Episode Details

Small language models can generate three million fables at thirteen and a half cents per thousand, but the real question isn't price — it's whether the generator understands what it's making. This episode maps the parameter tiers from three billion upward, separating tasks where small models excel (controlled transformations of verified facts) from tasks where they quietly fail (inventing and labeling expert judgment). Along the way: a maritime intelligence pipeline that cut inference costs 261-fold by letting GPT-4o and o3-mini teach a 7B student, why instruction-following benchmarks reward "constraint hacking," and why a thesaurus is not a curriculum. Key sources: • HuggingFace, SmolLM3 blog, 2025 — https://huggingface.co/blog/smollm3 • model card — https://huggingface.co/Qwen/Qwen2.5-3B-Instruct • model card — https://huggingface.co/google/gemma-3-4b-it Episode 6237 — myweirdprompts.com/n/6237 (permanent short link: myweirdprompts.com/854584)