Jul 11, 2023
ChatGPT, until the release of Meta’s Twitter competitor Threads, the fastest-growing consumer app ever (100m users in just two months), saw its first dip in traffic last month. Looking beyond the headlines, one realizes that AI usage overall keeps rising (which shouldn’t surprise anyone) – it’s just that specific usage for AI-posterchild ChatGPT has dropped as its novelty has (somewhat) worn off.
Meanwhile, the team behind ChatGPT released a new module for its AI assistant – ChatGPT Code Interpreter (available only to paying ChatGPT Plus subscribers). Upload pretty much any form of data (e.g., a CSV file) up to 100MB, and you can start asking ChatGPT to create visualizations (including animated GIFs) and analysis. It’s mind-blowingly good – visualizations that took you hours to create and fine-tune are done in minutes – and even more impressively – having ChatGPTs text interface able to interact with your data allows you to ask complex questions in natural language. Your $20/month ChatGPT Plus subscription just got you a personal data analyst.
Looking beyond the obvious – from the pressure this will put on specialized data analytics tools to entry-level data analyst jobs – it presents a clearer picture of our AI-enabled future. The interesting feature enabling the powerful features of the ChatGPT Code Interpreter is ChatGPT’s use of its language model to make sense of the data (allowing it to work with rather unstructured data inputs) which are then piped into a Python interpreter for which ChatGPT develops the code to run the analysis. This solves a good chunk of the “hallucination” problem plaguing LLMs and provides clear(er) error messages when something doesn’t work as intended.
All of this brings down the time it takes to “try something out” rapidly – turning work which took weeks to complete into queries done in a few minutes. I believe this will result in massively changing workflows inside corporations where we move into a mode of hyper-fast iteration, a massively increased volume of ideas being tried out, and – ultimately/hopefully – better results for the enterprise.
My friends at the global consulting firm EY developed a concept around the “Superfluid Enterprise” – mostly focussed on an enterprise’s ability to become super-modular (something we talk about in our work on Hourglass Economics). ChatGPT’s Code Interpreter might be another glimpse into the future – a world in which information turned superfluid. (via Pascal)
