{"id":84196,"date":"2022-11-29T12:11:20","date_gmt":"2022-11-29T17:11:20","guid":{"rendered":"http:\/\/danielschristian.com\/learning-ecosystems\/?p=84196"},"modified":"2022-11-29T13:01:28","modified_gmt":"2022-11-29T18:01:28","slug":"a-bot-that-watched-70000-hours-of-minecraft-could-unlock-ais-next-big-thing-douglas-heaven-plus-several-other-items-re-ai","status":"publish","type":"post","link":"https:\/\/danielschristian.com\/learning-ecosystems\/2022\/11\/29\/a-bot-that-watched-70000-hours-of-minecraft-could-unlock-ais-next-big-thing-douglas-heaven-plus-several-other-items-re-ai\/","title":{"rendered":"“A bot that watched 70,000 hours of Minecraft could unlock AI\u2019s next big thing” [Douglas Heaven] plus several other items re: AI"},"content":{"rendered":"

A bot that watched 70,000 hours of Minecraft could unlock AI\u2019s next big thing<\/strong><\/a> — from technologyreview.com by Will Douglas Heaven
\nOnline videos are a vast and untapped source of training data\u2014and OpenAI says it has a new way to use it.<\/em><\/p>\n

Excerpt:<\/em><\/p>\n

OpenAI has built the best Minecraft-playing bot yet by making it watch 70,000 hours of video of people playing the popular computer game. It showcases a powerful new technique that could be used to train machines to carry out a wide range of tasks by binging on sites like YouTube, a vast and untapped source of training data.<\/p>\n

The Minecraft AI learned to perform complicated sequences of keyboard and mouse clicks to complete tasks in the game, such as chopping down trees and crafting tools. It\u2019s the first bot that can craft so-called diamond tools, a task that typically takes good human players 20 minutes of high-speed clicking\u2014or around 24,000 actions.<\/p>\n

The result is a breakthrough for a technique known as imitation learning<\/em>, in which neural networks are trained to perform tasks by watching humans do them.
\n…
\nThe team\u2019s approach, called Video Pre-Training (VPT), gets around the bottleneck in imitation learning by training another neural network to label videos automatically.
\n<\/strong><\/span><\/p><\/blockquote>\n

Speak lands investment from OpenAI to expand its language learning platform<\/strong><\/a> — from techcrunch.com by Kyle Wiggers<\/p>\n

Excerpts:<\/em><\/p>\n

\u201cMost language learning software can help with the beginning part of learning basic vocabulary and grammar, but gaining any degree of fluency requires speaking out loud in an interactive environment,\u201d Zwick told TechCrunch in an email interview. \u201cTo date, the only way people can get that sort of practice is through human tutors, which can also be expensive, difficult and intimidating.\u201d<\/p>\n

Speak\u2019s solution is a collection of interactive speaking experiences that allow learners to practice conversing in English. Through the platform, users can hold open-ended conversations with an \u201cAI tutor\u201d on a range of topics while receiving feedback on their pronunciation, grammar and vocabulary.<\/p>\n

It\u2019s one of the top education apps in Korea on the iOS App Store, with over 15 million lessons started annually, 100,000 active subscribers and \u201cdouble-digit million\u201d annual recurring revenue.<\/strong><\/span><\/p><\/blockquote>\n

 <\/p>\n

\n

If you last checked in on AI image makers a month ago & thought \u201cthat is a fun toy, but is far from useful\u2026\u201d Well, in just the last week or so two of the major AI systems updated.<\/p>\n

You can now generate a solid image in one try. For example, \u201cotter on a plane using wifi\u201d 1st try: pic.twitter.com\/DhiYeVMEEV<\/a><\/p>\n

\u2014 Ethan Mollick (@emollick) November 26, 2022<\/a><\/p><\/blockquote>\n