One Poisoned Source Is All It Takes
Our daily sweep of new AI research turned up a number that should worry anyone building research agents: 54.7%.
Read the record →WTK ENGINEERING JOURNAL
Research records, newest first.
Notes, findings, failures, and factory logs remain one chronological record of what WTK is learning and changing.
WTK reviews new AI research and retains work that bears directly on current architecture questions.
The most useful catches challenge a WTK assumption, narrow a claim, or define the next experiment.
The Agents Did All the Engineering and None of the Science
In two case studies, frontier agents received six days and substantial compute, completed the engineering, and made no substantial progress on the research questions. A second model and scaffold reproduced the failure pattern.
Read the record →"Held-Out" Doesn't Mean Clean
In one multimodal fact-checking study, up to 29% of post-cutoff claims remained potentially contaminated, and the effect changed system rankings. That challenges how comparative value is measured, including in WTK.
Read the record →Verify Once, Answer Forever?
A frozen 12B model claims perfect accuracy at zero tokens by replaying verified answers. Big if true, and "if" is doing a lot of work.
Read the record →The Best Controls Might Be the Ones You Can't Ship
A new paper puts agent decision-making inside the model's hidden states. That's where WTK requires a tighter boundary, and we can now say why.
Read the record →