Unlocking Reasoning
The conversation delves into the significance of coding and math in enhancing reasoning within the startup ecosystem. Joe highlights how more data and logical structures, like those found in scientific papers, contribute to model performance. Interestingly, the smaller 8B model outperformed the larger 70B model, challenging common assumptions about model size and capability.In this clip
From this podcast
Training Data
Meta’s Joe Spisak on Llama 3.1 405B and the Democratization of Frontier Models | Training Data
Related Questions