GitHub Copilot: First Impressions
tl;dr: The good - you don't have to write boilerplate code. The bad - it's inconsistent, so you need to be aware of it at all times. Vlad believes there is a big opportunity for non-programmers and will impact how we perform code reviews.
featured in #240
Risk Assessment Of GitHub Copilot
tl;dr: The author is given access to the test phase of Copilot and asks the question - "how risky is it to allow an AI to write some or all of your code?"
featured in #236
Language Models Like GPT-3 Could Herald A New Type Of Search Engine
Will Douglas Heaven
tl;dr: Google researchers have published a proposal that could change search. Instead of leveraging Pagerank - Google's search algorithm - the new interface would allow users to ask questions and have a language model trained on all pages and answer questions directly.
featured in #230
Measuring Trends in Artificial Intelligence
tl;dr: AI investment in drug design and discovery increased significantly, more people are working in industrial roles, systems can compose text, audio, and images to a high standard that humans have a hard time telling the difference between "synthetic and non-synthetic outputs," and more.
featured in #228
Timnit Gebru’s Exit From Google Exposes a Crisis in AI
tl;dr: Google's firing of Timnit Gebru shows that "corporate-funded research can never be divorced from the realities of power." AI will reinforce discrimination unless action is taken: (1) Tech workers need to unionize as a "key lever for change." (2) We need protection and funding for research. (3) Regulation.
featured in #220
DALL·E: Creating Images from Text
tl;dr: "We’ve trained a neural network called DALL·E that creates images from text captions for a wide range of concepts expressible in natural language."
featured in #220
Experimenting With Automatic Video Creation From A Web Page
tl;dr: "we envision a future where creators focus on making high-level decisions and an ML model interactively suggests detailed temporal and graphical edits for a final video creation on multiple platforms."
featured in #215
OpenAI Presents GPT-3, A 175 Billion Parameters Language Model
tl;dr: “We find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans.”
featured in #194
Giving GPT-3 a Turing Test
tl;dr: "GPT-3 is quite impressive in some areas, and still clearly subhuman in others." Kevin shows the questions he asks OpenAI’s new GPT-3 language model, along with its answers.
featured in #193