4 ms·
Only on specific domains, these models don't become generalists like GPT-4, they can become task experts for a single task.
by visarga 3y ago
Only on specific domains, these models don't become generalists like GPT-4, they can become task experts for a single task.
- famouswaffles 3y agoFine-tuned MedPalm is worse than GPT-4 on most Medical Challenge Tests. Fine-tuned Minerva is much worse on arithmetic benchmarks. The LLM space is just different. There's no guarantee a fine-tuned model will beat a bigger generalist one.