Treasure Hunt Real Time Targeting Of The Long Tail Using Training Time Markers 2025 06 18
Source
cohere.com
Author
Cohere
Date
Context
Treasure Hunt teaches a LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → to recognize explicit markers for task types and desired output characteristics. These markers give users controls the model has encountered during training, addressing requests that appear infrequently in its training data. The model also learns to infer the markers, so users can omit them. Cohere's publisher abstract reports relative improvements of up to 14.1% on underrepresented tasks such as code repair.
The contribution is a training method that makes later instructions more effective for neglected use cases. The reported gain applies to the authors' evaluations.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
postCohere releases open-weights translation model North Small Translate
articleThe Data Provenance Initiative A Large Scale Audit Of Dataset Licensing And Attribution In Ai 2023 10 25