亚洲量化策略：资产管理中的生成式AI：与Perplexity对话——关键要点及回放

研报英文原文证据摘录

15 September 2026

Asia Quantitative Strategy

Gen AI in Asset Management: Chat with Perplexity - Key

Takeaways & Replay

Rupal Agarwal

Cheng Zhang, CFA, CQF

We kick-started Season 2 of our Gen AI in Asset Management Innovator Series last week

with Perplexity and hosted, Jeff Grimes, who heads their Computer Capabilities product. In

this short note, we present key takeaways from the session. For replay - click here.

Perplexity, the AI orchestrator: Perplexity's core differentiation lies in its ability to

orchestrate across multiple AI models (24 as of now across frontier and open source)

while maintaining governance, transparency, security, and cost efficiency. As the number

of competitive models continues to rise, value may increasingly accrue to platforms that

intelligently orchestrate models rather than to any single model provider. This reduces

vendor lock-in while improving cost efficiency and performance.

Perplexity Computer: The platform combines multiple frontier and open-source

AI models, enterprise data connectors, automation capabilities, persistent memory,

collaborative workspaces, and hybrid cloud/local computing infrastructure into a single

operating framework. There are four core pillars of Perplexity Computer: 1) Trusted Data

- Either user can connect to their existing licensed data sources through an extensive

connector ecosystem or use datasets provided by Perplexity (through pre-negotiated

terms), 2) Work Execution - Computer can output emails, spreadsheets, presentations,

documents, PDFs, markdown files, deep research reports etc. It also interacts with

connected applications and systems via MCP connectors whenever users require, 3)

Auditable Answers - The system displays source references, identifies the original filing

or transcript, and reconstructs the calculation path for derived metrics, 4) Take work

everywhere-The user’s work progress in Excel, Word, PowerPoint is all synced across local

and cloud; allowing users to set up and monitor tasks on the go.

5 powerful use-cases: 1) Model Council: This capability allows multiple models to

independently evaluate the same scenario/task and highlight areas of consensus and

disagreement.…