OpenAI Has Trained a Neural Network To Competently Play Minecraft

In a blog post today, OpenAI says they’ve “trained a neural network to play Minecraft by Video PreTraining (VPT) on a massive unlabeled video dataset of human Minecraft play, while using only a small amount of labeled contractor data.” The model can reportedly learn to craft diamond tools, “a task that usually takes proficient humans over 20 minutes (24,000 actions),” they note. From the post: In order to utilize the wealth of unlabeled video data available on the internet, we introduce a novel, yet simple, semi-supervised imitation learning method: Video PreTraining (VPT). We start by gathering a small dataset from contractors where we record not only their video, but also the actions they took, which in our case are keypresses and mouse movements. With this data we train an inverse dynamics model (IDM), which predicts the action being taken at each step in the video. Importantly, the IDM can use past and future information to guess the action at each step. This task is much easier and thus requires far less data than the behavioral cloning task of predicting actions given past video frames only, which requires inferring what the person wants to do and how to accomplish it. We can then use the trained IDM to label a much larger dataset of online videos and learn to act via behavioral cloning.

We chose to validate our method in Minecraft because it (1) is one of the most actively played video games in the world and thus has a wealth of freely available video data and (2) is open-ended with a wide variety of things to do, similar to real-world applications such as computer usage. Unlike prior works in Minecraft that use simplified action spaces aimed at easing exploration, our AI uses the much more generally applicable, though also much more difficult, native human interface: 20Hz framerate with the mouse and keyboard.

Trained on 70,000 hours of IDM-labeled online video, our behavioral cloning model (the âoeVPT foundation modelâ) accomplishes tasks in Minecraft that are nearly impossible to achieve with reinforcement learning from scratch. It learns to chop down trees to collect logs, craft those logs into planks, and then craft those planks into a crafting table; this sequence takes a human proficient in Minecraft approximately 50 seconds or 1,000 consecutive game actions. Additionally, the model performs other complex skills humans often do in the game, such as swimming, hunting animals for food, and eating that food. It also learned the skill of “pillar jumping,” a common behavior in Minecraft of elevating yourself by repeatedly jumping and placing a block underneath yourself. For more information, OpenAI has a paper (PDF) about the project.

Read more of this story at Slashdot.

New Linux Foundation Podcast: ‘Untold Stories of Open Source’

The nonprofit Linux Foundation pays Linus Torvalds’ salary and supports many other open source projects. But they also launched a new podcast series this week covering “The Untold Stories of Open Source.”

“Each week we explore the people who are supporting Open Source projects, how they became involved with it, and the problems they faced along the way,” explains the podcast’s GitHub page (where you can put in a pull request to suggest future episodes or track the project’s progress.)

The podcast is available on its official web page, as well as on Spotify, Apple, Google, or “wherever you listen to your podcasts,” according to an announcement from the Linux Foundation. An introductory page says the podcast will be “used to inform the Linux and Open Source communities as to the current state in development of open source initiatives and Linux Foundation Projects. It is vendor neutral, with no interviews of commercial product vendors or sales teams.”

Here’s the first four episodes:

Balancing Priorities at the Cloud Native Computing Foundation, with Priyanka Sharma, general manager
A Life in Open Source, with Brian Behlendorf, general manager at Open Source Security Foundation
A New Model for Technical Training, with Clyde Seepersad, senior vice president of the Linux Foundation’s training/certification project
The Business Side of Open Source, with Patrick Debois, “godfather of DevOps”

Read more of this story at Slashdot.

US Proposes New Rules to Curb ‘Meme Stock’ Rallies

America’s Securities and Exchange Commission “is considering broad changes to curb the frenetic trading of stocks based on social media activity,” reports Reuters:

The proposed overhaul would be the biggest change to Wall Street’s rules since 2005 and would affect nearly every corner of the market, from commission-free brokerages to market makers and exchanges. The U.S. House Committee on Financial Services on Friday called for the SEC, along with other regulators, to do more to protect the markets from similar events….

The U.S. House Financial Services Committee on Friday urged Congress to adopt legislation mandating the SEC study how its rules need to change to address new technological developments, such as digital engagement practices and social media-driven market activity.

Read more of this story at Slashdot.

Meme-Stock Probe Finds Robinhood Woes Were Worse Than It Let On

Bloomberg writes that the makers of the Robinhood app “faced a more dire situation during the height of last year’s meme-stock frenzy than executives at the online brokerage let on publicly, according to a report from top Democrats on a key congressional committee.”

A more-than-yearlong investigation by staff on the House Financial Services Committee concluded Friday that the frenzied trading in GameStop Corp. and AMC Entertainment Holdings Inc. posed a significant threat to the online brokerage. Robinhood avoided defaulting on its regulatory collateral obligations in late January 2021 only because it received a waiver from its clearinghouse, according to the findings… “The company was only saved from defaulting on its daily collateral deposit requirement by a discretionary and unexplained waiver,” according to the report. “Robinhood’s risk-management processes did not work well to predict and avert the risk of default that materialized….”

The 138-page document released on Friday provides the most detailed look yet at how alarmed Robinhood executives grew over the situation in late January 2021. According to the findings, those actions didn’t match the firm’s public assertions.

Read more of this story at Slashdot.

Engineers Demonstrate Quantum Integrated Circuit Made Up of Just a Few Atoms

Engineers in Sydney have demonstrated a quantum integrated circuit made up of just a few atoms. By precisely controlling the quantum states of the atoms, the new processor can simulate the structure and properties of molecules in a way that could unlock new materials and catalysts. New Atlas reports: The new quantum circuit comes from researchers at the University of New South Wales (UNSW) and a start-up company called Silicon Quantum Computing (SQC). It’s essentially made up of 10 carbon-based quantum dots embedded in silicon, with six metallic gates that control the flow of electrons through the circuit. It sounds simple enough, but the key lies in the arrangement of these carbon atoms down to the sub-nanometer scale. Relative to each other, they’re precisely positioned to mimic the atomic structure of a particular molecule, allowing scientists to simulate and study the structure and energy states of that molecule more accurately than ever before.

In this case, they arranged the carbon atoms into the shape of the organic compound polyacetylene, which is made up of a repeating chain of carbon and hydrogen atoms with an alternating pattern of single and double carbon bonds between them. To simulate those bonds, the team placed the carbon atoms at different distances apart. Next, the researchers ran an electrical current through the circuit to check whether it would match the signature of a natural polyacetylene molecule — and sure enough, it did. In other tests, the team created two different versions of the chain by cutting bonds at different places, and the resulting currents matched theoretical predictions perfectly. The significance of this new quantum circuit, the team says, is that it could be used to study more complicated molecules, which could eventually yield new materials, pharmaceuticals, or catalysts. This 10-atom version is right on the limit of what classical computers can simulate, so the team’s plans for a 20-atom quantum circuit would allow for simulation of more complex molecules for the first time. The research has been published in the journal Nature.

Read more of this story at Slashdot.

Goldman Sachs Raising Funds to Buy Celsius Assets

Goldman Sachs is looking to raise $2 billion from investors to buy up distressed assets from troubled crypto lender Celsius, according to two people familiar with the matter. CoinDesk reports: The proposed deal would allow investors to buy up Celsius’ assets at potentially big discounts in the event of a bankruptcy filing, the people said. Goldman Sachs appears to be gauging interest and soliciting commitments from Web3 crypto funds, funds specializing in distressed assets and traditional financial institutions with ample cash on hand, according to a person familiar with the situation. The assets, most likely cryptocurrencies having to be sold on the cheap, would then likely be managed by participants in the fundraising push. Celsius has tapped restructuring advisory firm Alvarez & Marsal, the Wall Street Journal reported Friday afternoon. Earlier this month, Celsius abruptly paused withdrawals, swaps, and transfers between accounts, citing “extreme market conditions.” The disclosure sent bitcoin’s price below $20,000 and prompted the firm’s token to take a 60% tumble.

As of Monday, the company said it’s still working on “stabilizing [their] liquidity and operations.”

Read more of this story at Slashdot.

The Mars Express Spacecraft Is Finally Getting a Windows 98 Upgrade

Engineers at the European Space Agency (ESA) are getting ready for a Windows 98 upgrade on an orbiter circling Mars. The Verge reports: The Mars Express spacecraft has been operating for more than 19 years, and the Mars Advanced Radar for Subsurface and Ionospheric Sounding (MARSIS) instrument onboard has been using software built using Windows 98. Thankfully for humanity and the Red Planet’s sake, the ESA isn’t upgrading its systems to Windows ME. The MARSIS instrument on ESA’s Mars Express was key to the discovery of a huge underground aquifer of liquid water on the Red Planet in 2018. This major new software upgrade “will allow it to see beneath the surfaces of Mars and its moon Phobos in more detail than ever before,” according to the ESA. The agency originally launched the Mars Express into space in 2003 as its first mission to the Red Planet, and it has spent nearly two decades exploring the planet’s surface.

MARSIS uses low-frequency radio waves that bounce off the surface of Mars to search for water and study the Red Planet’s atmosphere. The instrument’s 130-foot antenna is capable of searching around three miles below the surface of Mars, and the software upgrades will enhance the signal reception and onboard data processing to improve the quality of data that’s sent back to Earth. “We faced a number of challenges to improve the performance of MARSIS,” explains Carlo Nenna, a software engineer at Enginium who is helping ESA with the upgrade. “Not least because the MARSIS software was originally designed over 20 years ago, using a development environment based on Microsoft Windows 98!”

Read more of this story at Slashdot.

Government Scientists Discover Biggest Bacteria Ever, Visible To Naked Eye

An anonymous reader quotes a report from Motherboard: Scientists have discovered a bacteria with cells that measure a full centimeter in length, an astonishing size that makes it by far the largest bacterial species ever found and even “challenges our concept of a bacterial cell,” reports a new study. Bacteria are an extraordinarily diverse group of organisms that have inhabited Earth for billions of years and have evolved to occupy a dizzying variety of niches. Still, almost all of these microbes are composed of simple cells that measure about two microns in diameter, which is about 40 times smaller than a strand of human hair.

Thiomargarita magnifica, a bacteria discovered on sunken red mangrove leaves in Guadeloupe, Lesser Antilles, has blown this standard scale out of the water. The species has evolved filamentary cells that are “larger than all other known giant bacteria by ~50-fold,” making them “visible to the naked eye,” according to a study published on Thursday in Science.
Scientists led by Jean-Marie Volland, a marine biologist who holds joint appointments at the Laboratory for Research in Complex Systems and the Joint Genome Institute (JGI), a U.S. Department of Energy office at Lawrence Berkeley National Laboratory, suspect that this record-breaking adaptation is partly due to the astonishing number of duplicated genes wielded by T. magnifica, an ability that is known as polyploidy. […]

The results revealed that these bacteria contain DNA clusters in their cells, which are located in compartments bordered by membranes that the team called “pepins.” These organized pepins provide a stark contrast to the free-floating DNA seen in the cells of most bacteria. In addition, the team’s genetic sequencing revealed that T. magnifica contains hundreds of thousands of genome copies that are dispersed across the cell, adding up to about three times the number of genes in most bacteria, which is an extreme example of polyploidy. “These cellular features likely allow the organism to grow to an unusually large size and circumvent some of the biophysical and bioenergetic limitations on growth,” Volland and his colleagues said.

Read more of this story at Slashdot.

China Bans 31 Live-Streaming Behaviors

Long-time Slashdot reader Mr_Blank shares a report from Gerona: China has enacted new regulation for the live-streaming industry, listing 31 prohibited conducts and raising the bar for influencers to speak out on specific topics, in the government’s latest effort to regulate the booming digital economy. The 18-point guideline, released Wednesday by the National Radio and Television Administration and the Department of Culture and Tourism, requires influencers to have relevant qualifications to cover some subjects, including law, finance, medicine and education discuss, although the authorities have not specified the necessary qualifications.

The 31 prohibited conducts during live-streaming sessions include posting content that weakens or distorts the leadership of the Chinese Communist Party, the socialist system, or the country’s reform and opening-up. Other prohibited behaviors include using deepfake technologies to manipulate the images of party or state leaders and intentionally ‘building up’ sensitive issues and attracting public attention. Live streamers are also prohibited from showing an extravagant lifestyle, such as showing luxury products and cash, the policy said. This article originally appeared in the South China Morning Post.

Read more of this story at Slashdot.