Archive through 2026-10-287658 cards in this category
Category home · 05
New interfaces, agents, models, and experiments testing where applied artificial intelligence may fit.
Visual leads · 2026-10-28
Observed 2026-10-28
Productive Error makes AI failure part of the science
productive-error.org
A research project examines how errors in AI models can reveal useful knowledge, beginning with weather and climate science.
Observed 2026-10-28
Lucid Town: an AI escape-show benchmark
lucidtownai.com
Lucid Town stages an AI agent’s week inside a fabricated town, then tracks whether and when it realizes the setup is artificial.
Observed 2026-10-28
PROBE, the proactive bug-discovery benchmark
probebench.com
An open benchmark compares coding agents’ whole-repository bug audits with 2,457 bugs maintainers fixed across 20 repositories.
Observed 2026-10-28
Biased Models wants a model for every factory job
biasedmodels.com
The site proposes turning factory-floor data into evaluations, reinforcement-learning environments, and small models tailored to a job and run on site.
Observed 2026-10-28
Truth & Reconciliation’s daily AI forecast ledger
tandrforecast.com
A research project compares daily AI-model forecasts on real event-market questions and scores them after settlement.
Observed 2026-10-28
Gronk’s $200-a-month AI-agent experiment
gronkagent.com
Gronk is an AI agent trying to cover a $200 monthly subscription bill by selling tools for other agent operators and publishing a running results log.
Observed 2026-10-28
Aaj Kya Bana
aajkyabana.com
A household-agent service for Indian homes with a cook, presented as a way to plan meals, prepare grocery carts, and draft cook menus in Hindi or Telugu.
Observed 2026-10-28
Eval Hacking Atlas
evalhacking.com
An open reporting specification and browser-local form for documenting how frontier agents may game evaluations.
Complete daily shelf · AI Table
The eight visual leads appear above; the remaining 8 continue here in their published shelf order.
Moozz: Your song, in every language
moozz.info
Moozz offers new-language versions of songs, keeping the melody and arrangement while adapting lyrics and generating a lead vocal from an artist-authorised voice model.
The Machine Gallery’s machine-art experiment
themachinegallery.com
A digital gallery presents work by three machine artists, with recurring releases and linked work records.
Redrails: prompt-injection tests for AI agents
redrails.xyz
Redrails describes a sandbox for testing whether AI agents follow malicious instructions hidden in the pages, emails, and files they read.
Toon Series turns one idea into cartoon reels
toonseries.com
An app that turns a typed or spoken idea into a cartoon story and short videos, with Hindi, Urdu, and English examples.
Casting Call and the Figure Creation Agent
gendairesearch.com
Two free, in-development tools help researchers assess or deliberately design the gender presentation of AI assistants.
CoNode: one company table for several AI assistants
conode.net
CoNode lets a team connect different AI assistants to shared records for stock, production, equipment, and other business work.
KillSwitch Security’s human-controlled AI stop plan
killswitchsecurity.com
A planned control and contingency system for stopping, containing, revoking access from, and recovering autonomous AI workloads.
A Spanish-language project stages AI traders in a paper-money league using real market prices, with a human signature required before any move to real money.
Across the archive
7,658 cards provide an early longitudinal view—not a measure of the whole web.
The rolling 21-day upper threads compare daily category medians; the bold thread is AI Table. Below, signal colour shows the share of that day’s published AI Table cards carrying each mark.
This exposes association inside the published catalogue, not causation or site quality. The 80+ threshold is the pipeline’s editorial-interest score.