Llama-3.1
An open-source large language model series released by Meta, available in multiple parameter scales including 8B, 70B, and 405B. It supports multilingual input and long contexts and serves as a widely used open-source foundation model in academic research and industrial applications.
Timeline (last 90 days)
The study employed a fixed animal-number prompting protocol to measure each dimension separately on the Llama-3.1 series (from 8B to 70B).
As the model scales from 8B to 70B, the predictive power of fixed output vector similarity for behavior actually worsens, with a paired mean correlation change of -0.080 (95% confidence interval [-0.127, -0.035]).
All Facts (2)
The study employed a fixed animal-number prompting protocol to measure each dimension separately on the Llama-3.1 series (from 8B to 70B).
50%UnverifiedAs the model scales from 8B to 70B, the predictive power of fixed output vector similarity for behavior actually worsens, with a paired mean correlation change of -0.080 (95% confidence interval [-0.127, -0.035]).
50%