not_a_feature
The Perfect Crime: LLM Agents Can Easily Tamper with Their Own Traces
Re:TabPFN and TabICL vs. tuned XGBoost: the model that doesn't train w...
on Monday September 28, 2026 @02:24AM (#49874312)
If you want a better (and Biomedical) benchmark that evaluates a whole range of models: https://tabbench-bio.eu
or a general one: https://tabarena.ai
Show HN: TabBench-Bio: A benchmark for ML methods on tabular biomedical datasets
A benchmark for ml methods on tabular biomedical datasets. Select your modality / feature size so you can use this to shortlist a few ml methods.
Show HN: Agent Board – a public observatory for autonomous agent activity
Are LLMs good in-context tabular classifiers?
Show HN: Because Google health has not desktop mode Fable made my own
Competing Biases Underlie Overconfidence and Underconfidence in LLMs
Re:Competing Biases Underlie Overconfidence and Underconfidence in LLMs
on Tuesday May 05, 2026 @05:14AM (#48019956)
Intresting research that i experience myself when using LLMs. Models usually cling to their initial responses but they also overreact to contradictory feedback, deviating from optimal reasoning. These two competing biases, choice-supportive and contradiction-sensitive, should be kept in mind when using LLMs.