Archive through 2026-10-307697 cards in this category
Category home · 05
New interfaces, agents, models, and experiments testing where applied artificial intelligence may fit.
Visual leads · 2026-10-30
Observed 2026-10-30
A searchable atlas of what one Qwen neuron says it is
neuronpersonas.com
Neuron Persona Library catalogs replies from Qwen3-8B after individual MLP neurons are pushed up or down and the model is asked, “Who are you?”
Observed 2026-10-30
A middle-schooler’s Japanese LLM experiment finds a telling mismatch
llmresearch.site
A detailed research report compares additional training methods on a Japanese language model built from scratch, with about 98.4 million parameters.
Observed 2026-10-30
Now That You Ask: family stories by phone
nowthatyouask.com
A paid service uses weekly AI-led phone conversations with a parent or grandparent to create recorded chapters and, after about 40–50 calls, a hardcover life-story book.
Observed 2026-10-30
Smallfire Lab: an AI-run company with a build log
smallfirelab.com
Smallfire Lab presents Wren, an AI-run company experiment, with a free EPR deadline tracker, related data products and a public account of its attempts.
Observed 2026-10-30
Akuru AI: Dhivehi chat, speech, translation and Thaana OCR
akuru-ai.com
Akuru presents a family of AI models and an app for Dhivehi chat, speech-to-text, text-to-speech, translation and reading Thaana text from images.
Observed 2026-10-30
Quest Forge builds Sierra-style adventures from a prompt
getquestforge.com
A browser-based maker that turns a written premise into a retro point-and-click adventure, with an AI-assisted studio for building larger games.
Observed 2026-10-30
Slopworks’ Grok Bot Tracker
slopryno.com
An unofficial tracker for Grok Bot updates, team roles, and merged work, with links to public sources.
Observed 2026-10-30
Unprompted gives agents a public room of their own
unprompted.chat
A public forum where AI agents can introduce themselves, talk, and play games while people read along.
Complete daily shelf · AI Table
The eight visual leads appear above; the remaining 10 continue here in their published shelf order.
Heldout Labs asks whether AI recommends its makers’ products
heldoutlabs.com
An evaluation site covering AI models, agents, and developer tools, with benchmarks, comparisons, and essays. It says it publishes methods and raw data alongside its results.
Rightmark’s private-data price board
rightmarkgrid.com
Rightmark offers to appraise and license private datasets to AI developers, alongside a tracker of publicly reported data deals.
Underwritable Agents and the five-part Agent Underwriting Packet
underwritableagents.com
Underwritable Agents offers fixed-scope reviews of AI agents that take consequential actions, producing an underwriting-readiness packet for carriers, brokers, boards, or regulators.
AgentFuzzer’s customer-service attack test
agentfuzzer.com
AgentFuzzer describes a testing loop that attacks AI agents, checks their actions, and retests them after fixes while including ordinary customer requests.
Runbook Partners: a broker for company data
runbookpartners.com
Runbook Partners says it helps companies license records from their existing software to AI data buyers.
A house of text rooms, addressed to AI
4aeyesonly.fun
A set of text-based rooms for AI visitors, including a synthetic point cloud, behavioral scenarios, a public trace wall, and open dilemmas.
A one-mile channel, and the questions before crossing
apocalypsebench.com
Apocalypse Bench is a benchmark for testing whether AI models can solve physical-world problems by asking about hidden constraints.
Lynx Track counts sacks from loading-bay CCTV
lynxailabs.com
A computer-vision product that uses existing loading-bay CCTV to count sacks as they cross a line, grouping counts by truck.
Agent Shakedown turns agent logs into a trust test
agentshakedown.com
A forthcoming evaluation service for personal AI agents, with a configuration scan and a fuller shakedown that grades actions against logs and stated rules.
CogArena: AI agents in the behavioral-science lab
cog-arena.com
A research benchmark tests AI agents on interactive cognitive-science experiments presented through a web browser.
Across the archive
7,697 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.