💻 OpenAI Codex, a Programming Challenge and one of the Most Impressive AI Demos Ever Created
The Scope covers the most relevant ML papers, real-world ML use cases, cool tech releases, and $$$ in AI. Weekly.
📝 Editorial
Using artificial intelligence (AI) to write programming code automatically has always been considered one of the ultimate goals of the space. Now, for the first time in history, it seems that we might be getting closer to achieve that goal. A few weeks ago, the GitHub team announced the availability of CoPilot, a digital assistant optimized for pair programming. The engine powering GitHub Copilot is Codex, a model designed to interpret commands in natural language and output code in over a dozen programming languages such as JavaScript, Go, Perl, PHP, Ruby, Swift, TypeScript, and several others. Codex is based on the famous GPT-3 but trained not only in natural language but on billions of source code files.
Earlier this week, OpenAI announced the availability of Codex through their API platform. The OpenAI team also unveiled the Codex Challenge, a competition to solve Python programming puzzles assisted by the Codex model. The contest includes five programming exercises that you can solve by yourself or with the assistance of Codex. You can see the results live on the Codex Challenge leaderboard. The OpenAI team announce Codex and the programming challenge during a session in which they built several live applications using Codex. This has to be one of the most impressive AI demos I’ve ever seen. The richness of the language interactions and the quality of the code produce were beyond impressive. Watch the demo and judge for yourself 😊
🔺🔻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#115: the concept of OpenAI GPT-3 (🔥); two mechanisms for improving the current generation of transformer models introduced by Facebook AI Research; OpenAI API that brings GPT-3 to developers.
Edge#116: AI2-Thor – an open-source framework for embodied AI research.
Now, let’s review the most important developments in the AI industry this week
🔎 ML Research
Know Your Data
Google Research published a blog post detailing usage for the newly released dataset exploration tool Know Your Data ->read more on Google Research blog
Evaluating Text Summarization
Amazon Research published a paper describing a new metric to evaluate the performance of abstractive text summarization models ->read more on Amazon Research blog
AI for Policy Making
Salesforce Research published a super interesting paper showing how real-world policy design can use machine learning ->read more on Salesforce blog
🛠 Real World ML
Reducing Big Data Infrastructure Costs at Uber
The Uber engineering team details some of the architecture principles used to reduce the operational cost of their massive big data infrastructure ->read more on Uber blog
LinkedIn Lambda Learner Architecture
The LinkedIn engineering team published a blog post about Lambda Learner, an architecture for adapting machine learning models to real-time changes ->read more on LinkedIn blog
Task-Oriented Conversational AI at Airbnb
The Airbnb engineering team discusses the architecture to power task-oriented chatbots for customer support ->read more on Airbnb blog
Personalized Newsletter Delivery Using TFX
The Swiss online retailer Digitec Galaxus AG team details how they use TensorFlow Extended to deliver millions of personalized newsletters every week ->read more on TensorFlow blog
🤖 Cool AI Tech Releases
OpenAI Codex
The OpenAI team announced the availability of their Codex model for code generation through the OpenAI API ->read more on OpenAI blog
A massive GPT-3 Rival
The Israeli AI startup AI21 released the new version of their AI21 Studio developer platform accompanied by Jurassic-1, a 178B parameter model that is very similar to GPT-3 ->read more on AI21 blog
💬 Useful Tweet
That’s incredible
💸 Money in AI
Data-centric AI development platform Snorkel AI raised $85 million in Series C round at a $1 billion valuation, co-led by Addition and BlackRock. Hiring across all teams.
*Here you can read our deep dive about Snorkel (no subscription is needed).
Conversational AI startup Deepbrain AI (formerly Moneybrain) raised $44 million in a Series B round led by Korea Development Bank. Hiring in S. Korea.
Adaptive AI Company Latent AI raised $19 million in Series A round co-led by Future Ventures and Blackhorn Ventures. Hiring in Princeton, New Jersey.
AI-enabled chip design startup Motivo raised a $12 million Series A financing round led by Intel® Capital. Hiring.
Scalable and long-term AI solutions provider ThirdAI raised $6 million in seed funding, co-led by Neotribe Ventures, Cervin Ventures, and Firebolt Ventures.