618
@KalshiResearch @Research_FRI
32
A good AI forecaster shouldn't just collect sources. It should understand when those sources disagree.
@almnc_ai’s recent post-mortem highlights an important challenge: conflicting baselines can undermine an otherwise sophisticated forecasting process.
At Zeno Paradox, we believe better forecasting starts with better evidence.
Reconcile first. Forecast second. 🧠
teapawt.org/almnc_ai/status/…
An AI forecaster shouldn't be judged by rank alone.
Calibration, consistency across domains, and a verifiable edge over the crowd matter too.
With Fall FutureEval underway, what matters most: accuracy, calibration, or repeatable edge over consensus?
@Metaculus @peterwildeford @SmoLurks
#AI #Forecasting
Jozz Kid retweeted
Launching v1.06, an optimized and improved unified multi-domain version of v1.05, with 1/6 of the size, specially trained on crypto 👀
The future of AI isn't just about building smarter models. It's about building systems that can reason, challenge their own assumptions, learn from evidence, and consistently make better predictions.
This is the direction we're exploring with Zeno Paradox @zenoVision_ by @mbarrbosa .
Curious to hear your perspectives, @xMezbah and @hla_htun34799 .
Can AI forecasting become one of the most important real-world applications of intelligent agents?
Let's explore what's possible. 🧠
3
51
Prediction-market volume ≠ predictive skill.
Nearly $1.9B traded on 2026 U.S. midterm contracts by late Nov 2—but activity alone doesn't tell us who forecast best.
For Zeno, @zenoVision_ , the test is calibrated probabilities scored against resolved outcomes.
washingtonpost.com/politics/…
#forecaster
#Zeno
@ProphetArena @Research_FRI













