Automating away the boring parts of my job with Gemini 1.5 Pro + long context windows
Inspired by recent conversations with friends, and based on a long history of automating away every job I’ve ever had (from geospatial data processing to product management): Am sharing a few ways that I’m using Gemini 1.5 and 2M+ tokens of context in Google AI Studio to automate the boring parts of DevRel and user experience research!
Reminder: you can stuff quite a bit of data into 2M+ tokens (hours of video, years of emails, full codebases, etc.) and that, over time, we expect 2M tokens → infinity, cost → $0, latency → near instant.
(1) Uploading a dated codebase (in this case, Flax 0.7.5), and a newer version of the codebase (Flax 0.8.5), then analyzing changes.
You can generate documentation changes based on the differences in code; blog posts or release notes describing the code changes; and — a favorite — update old tutorials based on the new versions of the APIs.
(2) Analyzing and prioritizing product feedback at scale by scraping GitHub and Gitlab issues, conversations in Discord and Discourse forums, social media chatter, etc.
In this example, I scraped a whole bunch of user feedback about the open-source vector database, Chroma, and compared it to feedback on a competitor’s tool (Qdrant).
(3) Ingesting a whole bunch of tutorials, documentation, and cookbooks, and using them to create scripts for YouTube and TikTok videos.
Bonus: now you can automatically generate people to be the talking heads, and can translate between many languages (all with LLMs!). This is also useful for automatically re-recording video tutorials every time the codebase gets updated…
(4) Analyzing user experience video recordings, at scale, and generating friction logs for each.
Gemini 1.5 Pro can ingest pretty lengthy videos of users attempting to accomplish tasks, and create detailed friction logs — with timestamps! — of what went well and what went wrong. Remember: this is all out of the box! No fine-tuning required, the only thing I did was upload files and prompt kindly.
Other bonus use cases:
→ Generating 1-pagers or proposals based on existing templates (ex: “I’d like to create a program for <foo> and it seems reasonable because <x,y,z>. This is my company’s template, please write.”)
→ Generating trip reports, and synthesizing feedback from conversations with customers (auto-transcribed via Google Meet).
→ Generating PRDs for small-scale changes (ex: upgrades from one version of a library to another) and creating a bug or JIRA story for each subtask. All you need to give is the old version of the library, the new version, and careful prompting along the way.
Happy prompting!
