👩🏻✈️The GitHub CoPilot Milestone
Using AI to automate programming has been one of the aspirational goals of machine learning since its early days. There is something incredibly seductive about the idea of using machine intelligence to generate code, but the specifics of the problem have made it really impractical for machine learning techniques. From the lack of high-quality code datasets to train supervised learning models, the diversity of different programming languages, to the contextual nuances of each software application, there are numerous challenges that make the intelligent code generation problem more difficult than it seems at first glance. Recent breakthroughs in NLP, with architectures such as transformers, seem to finally bring us closer to a potential solution to this difficult problem.
GitHub CoPilot is a new ML system created as part of a collaboration between Microsoft and OpenAI to streamline the generation of code snippets across different programming languages. You can think of CoPilot as AI-powered pair programming which, if nothing else, sounds kind of cool 😊. The new technology is powered by OpenAI Codex, which is a variation of GPT-3 but specialized in code generation. Just like GPT-3 is able to understand English, Spanish or French, Codex can analyze contextual information in programming languages like Python, TypeScript, Java, and many others. Codex was trained in large amounts of source code repositories including those in GitHub. With Microsoft and GitHub developer reach, CoPilot has the opportunity to achieve mainstream adoption by the developer community and, for the first time, it feels that we might be getting closer to a real solution to the intelligent code generation problem in machine learning.
🔺🔻TheSequence Scope – our Sunday edition with the industry’s development overview – is free. To receive high-quality content about the most relevant developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻
🗓 Next week in TheSequence Edge:
Edge#103: Recap of Reinforcement Learning series
Edge#104 is about AllenNLP which makes cutting-edge NLP models look easy
Now, let’s review the most important developments in the AI industry this week
🔎 ML Research
New NLP Research and Datasets
Facebook AI Research (FAIR) published new research and datasets that can help advance conversational systems on a large scale ->read more on FAIR blog
Continuous ML at Uber
The Uber engineering team published an insightful blog post about the best practices to enable continuous integration in its ML infrastructure ->read more on Uber blog
Microsoft Research published a paper detailing CausalCity, an open-source, high fidelity environment to improve causal reasoning in ML systems ->read more on Microsoft Research blog
Google Research published a detailed blog post unveiling a new ML system that can improve game testing at scale->read more on Google Research blog
🤖 Cool AI Tech Releases
Microsoft and OpenAI collaborated in the release of CoPilot, an AI agent that can automate code recommendations ->read more on GitHub blog
Facebook AI Research (FAIR) open-sourced Habitat 2.0, a new version of its environment to advance embodied AI research ->read more on FAIR blog
📌 Join us July 14th at MLCon
We are happy to partner with cnvrg.io on MLCon! It is a virtual ML community conference meant to break down silos, share lessons learned, pro tips, and proven strategies for building real-world AI applications.
💬 Useful tweet
We find the best courses, articles and make our own treads to make it easier for you to learn and navigate the ML world
💸 Money in AI
What do you think about this new format of Money in AI? Is it more useful for you this way, or is it better with more details about the companies? Let us know by replying to this email.