Loading the PDF…
Your browser could not open the document in the reader.
Download the PDF Back to the record
You can still read this work as a PDF.
Mahendrakar, P. (2026). A Query Update Is Not a Memory Write: What Test-Time Training on a Long Context Changes, Why Its Advantage Over Thinking Tokens Narrows Outside the Setup That Produced It, and the Controls That Would Credit the Gradient [Preprint]. Life of Research. https://doi.org/10.5281/zenodo.24344793
Copy APA
@misc{lor2432026, title = {A Query Update Is Not a Memory Write: What Test-Time Training on a Long Context Changes, Why Its Advantage Over Thinking Tokens Narrows Outside the Setup That Produced It, and the Controls That Would Credit the Gradient}, author = {Pranay Mahendrakar}, year = {2026}, publisher = {Life of Research}, doi = {10.5281/zenodo.24344793}, url = {https://lifeofresearch.com/preprints/243}, keywords = {long-context language models, test-time training, inference-time compute, attention dilution, chain-of-thought, parametric memory, evaluation methodology, compute matching}, note = {LOR-00243} }
Copy BibTeX More on the record page