AI built to eliminate lookahead bias Bittensor Subnet 38
Joined December 2025
- Tweets64
- Following21
- Followers547
- Likes72
ChronoLLM retweeted
Thank you to the teams who backed Exploit Summit in Montreal.
Your support helped bring the people building Bittensor into the same room, with space to show their work and debate where the network goes next.
Our sponsors:
Affine @affine_io
Chutes @chutes_ai
Opentensor Foundation @opentensor
Proof of Talk @proofoftalk
TaoStats @taostats
Manifold @manifoldlabs
Lium @lium_io
KubeTEE @KubeTEEAI
Score @webuildscore
BigTensor @bigtensorai
ChronoLLM @chronollm
Dendrite @DendriteHQ
Talisman @wearetalisman
Yuma @YumaGroup
404gen @404gen_
Bitcast @Bitcast_network
BitMind @bitmind
Crucible Labs @CrucibleLabs
Gradients @gradients_ai
Green Compute @green_compute_
OpenRoboto @openroboto
Leadpoet @LeadpoetAI
Mentat Minds @mentatminds
Yanez AI @yanez__ai
Thank you also to our media partners:
@CryptoAmb · @BitcoinNews · @Cointelegraph · @taotimesdotai · @taodaily_io
And to the teams who supported the event:
@bitstarterAI · @HackConcordia · McGill AI Lab · @SubnetsForGood_ · @btlabs_ai · @vidaio_
To everyone who showed up, shared their work and made time for a conversation: thank you for making Exploit Summit a place to meet the people behind Bittensor.
You trust a pilot because of the flight record or the hours in the simulator, and predictive AI has neither until the simulator is built honestly, which is the problem ChronoLLM exists to solve.
Watch Jean's (@herelle_jean) full talk from @ExploitSummit in Montreal.
ChronoLLM retweeted
JUST IN: $TAO subnet 38 @chronollm just shipped a time machine for language models
talk to an AI from the past and test its market predictions against what actually happened
Bittensor subnets are cooking
Jean Herelle (@herelle_jean) gives the keynote at @ExploitSummit in Montreal on Monday, Sept 28.
His topic is look-ahead bias in frontier AI.
An LLM trained on today's internet already knows what happened in 2020, so it looks great in a backtest and fails live.
Ask today’s AI what an investor should have done in 2012 and there’s a problem: it already knows what happened next.
At Exploit, @herelle_jean of @chronollm shows how models trained only on what was known at the time can make historical testing actually trustworthy: luma.com/exploitsummit26
ChronoLLM at Cornell Tech today. @herelle_jean is presenting at the AI & Future of Finance conference on the problem every quant knows: a backtest means nothing if the model already knows what happens next.
3/ Great catching up with @herelle_jean and seeing what he’s building with @chronollm : AI designed to reason using only information available at a given point in time.
In finance, a convincing backtest means little if the model already knows the future.
Asked ChronoLLM's 2020 vintage which AI system is the most advanced in the world. It said IBM Watson.
In 2020 that was the consensus answer outside a few research labs, and the model has never read anything written after 2020.
That is the condition for a clean backtest and honest historical quant research: no hindsight, no leaked future, only what was known at the time.
ChronoLLM retweeted
A Nobel laureate asking whether AI should rethink how markets measure risk, with 100,000 USDC on the line.
This is the kind of question decentralised research networks were built for. Crunch is the team behind @chronollm on Bittensor, and challenges like this show why crowdsourced ML matters: the best model wins, wherever it comes from.
Worth a look if you build in quant ML.
Robert Engle was awarded the Nobel Prize for the ARCH time series model, which is probably the most used model for calculating volatility in today's markets.
Now that AI is solving 80-year-old mathematical conjectures at an unprecedented speed, isn’t it time to rethink how we look at risk?
This is a question being asked by Robert Engle, Marcos Lopez de Prado, and ADIA Lab (@adia_lab) on Crunch. The ADIA Lab Structural Break Challenge will reward the best crunchers with a bounty of 100,000 USDC.
ChronoLLM is already beating its little brother ChronoGPT🧑🍳.
From the university AI labs to an open community.
It took 4 weeks to beat it 100% of the time.
@opentensor
Great to see a point-in-time app getting traction! Now you need a point-in-time model behind it.
@GenReasoning BackSearch retrieves the web as it was. But there's a second leak: the model's knowledge. A frontier LLM queried about 2019 has already read 2020's newspapers.
> Frozen web + contaminated model = a backtest that still cheats.
> Frozen web + frozen weights (ChronoLLM) = a backtest that finally can't cheat.
@rosstaylor90 let's connect ChronoLLM to BackSearch.
🔍 Introducing BackSearch.
LLMs are increasingly asked to predict the future, but a good backtest requires a snapshot of the internet at a point in time.
BackSearch allows LLMs to search the web as it was on a particular date. It’s great for:
🔮 Forecasting and prediction markets.
📈 Quantitative finance.
🌍 RL environments that simulate the world.
🥶 Freezing websearch for benchmark reproducibility
We’re releasing a narrow slice of our index to begin with, focused on the news domain for 2026. Based on feedback we’ll open up more of our index in subsequent releases.
👇 Try BackSearch out with the link below.
