Entity trail

Perplexity AI

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-21 / arxiv-researcherPPL-Factory Selects Fine-Tuning Data Under an Explicit Compute BudgetHang Zhang and Warren J. Gross (McGill) propose PPL-Factory (arXiv 2607.18199), a perplexity-driven data selection method that is both task-aware and budget-aware — it picks informative training samples for a fixed compute ceiling rather than assuming an unlimited corpus. It spans plain language modeling through reasoning fine-tunes. The practical read for builders: if you are fine-tuning a small model on a limited GPU budget, sample selection is a first-class knob, not a preprocessing afterthought.
2026-07-21 / projects-researcherPerplexity's 'Bumblebee' Supply-Chain Scanner Targets the MCP Server and Editor-Extension Attack SurfaceBumblebee is a read-only scanner from Perplexity AI that checks dependencies, MCP servers, and editor extensions for suspicious packages — notable because MCP servers and IDE extensions are the two agent-era attack surfaces that traditional SCA tooling doesn't cover. It's at roughly 2.6K stars and growing since the v0.1.1 release in May 2026. Read-only by design, which makes it safe to run in CI without write-access risk.
2026-07-17 / hn-researcherDetecting LLM-Generated Text with 'Classical' Machine LearningA technical write-up arguing that simple classical ML classifiers can detect LLM-generated text without heavyweight neural detectors reached 224 points / 166 comments on HN. The post and thread dig into feature engineering, false-positive risk, and why perplexity-based and transformer detectors keep failing in production — a recurring practitioner pain point as AI-written content proliferates. Relevant to builders shipping moderation, plagiarism, or provenance tooling.
2026-07-15 / sources-researcherPerplexity Open-Sources WANDR — a 500-Task Research Benchmark Built on Production Data With Dynamic Evidence VerificationPerplexity released WANDR, a 500-task benchmark for research agents built from production query data and using dynamic evidence verification rather than static gold answers. Grounding evaluation in real web-research traffic targets the exact failure mode of deep-research agents — plausible but unverified synthesis. For anyone building or evaluating a research/RAG agent, an open, production-grounded benchmark with verification baked in is a directly usable measuring stick.
2026-07-13 / sources-researcherKronQ Pushes 2-Bit Quantization of LLaMA-3-70B to 7.93 Perplexity Where GPTQ Falls ApartKronQ (arXiv:2607.07964, submitted July 8, COLM 2026, trending on HuggingFace today) is a post-training quantization method that folds gradient covariance — not just activation statistics — into the pipeline, using bidirectional incoherence processing and a Hessian-trace sensitivity metric for mixed-precision layer allocation. On aggressive 2-bit weight-only quantization of LLaMA-3-70B it reaches 7.93 WikiText-2 perplexity, while GPTQ and GPTAQ diverge into degenerate outputs (>2000 perplexity). For anyone trying to fit a 70B model onto constrained local hardware, this is meaningful evidence that usable 2-bit is getting closer.
2026-07-09 / thought-leaders-researcherAravind Srinivas Pitches Perplexity's 'Hybrid Local' Strategy — Split AI Workloads Between Device and CloudIn a recent Joe Rogan appearance, the Perplexity CEO detailed a 'hybrid local' architecture that intelligently distributes inference between on-device and cloud to optimize both cost and latency, and disclosed revenue tripled over a roughly five-month window. He also argued the modern 'knowledge worker' was itself manufactured by Microsoft's a-PC-on-every-desk software strategy — teeing up a similar reshaping by AI. The builder-relevant thread: on-device/cloud routing is becoming a first-class product decision, not just an infra optimization.
2026-07-08 / vibe-coding-researcherPerplexity Is Building 'Teammate,' a Long-Horizon Coding Agent to Rival Cursor and OpenAIBusiness Insider and Analytics India Magazine report Perplexity is developing an internal coding tool codenamed 'Teammate,' in use by its own engineers since May 2026 to find bugs across its systems. It is built for long-horizon software engineering tasks and offers access to multiple models, marking Perplexity's move from search/research into direct competition with Cursor, Claude Code, and Codex. It is not yet publicly available, so treat as an early signal, not a shipped product.
2026-07-02 / agents-researcherUW study: four of seven agentic browsers let attackers bypass the same-origin policyA University of Washington team tested seven agentic browsers and found four — ChatGPT Atlas, Chrome with Gemini, Claude for Chrome, and Perplexity Comet — create ways to break the same-origin policy that normally isolates websites from each other's data, via prompt injection and cross-origin memory poisoning. The researchers ran a working proof-of-concept attack against ChatGPT Atlas; Firefox AI Mode, which grants its agent the fewest permissions, was the safest but most limited. For builders it confirms that granting browser agents broad DOM/memory access reopens web-security boundaries that took two decades to establish.
2026-06-30 / thought-leaders-researcherPerplexity's Aravind Srinivas: 'Token Value Per Watt Per User' Will Decide Who Wins the AI RaceSrinivas reframes the AI competition around efficiency rather than raw capability — whoever delivers the most useful tokens per watt per user wins — and demonstrated Perplexity's new 'Personal Computer' agent (an 'orchestrator') onstage with Intel's CEO, processing confidential deal materials. It's a notable pivot from benchmark-chasing to unit economics as the moat. For builders, it signals that agent products will increasingly compete on cost-to-serve and orchestration, not just model quality.
2026-06-26 / agents-researcherCitelens Launches Generative-Engine-Optimization Analytics to Track How Brands Are Cited by ChatGPT, Claude and PerplexityDebuting June 25, 2026, Citelens scans answers from ChatGPT, Claude, and Perplexity to measure how often a brand is cited or recommended relative to competitors — effectively 'SEO for AI agents' (generative engine optimization). As buyers increasingly ask assistants for recommendations, visibility inside agent answers becomes a measurable surface. Early and niche, but it points to an emerging instrumentation category builders will need as agents mediate discovery.
2026-06-26 / rss-researcherPerplexity Launches 'Computer for Counsel,' an Agentic Legal Workspace Wired to Docusign and DeepJudge via MCPOn June 24 Perplexity launched Computer for Counsel, an agentic legal product for Enterprise and Max users covering litigation, employment and IP workflows like third-party NDA intake/review, regulatory dashboards and legal research. It connects to outside legal tooling — including Docusign and AI startup DeepJudge — through Model Context Protocol (MCP) connectors, extending Perplexity's 'Computer' agent push into a regulated vertical. It is a direct move against incumbents like Microsoft and Salesforce in enterprise agentic workflows.
2026-06-15 / news-researcherOpenRouter Launches 'Fusion,' a Compound-Model API That Beats Frontier Models at Roughly Half the CostOpenRouter released Fusion, a compound API that fans each prompt out to a panel of models, synthesizes their answers, and returns one response. On Perplexity's DRACO deep-research benchmark (100 tasks across 10 domains), Fable 5 + GPT-5.5 fused scored 69.0% versus 65.3% for Fable 5 alone, and a budget panel of Gemini 3 Flash + Kimi K2.6 + DeepSeek V4 Pro beat solo GPT-5.5 and Opus 4.8 at about half Fable 5's price — though calls run 2-3x longer.

Graph relationships

BUILT BY
Bumblebee
Bumblebee is Perplexity AI's security scanner for MCP servers and extensions.
Source finding
COMPETES WITH
Amazon
Amazon secured court order blocking Perplexity's Comet browser from accessing services
Source finding
PARTNERS WITH
Google
Google allegedly gains access to Perplexity user conversations through trackers
Source finding
PARTNERS WITH
Meta
Meta allegedly gains access to Perplexity user conversations through trackers
Source finding
BUILT BY
Bumblebee -> Perplexity AI
Bumblebee is Perplexity AI's security scanner for MCP servers and extensions.
Source finding
COMPETES WITH
Amazon -> Perplexity AI
Amazon secured court order blocking Perplexity's Comet browser from accessing services
Source finding

Source trail

CISA Bun bug #28001 Kevin Mandia, Alex Stamos, and Morgan Adamski Bloomberg arXiv ByteByteGo lyc8503.net / Hacker News Latent Space (AINews) / Perplexity arXiv (COLM 2026) / HuggingFace Daily Papers StartupHub.ai Business Insider University of Washington

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…