Memory & Retrieval · paper · 12 min
From Noisy Retrieval to Decision Evidence: Silver Evidence Demonstrations for In-Context Learning
- Serves a goal: grounded answers from retrieval
- First of 42 on the day's shortlist
Activation 0.71
Every morning · one field · a few picks
What deserves your attention today, and why.
Hundreds of new papers land in an active field every day. Pool of Thought reads all of them each morning, notices which areas are unusually busy, weighs every paper against the goals you set, and picks the few worth your time. Every pick comes with its reasons, and so does everything it left out.
The demo · applied AI research · Mon 26 Oct
Four views of the same day
Same papers, same reasons. Pick the view that suits the moment.
- new papers read
- 520
- areas watched
- 12
- on the shortlist
- 42
- on the stage
- 1
Today Attend
A stage. The day's top pick in the spotlight, the runners-up in the wings, your goals in the director's chair.
Open the stage →
Explore Perceive
A chart of the field. Every arrival plotted by area; click any one to see why it did or didn't make the shortlist.
Open the map →
Field Survey
The whole field at a glance. Regions light up where an area is running well above its usual pace.
Open the field →
Focus Act
A heads-up triage display for heavy days: the stream ranked, a reader in the middle, your goals alongside.
Open focus →Every pick explains itself
Feeds rarely tell you what they didn't show you. Here every paper keeps its reasons for the day, whether it won or not.
Memory & Retrieval · arrived, not on the shortlist
EVOLVE: Uncertainty-Aware Multivariate Time Series Anomaly Prediction with Adaptive Prototype Memories
- 436th of 520 on the first pass, which weighs goal fit, the area's momentum and released code. Only the top 301 were scored in full.
- Serves none of the goals, so it could only compete for one of Memory's own places.
Both cards are from the demo's day of 26 October 2026, as its Explore view shows them.
You set the goals
A workspace starts from a few standing goals in plain words: what you're researching, what you follow, what you're learning. Each one leans on the ranking by a weight you choose. Switch one off and the day re-ranks in front of you.
If you want more judgment than keywords give, a language model you choose can grade papers against a rubric you write, under a monthly spending cap you set.
- Research · ×1.0Grounded answers from retrieval
- Follow · ×0.8Efficient inference
- Learn · ×0.6How to evaluate language models
- Team watch · ×0.5 · offOpen-weight model releases
Yours, and private by default
Its own address and database
Each workspace runs on its own hostname, and its projects, goals and days live in a database of its own.
Nothing shows until you sign in
A private workspace shows nothing to visitors. You sign in with a one-time link sent to your email; there's no password to leak.
It doesn't learn from your clicks
The ranking depends on the day's papers and your goals, nothing else. What you hold and switch off stays in your browser.
Want one for your field?
Private workspaces are set up by hand during the beta. Tell us the field and what you're trying to keep up with.