Improvements to data analysis in ChatGPT

Original Article ↗ Hacker News Discussion ↗

OpenAI’s upgraded data analysis features for ChatGPT, including direct access to Google Drive and OneDrive files, are seen as a significant boost for non-technical users who want to explore and visualize data without writing code. Commenters debate how this compares to Microsoft 365 Copilot and Google’s Gemini tools, whether enterprises will trust third-party AI with sensitive corporate data, and how much of analysts’ work can realistically be automated. Many expect “wrapper” startups built on OpenAI’s APIs to be squeezed as OpenAI folds popular use cases into its own product, while noting that domain expertise, governance, and transparent transformation logic remain critical limitations for AI-driven analytics.

Comparison to Microsoft/Google and native suites

  • Some see the feature as overlapping with Microsoft 365 Copilot and Google Gemini, questioning why a business wouldn’t just use native tools.
  • Others report ChatGPT being much better at data analysis than Office Copilot, which is described as immature outside of Teams.
  • The new Google Drive/OneDrive integration is noted as an important first‑party connection to users’ cloud documents.

Alternatives and ecosystem

  • A range of competitors is mentioned: standalone “chat with your data” tools (e.g., julius.ai), database‑centric tools (patterns.app, findly.ai), AutoML platforms (Akkio), visualization‑focused apps (minard.ai), and all‑in‑one data stacks with AI assistants (definite.app).
  • Some tools target embedding an “AI data scientist” inside customer apps, not just analyst self‑service.

Use cases, UX, and limitations

  • Many view this as “last‑mile” ad‑hoc analysis: upload a file, ask questions, get charts/tables without writing Python.
  • It’s valued for users who can’t or won’t code, or who lack permission to run code at work.
  • Others prefer using LLMs to generate explicit code (Python/SQL) for transparency and reuse in repos.
  • Complaints include basic/ugly charts and confusion about modes; clarification that “Data Analysis” is now just part of normal chats with any model.

Technical design, reliability, and trust

  • Under the hood, it’s said to generate deterministic Python, making transformations reproducible.
  • Critics worry about black‑box transformations and missing lineage; they argue for logged, declarative steps (e.g., SQL/ibis‑style primitives).
  • Skeptics doubt LLMs can correctly interpret ambiguous schemas and real‑world analytics nuances; proponents say newer models handle many everyday spreadsheet tasks well.
  • One user reports internal errors (e.g., AceInternalException) and difficulty finding a bug‑report channel.

Privacy, security, and enterprise adoption

  • Strong debate over sending corporate data to third‑party clouds and OpenAI: some see it as reckless; others note most corporate data already lives in major clouds.
  • Concerns raised about cloud security incidents, subpoena access, and whether confidential computing truly prevents provider access.
  • Several think this is attractive for individuals and small businesses; some doubt large enterprises will hand over sensitive data.

Jobs and startups impact

  • Discussion on whether analysts are being automated away: many say no, arguing that domain expertise and broader job context remain essential.
  • Some note executives may still use such tools as justification for downsizing line workers.
  • Broad agreement that “wrapper” startups with thin moats are at risk as OpenAI folds popular patterns directly into ChatGPT.