Manifold Grace
EMERGING HORIZON / RESEARCH DISCOVERY

AI Research Discovery Directory

Explore where researchers publish experiments, challenge findings, investigate engineering failures and assess AI safety. Original research and open problems—not just product announcements.

← Emerging Horizon of Research and Its Advancement · AI Research Laboratory

27 sources shown

AREA 1 OF 4

Emerging scientific discoveries

Find fresh papers, proposed methods, critical reviews and research before it becomes mainstream.

Peer review

OpenReview

Submissions, reviews, rebuttals and changing arguments.

AREA 2 OF 4

Frontier engineering

Examine implementation decisions, error reports, reproducible code and unresolved architectural problems.

Discovery

GitHub Trending

Discovery starting point; popularity alone is not evidence of frontier value.

Open issues

vLLM — Issues

Inference serving, throughput, architecture and integration problems.

AREA 3 OF 4

Independent capability evaluation

Ask what has actually been measured, under which conditions, and where benchmarks remain incomplete.

Evaluations

METR Research

Research into agent-task capability, evaluations and limitations.

Evaluation framework

Stanford CRFM HELM

Evaluation methodology and historical benchmark comparisons; check maintenance status.

Benchmark consortium

MLCommons

Community benchmarking and measurement initiatives.

AREA 4 OF 4

Safety and governance research

Track oversight, security, robustness, interpretability, accountability and lifecycle risk.

From discovery to verified gaps and tested solutions

Research does not stop at finding a paper or open issue. Compare contributors' previous attempts, competing explanations, evidence, remediation, architectural trade-offs, safety controls and the handover to subsequent researchers.

Explore the research questions and gap-to-solution register →

Research discovery and attribution practice

For each promising finding, capture the original authors, publication date, source URL, research problem, existing approaches, evidence, unresolved limitations, alternative designs and the next falsifiable question. Link back to original sources and avoid claiming independent validation where none exists.

These are curated destinations, not a real-time feed, an endorsement or an exhaustive map of all frontier research. Current research may also be unpublished, private or absent from indexed sources. Link availability, project activity and editorial status can change.

Directory established 5 November 2026 · 27 destinations · Safety and responsible research remain foundational priorities.