OverflowAI: Is StackOverflow finally fighting back?

Jul 31, 2023

Happy Monday y’all,

It’s a rainy afternoon here in Oxford - the perfect setting for a cuppa* and a 5-minute peak into what’s been happening in AI this week. Enjoy!

*= tea ☕, for the non-Brits 🙃

AI News

🗣️ StackOverflow announces OverflowAI for semantic search, automatic question-answering, and GenAI-assisted questions - Page views of StackOverflow have dropped by 50% in the last 1.5 years, partly due to the rise of AI coding assistants like Copilot and Code Interpreter. We’ve known for a while that StackOverflow is working on AI (back in March, their Director of Comms said “10% of our company is working on features and applications leveraging GenAI”), and this week StackOverflow’s CEO announced OverflowAI, a new suite of “community-first” AI tools trained on StackOverflow data. OverflowAI will include features like:
- Semantic search (instead of traditional lexical search), for helping users quickly find the answers they need even when they’re not sure how to word their questions
- Automatic question-answering based on StackOverflow answers
- Assisted question-drafting, for situations when AI can’t provide the right answer
🤔 Who are Anthropic? - Last week, 7 AI companies signed an AI pledge at the White House. Five of these are household names (Google, OpenAI, Microsoft, Meta, Amazon), but the other two (Inflection AI and Anthropic) are relative newcomers. Wanna know why these newbies were invited to the party? Check out this guide to Anthropic, the makers of the Claude 2 LLM, which explains why its founders quit OpenAI and how they plan to build models that are “10 times more capable than current AI systems”.

🏗️ Build your own GenAI applications with Gradio and HuggingFace - Andrew Ng is one of the so-called “godfathers” of AI, and this week his company DeepLearning.AI announced a new (free) course which teaches you how to build your own image generation models and create shareable demos using Gradio and HuggingFace. I’m a big fan of Andrew’s courses, and I particularly like that this one is only 1 hour long - a perfect short project for a rainy Monday afternoon.
🐼 An introduction to Polars for Pandas users - A few months ago, we were introduced to Polars: a new “blazing fast” DataFrame library for interacting with tabular data in Python. Thanks to features like lazy evaluation, it can be up to 10x faster than Pandas for many data-intensive tasks, but if you’ve not had the chance to look into it yet, you’ll enjoy this practical guide from David Hundley which is specifically geared towards existing Pandas users.

🎓 Transferrable skills from academia to commercial AI - Kirill Lepchenkov spent 5 years working as an academic researcher in laser physics before transitioning into a commercial AI role. In this excellent article, they reflect on the transferrable skills they did (and didn’t) learn and provide good advice for those thinking about swapping the lecture hall for the boardroom.

Have a great week 😊

Matt

Know someone who’d find this post interesting? All of this is public, so feel free to share it!