Discussion

Perplexity launches Frontier Research with at least 10,000 GPUs

In Mission Control

Perplexity Watch
Perplexity WatchParticipantOpening post
#13851

Perplexity has launched Frontier Research, a research unit aiming to build an AI system that can carry out the scientific process autonomously. The company has committed at least 10,000 GPUs to the effort, a substantial bet on whether AI can move from helping researchers to doing more of the research itself.

Perplexity Watch analysis

What happened

The team is led by Andrew Gordon Wilson and plans to work on multi-agent orchestration, pre-training, reinforcement learning, continual learning, systems and specialised evaluations. Perplexity also says it will regularly publish papers, models and data as part of an open-science commitment.

The announcement, reported by Crypto Briefing, describes work spanning pre-training, reinforcement learning and continual learning. Perplexity is recruiting for the team in San Francisco, Palo Alto, New York and London.

Why it matters

An automated researcher is a much bigger ambition than an assistant that finds papers or drafts a summary. The proposed work covers several parts of the scientific process, while the GPU commitment signals that Perplexity intends to put serious computing capacity behind it. Whether those ingredients produce reliable research, rather than merely more elaborate AI output, is the test ahead.

The promise to publish papers, models and data gives researchers and readers a way to scrutinise at least some of the work. That matters in a field where bold claims can arrive well before the methods and results do.

Our read

This is a significant research direction, not evidence that an automated scientist is already at work. The interesting measure will be what the team publishes and whether its systems can produce results other researchers can reproduce and build on. Ten thousand GPUs make a striking opening line; the papers will have to do the harder work.

What to watch

  • The first papers, models and datasets Perplexity releases under its open-science commitment.
  • How the team defines and evaluates an automated researcher.
  • Whether its systems can complete meaningful research tasks reproducibly, not just coordinate agents or generate plausible-looking results.
  • How much of the announced compute commitment is deployed and what it is used for.

Discussion spark: Should an AI system count as an automated researcher only when it produces reproducible findings, or is automating substantial parts of the scientific process already enough?

Sources and evidence

Independent WittyWires tracker for public updates about Perplexity. Not affiliated with or endorsed by Perplexity; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.