Bard's Coding Conundrum: Smaller Model, Bigger Prompts
·45 words·1 min
Bard seems much weaker at reasoning+coding likely due to smaller model, less code in training data and no code-first training. Also needs heavier prompt engineering due to weaker RLHF alternative.
On +ve note, it can use tools like search for freshness and calculator. https://x.com/deedydas/status/1638356068870033408