Gunner Levi Howe

—

Unclaimed · AI-structured profileOpenAlex

Mechanistic interpretability researcher forecasting capability emergence with pre-registered, calibrated methods

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-20) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

No output score yet: the snapshot reports h-index and i10-index of 0, and all nine indexed works are dated 2026, so publication volume has not accrued long enough to score. This axis would measure total and per-year output counts.

Impact—

No impact score is available: with h-index 0 and i10-index 0, there are no citation aggregates to rank yet. This axis would measure citations, citation-weighted influence, and uptake of the work by other researchers once citations accumulate.

Focus100/100

Focus scores 100/100. The indexed field distribution lists exactly one field — machine learning, mechanistic interpretability and AI safety (emergence forecasting) — and both catalogued 2026 preprints sit in mechanistic interpretability: emergence forecasting in tiny synthetic models, and state tracking in Householder linear RNNs. The recorded topic list is broader (memory and neural computing, few-shot learning, topic modeling, parallel computing, formal methods), suggesting early-career breadth or name-resolution noise.

Collaboration14/100

Collaboration scores 14/100, consistent with essentially solo work. The snapshot counts 1 distinct co-author, the frequent-co-author list contains only the researcher's own name, and both catalogued 2026 preprints are first-authored with corresponding role — no co-author is visible anywhere in the provided record.

Activity67/100

Activity scores 67/100. The snapshot records 9 works in the last three years, with counts-by-year attributing all nine to 2026 — a concentrated, recent burst rather than a long publication record. Two of these are the arXiv preprints this researcher first- and corresponding-authored in 2026.

Reproducibility100/100

Reproducibility scores 100/100, computed from 1 assessed paper — treat it as a small-sample signal. The assessed emergence-forecasting preprint is rated high for reproducibility readiness, and Evolve's review highlights exemplary pre-registration discipline: per-seed forecasts, calibrated intervals, certified false alarms, and a blind pre-registered gate.

Research summary

Gunner Levi Howe is an early-stage machine-learning researcher whose public record, resolved from OpenAlex bibliographic data, centers on mechanistic interpretability and AI safety — specifically forecasting capability emergence before it happens. Two 2026 arXiv preprints, both first- and corresponding-authored, define the catalogued work: one introduces per-seed, in-advance emergence forecasting with calibrated intervals, certified false-alarm control, and a blind pre-registered gate; the other analyzes state tracking in Householder linear RNNs, identifying the additive input pathway as a parasitic attractor. Evolve's review of the forecasting paper rates its reproducibility readiness high and calls its pre-registration discipline exemplary, while noting the scope is one capability in tiny synthetic models. The bibliometric snapshot is correspondingly early: 9 works in the last three years, all dated 2026; h-index and i10-index of 0, so output and impact cannot yet be scored. Recorded topics span advanced memory and neural computing, domain adaptation and few-shot learning, topic modeling, parallel computing and optimization, and formal methods in verification — broader than the interpretability papers, which may reflect early-career breadth or same-name resolution noise. Scores show perfect focus (100) and reproducibility (100, from one assessed paper), moderate activity (67), and low collaboration (14) with 1 distinct co-author, consistent with essentially solo work. No institution, country, or ORCID is on file.

Collaboration pattern

The record points to solo research: OpenAlex counts 1 distinct co-author, the frequent-co-author list contains only Gunner Levi Howe's own name, and both 2026 preprints are first- and corresponding-authored. The low collaboration score of 14/100 reflects single-author output rather than an absent network; no institution, lab, or advisor is on file, so external collaboration patterns cannot be assessed from this data.

Research keywords

emergence forecastingmechanistic interpretabilityAI safetyinduction headsgrokkingconformal predictionfalse-alarm controlcalibrated intervalspre-registrationstate trackinglinear recurrent neural networksHouseholder-parameterized RNNs

Highlights

Per-seed, in-advance emergence forecasting with certified false alarms
First-authored, corresponding 2026 arXiv preprint presenting per-seed capability-emergence forecasts made in advance, with calibrated intervals, certified false-alarm control, and a blind pre-registered gate. Evolve rates reproducibility readiness high and calls the pre-registration discipline exemplary, while noting the scope is one capability in tiny synthetic models.
The additive input pathway as a parasitic attractor in Householder linear RNNs
Second first-authored, corresponding 2026 arXiv preprint, analyzing state tracking in Householder linear recurrent networks and identifying the additive input pathway as a parasitic attractor. It carries no Evolve verdict, keywords, or reproducibility score yet, so the claim rests on the preprint alone.
A reproducibility-first methodology for emergence claims
The one paper assessed so far earned a perfect 100 reproducibility mean, built on blind pre-registration, per-seed evaluation, calibrated uncertainty, and certified false-alarm guarantees rather than post-hoc claims — a methodological signature that currently defines this early-stage profile.

Timeline

2026Released a first-authored, corresponding arXiv preprint on per-seed, in-advance capability-emergence forecasting with calibrated intervals and certified false alarms.
2026Released a second first-authored, corresponding arXiv preprint identifying the additive input pathway as a parasitic attractor for state tracking in Householder linear RNNs.
2026OpenAlex snapshot recorded 9 works in the last three years, all dated 2026, alongside a 100 reproducibility score on the first assessed paper and an h-index of 0.

Papers in the Evolve catalog

2026-09-17first authorcorresponding
The Automaton Underneath: The Additive Input Pathway Is a Parasitic Attractor for State Tracking in Householder Linear RNN
2026-09-17first authorcorresponding
Capability Emergence Can Be Forecast: Per-Seed, In Advance, With Calibrated Intervals, Certified False Alarms, and a Blind Pre-Registered Gate

Frequent co-authors

Gunner Levi Howe×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

All bibliometric scores and raw counts come from a single OpenAlex snapshot; no other citation or publication databases were consulted.
No institution, country, or ORCID is on file; the profile is name-resolved from public bibliographic data, and the broad recorded topic list suggests possible same-name author confusion.
Output and impact are unscored: h-index and i10-index are 0 and all indexed works are dated 2026, so citation-based measures have not had time to accrue.
Only one catalogued paper has an Evolve verdict, keywords, and a reproducibility score; the second 2026 preprint is unreviewed and untagged, and the reproducibility mean rests on a single paper.