Search results for "DRIFT"
2026-04-25
04:54

Perplexity Discloses Web Search Agent Post-Training Method; Qwen3.5-Based Model Outperforms GPT-5.4 on Accuracy and Cost

Perplexity uses SFT followed by RL with Qwen3.5 models, leveraging a multi-hop QA dataset and rubric checks to boost search accuracy and efficiency, achieving best-in-class FRAMES performance. Abstract: Perplexity's post-training workflow for web-search agents combines supervised fine-tuning (SFT) to enforce instruction-following and language consistency with online reinforcement learning (RL) via the GRPO algorithm. The RL stage uses a proprietary multi-hop verifiable QA dataset and rubric-based conversational data to prevent SFT drift, with reward gating and within-group efficiency penalties. Evaluation shows Qwen3.5-397B-SFT-RL achieving top FRAMES performance, 57.3% accuracy with a single tool call and 73.9% with four calls at $0.02 per query, outperforming GPT-5.4 and Claude Sonnet 4.6 on these metrics. Pricing is API-based and excludes caching.
More
16:24

Solana cofounder toly: a base-layer stablecoin should be built that can only be frozen with authorization from the court

Solana co-founder toly noted that the industry needs a stablecoin that can only be frozen under a court order, opposing other freeze factors. He suggested that the protocol issue a stablecoin with custom freeze strategies on the base layer and strengthen security measures. This view stems from a recent response by Circle to the Drift protocol hack incident, sparking discussions about centralized stablecoins.
More