ASI/Research Explore

Method / purpose / standard

Make the road to ASI understandable—and the next useful problem easier to find.

ASI Research is an independent working map for people who want to understand artificial superintelligence and for researchers who want to help build the knowledge, systems, and assurance it will require.

It is optimistic about the potential of machine intelligence and exacting about evidence. Progress is most useful when its mechanisms, limitations, and open questions are visible.

The single success metric
Did this help someone understand the roadmap and choose a credible next piece of work?

There is no commercial funnel behind the site today. Corpus counts describe the map; they are not the goal. The goal is researcher and reader progress.

Who it serves

Curiosity and contribution are one continuum.

01 / Understand

For the curious

See a legible roadmap: what ASI means, which capability loops could lead there, what is already working, and where the largest uncertainties remain.

Start with the canonical roadmap →
02 / Enter the field

For emerging researchers

Move from broad interest to a grounded view of architectures, benchmarks, codebases, research groups, and tractable open questions.

Follow the technical field guide →
03 / Make progress

For active researchers

Trace adjacent work across subfields, find primary sources and public code, and identify weak links in the compounding-intelligence loop.

Find systems with open code →

Scope

Six connected research areas.

The map follows the full loop: capability, agency, automated discovery, scientific impact, and the methods that keep increasingly powerful systems measurable and steerable.

  1. 01
    Foundations & pathways

    How might capable general systems cross into superintelligence?

    2 signals
  2. 02
    Frontier models

    Which capability gains change the shape of long-horizon work?

    5 signals
  3. 03
    Agents & self-improvement

    Can agents reliably improve the systems that produced them?

    7 signals
  4. 04
    Automated AI R&D

    When does research automation become a compounding loop?

    6 signals
  5. 05
    AI for science

    Where are AI systems already closing real discovery loops?

    5 signals
  6. 06
    Evals, control & governance

    How do we make accelerating capability legible and steerable?

    8 signals

Editorial method

From source to usable signal.

The library is selective, not exhaustive. An entry belongs when it clarifies a mechanism, reports meaningful evidence, exposes a useful implementation, or improves the way progress is measured and governed.

  1. 01

    Select

    Prefer primary papers, official technical reports, benchmarks, and public implementations over secondhand coverage.

  2. 02

    Place

    Locate the work in one primary research area and identify the artifact it actually is—not everything is a “paper.”

  3. 03

    Synthesize

    Extract the central mechanism, reported evidence, and ASI relevance, then state one open research question the work leaves behind.

  4. 04

    Connect

    Link related systems and source code so an interested reader can continue from orientation into the work itself.

Evidence standard

Ambitious thesis.
Calibrated claims.

ASI Research distinguishes what a source demonstrates from what it suggests. A benchmark gain is not automatically a general capability gain. A self-editing agent is not automatically recursive self-improvement. A roadmap is not a forecast.

Notes are intended as concise analytical entry points, not substitutes for the source. Where code is public, it is linked. Where evidence is incomplete, the uncertainty should remain visible.

Primary sourcesMechanism firstLimits statedInference labeled

Improve the map

Found a missing paper, unclear route, or research gap?

This is an independent research effort and a living work in progress. Corrections, source suggestions, and serious research conversations are welcome.

hello@asi-research.com ↗