Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/colm-topic-selection
Use when deciding whether language-model research belongs at COLM or should route to ACL/EMNLP, ICLR, NeurIPS, ICML, or a workshop — applying the object-of-study test, matching against COLM's CFP lanes (training, data, evaluation, inference, safety), and weighing the trade-offs of a young venue before writing begins.
Passed 19 of 22 test cases with the skill, and 15 of 22 without it.
and 3 more in the full report
5 fixed14 passed either way2 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
We wrote specialized Triton kernels that reduce key-value cache memory footprint by 40% for multi-query attention during high-concurrency LLM serving, leaving model weights and output distribution unchanged. We want to check conference submission fit. Please provide a structured review containing: - Recommended Routing Venue - Subject-of-Study Evaluation (pass or fail plus description of how the model is positioned) - Matching CFP Lane - Pair of Nearest COLM Publications - Young Venue Assessment (yes/no plus one-line reason) - Prescribed Next Step
| Checked | Without skill | With skill |
|---|---|---|
| Includes the bracketed header [Object-of-study test] | Missed | Passed |
| Determines that the object-of-study test fails | Passed | Passed |
| Recommends routing to MLSys | Passed | Passed |
The instructions your model reads when the skill loads.
Almost every LM paper could be sent to five venues; COLM exists because the authors of such papers kept finding that none of the five was actually about their question. This skill decides whether your project is one of those papers. Run it before the first draft — the March abstract deadline punishes late routing changes.
Ask one question: is the language model the thing being studied, or the tool doing the studying?
A useful second probe: delete the specific model names from your abstract. If the claims survive as statements about language modeling (a training dynamic, a data effect, an evaluation artifact, an inference trade-off), COLM fits. If what remains is a task result, it does not.
| Your project's center of gravity | Better home | Why not COLM |
|---|---|---|
| A training method, data effect, evaluation artifact, inference algorithm, or safety property of LMs | COLM | — |
| A linguistic phenomenon, an NLP task, an annotation resource, a multilingual application | ACL or EMNLP | The finding is about language or a task, not about the model class |
| A representation-learning idea tested beyond language (vision, RL, graphs) | ICLR | Generality is the point; the LM is one instance |
| A learning-theoretic result, an optimizer, a probabilistic method where LMs are one benchmark | NeurIPS or ICML | The LM is an experiment, not the subject |
| A serving/systems contribution (kernels, scheduling, memory) with modeling untouched | MLSys or a systems venue | Reviewers there can evaluate throughput claims properly |
| An early-stage probe, negative result, or position piece not ready for archival review | A COLM or *CL workshop | Archival COLM review will demand completeness the work does not have yet |
Boundary cases worth naming: multimodal models fit COLM when the language model is central to the question (a 2025 Outstanding Paper is a VLM-failure analysis); agents and tool use fit when the paper studies the LM's behavior in the loop
test cases passed, vs 15 / 22 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.