§ feed · storyline
Learning from Real-World Experience
Perplexity Research reduces tool-call failures by 21% in its Computer model by post-training on user corrections and observed tool failures collected from real-world interactions.
Perplexity Research describes post-training a Perplexity Computer model using user corrections, successful behaviors, and tool failures. In a later comparison, tool-call failures fell from 2.24% to 1.77%, a statistically significant 21.2% relative reduction; the authors note this was not a direct comparison with stock GLM 5.2 online.
§ sources1 publication · timeline below
- perplexity.aiLearning from Real-World Experienceprimary