The Labs
Cerebral Palsy Research
A scoping review of biomarkers and rehabilitation response in hypotonic cerebral palsy, published as a preprint. It is not peer-reviewed. It proposes research questions. No cure or treatment is claimed.
Read the analysis →Independent research lab · Poprad, Slovakia
We research the unsolved and show the work
The Labs is the independent research practice of David Šarlák, working across medicine, mathematics, physics and robotics with open data and free compute.
We say what we ran, how it was checked and what it does not show. Negative results are reported next to progress.
Research
4 programs. Honest status.
Erdős Problems
Finite computer checks on open problems from erdosproblems.com. Problem 699 was checked for every n up to 255,667 with no counterexample. Problem 942 was computed for every n below 3.1 billion. These are finite checks, not proofs and not new theorems.
Read the analysis →
Next-Gen Refrigerants
Planned computer screening of refrigerants for heat pumps. No results yet.
Robot Learning
Planner-based manipulation with a little learning, tested in simulation on KinDER tasks. On our internal panel, 152 of 250 episodes succeed on the hardest Cluttered task (25 obstacles). Simulation only. Internal protocol, not an official benchmark result.
Read the analysis →Articles
Longer write-ups with data, sources and stated limits.
Technology
Cutting the memory of a language model
Measured, mostly negative. Throwing away most of the KV cache lost the needed facts almost every time on our test.
Read the article →
Ternary weights
Measured, negative. Our small-scale training attempts missed their own targets.
Read the article →
Network congestion control for AI Ethernet
Simulation. A one-line change helps cross-traffic and loses slightly on incast. Not affiliated with any vendor.
Read the article →
Formula 1 modelling
Simulation and public race data. Gains are within model error. No claim about real cars.
Read the article →Health
Colorectal cancer: what public data can tell us
Hypotheses from public data, with cited literature kept separate. Not medical advice.
Read the article →
Migraine: two attempts with public genetics
One failed control and one small finding on a known gene. Not medical advice.
Read the article →Contact
Questions, corrections, data or collaboration - write to us.
General contact
About the site, the articles or the data behind them.
Join the research
Want to take part? Tell us what you would like to work on and how you would like to contribute. Joining is not automatic and not every request can be accepted.