§ feed · storyline
Learning from Real-World Mistakes
Perplexity Research post-trained its Computer model on user corrections and tool failures, cutting tool-call errors by 21.2% relative, though broader task-success gains were not confirmed.
Perplexity Research describes post-training a Perplexity Computer model using user corrections and tool failures. The reported online evaluation reduced tool-call failures from 2.24% to 1.77%, a statistically significant 21.2% relative reduction, while noting that fewer tool errors did not necessarily translate into broad task-success improvements.
§ sources1 publication · timeline below
- perplexity.aiLearning from Real-World Mistakesprimary