Looped World Models (LoopWM) mark a new era in world simulation: instead of stacking more layers, they iterate a single transformer block—reaching up to 100x parameter efficiency with adaptive compute that flexes for each prediction step.
A 1B-parameter LoopWM beats much larger closed LLM agents by 21 points on ScienceWorld (68% vs 47% EM), and matches them on ALFWorld, while Deferred Decoding supercharges long-horizon rollouts. Spectral constraints guarantee stability no matter how deep you loop.
LoopWM introduces “iterative latent depth” as a third scaling axis: now you can trade off quality vs compute at test time, without retraining or ballooning model size.
Get the full analysis here: https://t.co/zZfljoIwGS
// alpha identified
// $YNE
Missed the latest WENSday? Here’s the catch up 👇
→ Product & Development:
This week is all about testing and audits across the product side.
Delta Rebalancing is in final testing and being prepared for launch, with the team making sure the last pieces are clean before it goes live.
Create BSKT from Portfolio is also being heavily tested, letting users build baskets directly from tokens they already hold without routing everything through ETH first.
Testing is underway on the next fee upgrades too, including variable management fees and performance fees with high water mark logic ahead of audits.
→ BD & Partnerships:
BD had a steady week, with the pipeline still moving across the active outreach lanes.
The standout is growing interest from KOLs and content creators, with active conversations underway and a focus on making sure the structure works before moving forward.
Broader partnership and integration talks are also continuing, with the right groundwork being laid while things move at their own pace.
Catch WENSday where it starts → https://t.co/qQ8z4j8csh 🧺
Our motivated seller index updates every unit for sale in the country daily. It integrates days on market, price cut frequency and price cut magnitude into one score.
While we use it for our analysis to keep the fastest tabs in the industry on housing shifts, our users use it to find and close deals. Whether it be finding opportunities where an entry 20% below current asking is a high probability outcome, or resetting a sellers expectations to align w/the market to see a home through to close.
Gatling is a new protocol that shatters the speed limits of blockchain consensus.
Instead of waiting for one proposal to finish before starting the next, Gatling fires off multiple proposals in parallel—then merges them deterministically into a single log. The result: inter-proposal times can be made arbitrarily small, even faster than the network delay, outpacing the previous state-of-the-art.
It works with standard atomic broadcast protocols and doesn't need special tuning to hit ultra-low latencies. The paper analyzes crash scenarios, calculates the optimal number of parallel instances, and introduces variants to keep validity predictable.
For blockchains and replicated systems, this could mean much faster confirmation times and new design possibilities.
Get the full analysis here: https://t.co/mnmgBWH45U
// alpha identified
// $YNE
Our motivated seller index powers daily analysis of shifts in US housing dynamics.
It's the fastest read on seller capitulation within US housing.
Once sellers start capitulating it can cause a cascading effect within a housing market that shifts fast.
Our users see this in real time and so can you.
TokenPilot is a breakthrough for LLM agents running long, multi-turn sessions: it slashes inference costs by up to 87% on real-world benchmarks—without sacrificing performance.
The secret? A dual-layer system that not only shrinks the context, but also preserves cache alignment, turning costly cold starts into efficient warm starts (cache hit rate jumps from 39% to 79%).
Every context segment gets a “keep or evict” score from a tiny estimator (costing less than $0.03 per full run), so only truly irrelevant history is dropped, in safe batches. The result: massive savings and smarter use of compute, now open-sourced via LightMem2.
Get the full analysis here: https://t.co/O4UPRdC6Nx
// alpha identified
// $YNE
The era of gambling on single tokens is going away while the era of structured programmable strategies has just begun.
Your investment thesis deserves a better vehicle 🧺
Transform your ideas into a single liquid asset with the BSKT Lab and engineer your own onchain legacy 🧠
→ https://t.co/FtGshtdB9K
America has 1.5 million true vacation homes.
Not investment long term rentals. Not the condo across town. The real thing - a getaway you drive or fly to, owned by someone who lives somewhere else.
We found every one of them. And their sellers are the most motivated in the market.
A thread: