2026-06-08原文2 min readUpdated: 2026-06-08

Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics

This position paper argues that AI research should focus on studying the training dynamics that produce model behavior, rather than only analyzing final models post-hoc. It calls for a science of AI that can predict, intervene, and design training procedures to reliably achieve desired properties.

SourcearXiv AIAuthor: Stella Biderman, Mohammad Aflah Khan, Niloofar Mireshghallah, Catherine Arnett, Fazl Barez, Naomi Saphra

[2606.06533] Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics

[Submitted on 3 Jun 2026]

Title:Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics

View a PDF of the paper titled Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics, by Stella Biderman and Mohammad Aflah Khan and Niloofar Mireshghallah and Catherine Arnett and Fazl Barez and Naomi Saphra

View PDF

Abstract:What would it mean to have a scientific understanding of AI? Models are not static objects: they are snapshots of time-evolving processes shaped by data, objectives, architectures, and optimization dynamics. Yet much of AI research treats models as fixed artifacts, analyzing behaviors after training rather than asking why they emerge. This position paper argues that a science of AI must move beyond post-hoc fixes and study the training dynamics that produce model behavior. Such a science should support progressively stronger forms of understanding: predicting outcomes from early training signals, intervening when trajectories go wrong, and ultimately designing training procedures that more reliably produce desired properties. Scaling laws have made prediction routine for loss; the challenge is extending this success to capabilities, biases, robustness, and safety-relevant behaviors. We articulate requirements for such theories grounded in the history and philosophy of science, examine progress in mechanistic interpretability, fairness, memorization, and simplicity bias, and identify concrete open problems.

Comments: Accepted as an oral to the ICML: this https URL

Subjects:

Artificial Intelligence (cs.AI); Computation and Language (cs.CL)

Cite as: arXiv:2606.06533 [cs.AI]

(or arXiv:2606.06533v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2606.06533

arXiv-issued DOI via DataCite

Submission history

From: Stella Biderman [view email] [v1] Wed, 3 Jun 2026 17:58:14 UTC (97 KB)

Full-text links:

Access Paper:

View PDF

TeX Source

view license

Current browse context:

cs.AI

new | recent | 2026-06

Change to browse by:

cs cs.CL

References & Citations

NASA ADS

Google Scholar

Semantic Scholar

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)