Daily Archives: June 26, 2022

In 2020, OpenAIs machine learning algorithm GPT-3 blew people away when, after ingesting billions of words scraped from the internet, it began spitting out well-crafted sentences. This year, DALL-E 2, a cousin of GPT-3 trained on text and images, caused a similar stir online when it began whipping up surreal images of astronauts riding horses and, more recently, crafting weird, photorealistic faces of people that dont exist.

Now, the company says its latest AI has learned to play Minecraft after watching some 70,000 hours of video showing people playing the game on YouTube.

Compared to numerous prior Minecraft algorithms which operate in much simpler sandbox versions of the game, the new AI plays in the same environment as humans, using standard keyboard-and-mouse commands.

In a blog post and preprint detailing the work, the OpenAI team say that, out of the box, the algorithm learned basic skills, like chopping down trees, making planks, and building crafting tables. They also observed it swimming, hunting, cooking, and pillar jumping.

To the best of our knowledge, there is no published work that operates in the full, unmodified human action space, which includes drag-and-drop inventory management and item crafting, the authors wrote in their paper.

With fine-tuningthat is, training the model on a more focused data setthey found the algorithm more reliably performed all of these tasks, but also began to advance its technological prowess by fabricating wooden and stone tools and building basic shelters, exploring villages, and raiding chests.

After further fine-tuning with reinforcement learning, it learned to build a diamond pickaxea skill that takes human players some 20 minutes and 24,000 actions to accomplish.

This is a notable result. AI has long struggled with Minecrafts wide-open gameplay. Games like chess and Go, which AIs already mastered, have clear objectives, and progress toward those objectives can be measured. To conquer Go, researchers used reinforcement learning, where an algorithm is given a goal and rewarded for progress toward that goal. Minecraft, on the other hand, has any number of possible objectives, progress is less linear, and deep reinforcement learning algorithms are usually left spinning their wheels.

In the 2019 MineRL Minecraft competition for AI developers, for example, none of the 660 submissions achieved the competitions relatively simple goal of mining diamonds.

Its worth noting that to reward creativity and show that throwing computing power at a problem isnt always the answer, the MineRL organizers placed strict limits on participants: they were allowed one NVIDIA GPU and 1,000 hours of recorded gameplay. Though the contestants performed admirably, the OpenAI result, achieved with more data and 720 NVIDIA GPUs, seems to show computing power still has its benefits.

With its video pre-training (VPT) algorithm for Minecraft, OpenAI returned to the approach its used with GPT-3 and DALL-E: pre-training an algorithm on a towering data set of human-created content. But the algorithms success wasnt enabled by computing power or data alone. Training a Minecraft AI on that much video wasnt practical before.

Raw video footage isnt as useful for behavioral AIs as it is for content generators like GPT-3 and DALL-E. It shows what people are doing, but it doesnt explain how theyre doing it. For the algorithm to link video to actions, it needs labels. A video frame showing a players collection of objects, for example, would need to be labeled inventory alongside the command key E which is used to open the inventory.

Labeling every frame in 70,000 hours of video would beinsane. So, the team paid Upwork contractors to record and label basic Minecraft skills. They used 2,000 hours of this video to teach a second algorithm how to label Minecraft videos, and that algorithm, IDM, annotated all 70,000 hours of YouTube footage. (The team says IDM was over 90 percent accurate when labeling keyboard and mouse commands.)

This approach of humans training a data-labeling algorithm to unlock behavioral data sets online may help AI learn other skills too. VPT paves the path toward allowing agents to learn to act by watching the vast numbers of videos on the internet, the researcher wrote. Beyond Minecraft, OpenAI thinks VPT can bring new real-world applications, like algorithms that operate computers at a prompt (imagine, for instance, asking your laptop to find a document and email it to your boss).

Much to the chagrin of the MineRL competition organizers perhaps, the results do seem to show that computing power and resources still move the needle on the most advanced AI.

Never mind the cost of computing, OpenAI said the Upwork contractors alone cost $160,000. Though to be fair, manually labeling the whole data set wouldve run into the millions and taken considerable time to complete. And while the computing power wasnt negligible, the model was actually quite small. VPTs hundreds of millions of parameters are orders of magnitude less than GPT-3s hundreds of billions.

Still, the drive to find clever new approaches that use less data and computing is valid. A kid can learn Minecraft basics by watching one or two videos. Todays AI requires far more to learn even simple skills. Making AI more efficient is a big, worthy challenge.

In any case, OpenAI is in a sharing mood this time. The researchers say VPT isnt without risktheyve strictly controlled access to algorithms like GPT-3 and DALL-E partly to limit misusebut the risk is minimal for now. Theyve open sourced the data, environment, and algorithm and are partnering with MineRL. This years contestants are free to use, modify, and fine-tune the latest in Minecraft AI.

Chances are good theyll make it well past mining diamonds this time around.

Image Credit: SIMON LEE / Unsplash

Follow this link:

OpenAI's New AI Learned to Play Minecraft by Watching 70000 Hours of YouTube - Singularity Hub

Just under 402 years ago, in August of 1620, the Mayflower set sail from Southampton, England, bound for America. The 100-foot-long-long, triple-masted wooden vessel with canvas sails took more than two months to cross the Atlantic. It carried 102 passengers, had a max speed of three knots an hour (thats about 6 kilometers or 3.7 miles an hour) and required a crew of 30 to operate.

Earlier this month, another Mayflower crossed the Atlantic, but it couldnt have been more different from its namesake in almost every way. The one similarity was that, well, it was also a boat.

The differences? The new Mayflowerlogically dubbed the Mayflower 400is a 50-foot-long trimaran (thats a boat that has one main hull with a smaller hull attached on either side), can go up to 10 knots or 18.5 kilometers an hour, is powered by electric motors that run on solar energy (with diesel as a backup if needed), and required a crew of zero.

Thats because the ship was navigated by an on-board AI. Like a self-driving car, the ship was tricked out with multiple cameras (6 of them) and sensors (45 of them) to feed the AI information about its surroundings and help it make wise navigation decisions, such as re-routing around spots with bad weather. Theres also onboard radar and GPS, as well as altitude and water-depth detectors.

The ship and its voyage were a collaboration between IBM and a marine research non-profit called ProMare. Engineers trained the Mayflower 400s AI Captain on petabytes of data; according to an IBM overview about the ship, its decisions are based on if/then rules and machine learning models for pattern recognition, but also go beyond these standards. The algorithm learns from the outcomes of its decisions, makes predictions about the future, manages risks, and refines its knowledge through experience. Its also able to integrate far more inputs in real time than a human is capable of.

The training included teaching the algorithm to identify objects in its path like cargo ships, fishing vessels, or shipping containers floating in the water.

For all its training and preparation, though, the Mayflower 400 ended up falling a bit short of its goal.

It set sail from Plymouth, England on April 29 and was meant to take three weeks to arrive in Washington, DCbut a mechanical issue ended up derailing it to the Canadian port of Halifax. Details werent specified, but it may have been something similar to what happened during the ships first attempted voyage in 2021, when a metal component on the backup generator fractured, and solar power alone wasnt enough for the ship to complete its journey.

The Mayflower 400s engineers will doubtless press on, though, and are likely already planning another voyage for the high-tech autonomous ship. Despite the snafus, its pretty amazing to contemplate how far technology has come since the original Mayflower crossed the Atlantic. Makes you wonder what a similar voyage will look like 400 years from now; from hydrogen-powered airships to civilian submarines to faster, sleeker AI-powered solar vessels, it seems anything is possible.

Image Credit: Oliver Dickinson for IBM/ProMare

Read the original here:

An Autonomous Ship Used AI to Cross the Atlantic Without a Human Crew - Singularity Hub

Reducing carbon emissions from roads, railways, and shipping requires implementing a range of solutions simultaneously. As far as cars are concerned, cutting the number of journeys altogether (by making it easier for people to walk and cycle and improving public transport), changing the fuel in vehicles, and making the most of those vehicles already on the road must all play a part. None of these solutions are sufficient on their own.

In 2030, the sale of new diesel and petrol passenger cars will be outlawed in the UK. The future of passenger motoring will be electric. But recent problems supplying parts and the high carbon cost of manufacturing electric vehicles could delay the climate benefits of this transition.

To make best use of existing petrol and diesel burning vehiclesand the carbon that was invested in creating themdrivers and manufacturers can reduce the emissions of a family of compounds called nitrogen oxides, which are linked to respiratory diseases, through better treatment of exhaust fumes. This way, the communities most blighted by air pollution can at least be protected before harmful vehicle emissions are finally eradicated.

My research team is developing a new generation of catalytic convertersthe devices fitted to exhaust pipes to reduce the release of toxic gases. Inspired by chemistry observed on the surface of extremely hot planets such as Venus, we have produced a synthetic material that could improve air quality.

The suns light destroys carbon dioxide (CO) in the atmospheres of planets, producing carbon monoxide (CO). Not fast enough to avert climate change, but enough that atmospheres like Venus should contain far more CO than we observe there.

Our group studies the effects of meteoric material (dust arriving from space) in atmospheres. An iron silicate powder we made which replicates this dust can speed up the conversion of CO to CO. This is what the first catalytic converters in cars were designed to do, since CO is a toxic gas.

That got us thinking about whether this material could help with other problems, such as nitrogen oxide pollution, which exceeds legal limits in the air of many UK cities. Poor air quality from vehicle exhausts costs tens of thousands of lives annually.

Weve found that not only can the powder simultaneously clean up CO and nitrogen oxide emissions, but it can convert nitrogen dioxide (NO, a harmful gas which is specifically regulated) to harmless molecular nitrogen (N) and water at room temperature.

Catalysts for processing nitrogen oxide (NOx) emissions installed in modern diesel vehicles only work at exhaust temperatures above 150C. Even if your car uses an additive fluid to reduce nitrogen oxide emissions, its unlikely to work while driving slowly when the exhaust is cooler. This is when vehicles emit the most NOoften in traffic jams where the most polluted air can accumulate.

When the electricity grid is decarbonized and sufficiently robust to charge millions of electric vehicles, catalytic converters capable of removing nitrogen oxides may still be important. For example, the natural gas fuel in industrial furnaces is likely to be replaced with hydrogen.

Unlike buses and cars running on hydrogen, which produce energy via a reaction in a fuel cell, larger applications such as furnaces in steelworks will burn hydrogen fuel directly. This high-temperature combustion will convert molecular nitrogen in the air to nitrogen oxide pollution, which will need to be removed.

Thats why were excited to be developing a prototype emissions converter that can work in most situations, with the potential to radically reduce toxic emissions from combustion engines and other sources in the future.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Image Credit: Hans / 20749 images

See more here:

Cosmic Dust From Venus Is Inspiring New Air Pollution-Busting Technology - Singularity Hub

Hit Reload is a weekly column on everything first-person shooter. This week, Jake unloads his thoughts on the unfortunately named blaster Trepang2, which channels the spirit of Black and F.E.A.R to great effect.

In Trepang2 you play as an amnesiac soldier that has been infused with supernatural abilities. In the Steam Next Fest demo, thats about all youre given before you are turned loose into the world to seek vengeance, teaming up with a mysterious group to mete out punishment.

Is vengeance even possible when you have no memory of your previous life? Trepang2 doesnt care, its too busy handing you an assault rifle to philosophise. Youll likely be too busy too, as smart enemies outnumber and overwhelm you as you try to make do with the gun in your hand and your two abilities: a toggleable slow-motion mode and a cloak that drops you into a blue-hued invisibility, offering just enough time to reposition during a firefight or stalk up to an enemy and get him in a headlock to use as a human shield.

Its heaps of fun, even if the demo only gives you access to a short but sweet prologue filled with corridors and men to wallpaper those corridors with, culminating in a set piece best described as elevators full of men in tactical gear charge at you until everyone dies. Then theres a few arena-based wave defence modes, which dont have the same charm as the prologue but have still kept me amused for a few hours due to the tense nature of the brutal combat.

Trepang2. Credit: Trepang Studios.

To gamers of a certain age, a mysterious commando working with a shadowy group while also having the ability to move in slow motion says one thing: F.E.A.R. In an age of constant remakes, remasters and re-imaginings, Monoliths stellar FPS has never been revisited, but Trepang2 has come along to offer its own interpretation of slow-motion first person blasting.

Theres something about the way Trepang2 has taken the core elements and remixed them the goons attacking you from all sides are in newer tactical gear sure, but this is a pulpy FPS at a time when the genre has largely died out in no small part due to the fact that Activision took Raven Software of Wolfenstein, Singularity and Soldier of Fortune 2 fame and locked them in a room to work on Call of Duty games. This is a game where secret experiments have given you powers and youre more than happy to jab yourself in the arm with a serum that will let you dual wield shotguns. Its B-movie jank, but whos complaining?

F.E.A.R was famous for three things four if you count the terrifying jump scare where monstrous little girl Alma shits you up at the top of a ladder and thats the nailgun, the smart AI and the games slow-motion abilities. All of this took place in the games least memorable feature: dreary light brown corridors just waiting for you to paint the walls with viscera. The sequels did away with the dingy corridors and clean styling and bizarrely lost their identity a little bit. A few hours with Trepang2 and its clear that its nailed the slow motion and the corridors, but theres not enough in the Summer Games Fest demo to see if its ticking all of the boxes just yet.

Trepang2. Credit: Trepang Studios.

I dont think it matters, either way. As a fan of F.E.A.R, obviously I want to see nail guns and proper reverence to a game Ive loved for years. But heres the thing F.E.A.R came out in 2005. Thats coming up on 20 years ago. While in my head I have a list of things I want Trepang2 to adhere to, actually Im in the wrong here its time for a new audience to find things they love. This is the best of both worlds, its clear to see elements of F.E.A.R spattered about the place but this is a game that seems to take its inspiration from a deep love for F.E.A.R. at Trepang Studios.

Now, we get to see Trepang Studios add its own twist. Theres already the ability to mod weapons and even shoot out lights to engage in some light stealth, in addition to grabbing enemies to use as human shields before pulling the pins from their grenades and launching them like an improvised explosive device.

From the demo,Trepang2 appears to be a good game for those looking for that F.E.A.R fix. But its also shaping up to be a solid game on its own merits, and should scratch that itch for a no-nonsense FPS with lashings of gore.

Id still like a nail gun, mind.

You can currently download a free demo for Trepang2 on Steam.

Read more:

'Trepang2' is the gold standard for spiritual successors - NME

A little over a year ago, a Scottish company called Orbital Marine Power was getting a two-megawatt tidal turbine up and running in the North Sea. The system was expected to generate enough electricity to power around 2,000 Scottish homes and offset 2,200 tons of CO2 annually.

Meanwhile, a similar ocean turbine called Kairyu was already in the water off the coast of Japan, undergoing a three-year trial to harness the power of deep ocean currents. The trial was deemed a success, and detailed in a report published by Bloomberg recently.

Ocean turbines are essentially an underwater version of wind turbines, functioning much the same way; the movement of waterwhether from tides or currentsturns blades that are attached to a rotor, turning the rotor and powering a generator.

The biggest tidal power plant in the world is Sihwa Lake station in South Korea, which has a capacity of 254 megawatts (MW), followed by a 240MW station in La Rance, France (this is also the worlds oldest tidal plant, built between 1961-1966).

Both these plants use whats called a barrage, which is a long underwater wall (the barrage at the La Rance plant, for example, is 476 feet long) with gates that open and close with the tides in a way that capitalizes on their energy production potential. This setup works well for generating high quantities of power with multiple turbines; Sihwa Lake has 10 turbines and La Rance has 24.

Kairyu works a little differently. The 330-ton turbine is designed to harvest energy from ocean currents (as opposed to tides). The system is made up of three connected cylindrical pods. The middle pod, which sits several feet higher than those on either side, holds a connector for the power transmitter as well as a device to adjust buoyancythat is, to enable the array to move to deeper or shallower water and thus capitalize on variations in the currents strongest spots.

The pods on either side both have turbine blades on one end that are 36 feet long, and they rotate in opposite directions to stabilize the generators position underwater. They also each contain a generator, a controller, and various measurement instruments.

Kairyu will be anchored to the seafloor, with its pods floating about 160 feet below the waters surface. The system will sit along the Kuroshio current, one of the largest ocean currents in the world. It starts east of the Philippines and flows north-east past Taiwan and Japan. Oceanographer Steven Jayne described it as the strongest current in the Pacific Ocean, and also one of the most intense air-sea heat exchange regions on the globe. It influences climate as far as North America.

The current flows at a rate of 2.5 meters (8.2 feet) per second, and Kairyus trial period showed that the tidal turbine system could steadily generate 100 kilowatts of power (50 per turbine unit) with a rate flow speed of three knots (thats about 3.45 nautical miles per hour). Because ocean currents dont change speed or direction much, systems like Kairyu could generate power some 70 percent of the time versus more intermittent sources like onshore wind (29 percent) and solar (15 percent). Japans New Energy and Industrial Technology Development Organization estimates this current could generate up to 200 gigawatts of reliable energy.

Being an island nation, it seems only logical that Japan should harness ocean currents for energy; according to Ken Takagi, a professor of ocean technology policy at the University of Tokyo Graduate School of Frontier Sciences, theres more potential for ocean current turbine power than offshore wind in the area. Ocean currents have an advantage in terms of their accessibility in Japan, he told Bloomberg. Wind power is more geographically suited to Europe, which is exposed to predominant westerly winds and is located at higher latitudes.

The Japanese are hoping power from the ocean will be a significant contributor to the countrys renewable energy stock and help it meet its net zero carbon emissions target by 2050. Building the submerged transmission lines that will carry the energy Kairyu generates to the grid will be no small project, but the tidal turbine is expected to be up and running in the next decade.

Image Credit: IHI Corporation

Read this article:

Japan's Kairyu Sea Turbine Will Harvest the Never-Ending Power of Deep Ocean Currents - Singularity Hub

Screenshot: YouTube

Welcome friends, to the Steve Carell Vortexthat place where all Steve Carell fandoms collide into one unholy singularity of comedy (with occasional affecting dramatic turns). In this magical wonderland of brand synergy, Thats what she said and Welcome to Marwen can interact in perfect harmony; you can love lamp, 40-year-old virgins, and catching foxes all in one fell swoop. And you can, of course, enjoy a professionally crafted, officially licensed version of The Office opening recreated by the Minions from Minions, because hey, why in the name of Steve couldnt you do that thing?

Minions Opening Credits - The Office US

Hence the above video, posted, not to the depths of Tik Tok or some bizarre fan channel, but to the official The Office YouTube page, in which the various interchangeable yellow monsters you know and fear from Carells Despicable Me franchiseinstead invade Scranton, apparently working in a paper company run by Carells character Gruwhos supposedly set to Rise whenever the second Minions movie ever actually manages to come out

Read more

If nothing else, its a staggering amount of thought and effort to put into such a weird concept; eagle-eyed viewers will see references to a whole host of Officein-jokes, including a Prison Mike callback, that time Dwight wore a terrifying human face mask, and, of course, the famed Kevins famous chili scene. (Is it driving us nuts that the Minion who drops the chili is labeled Otto, instead of Kevin? Trick question, baby: Were already in the Steve Carell Vortex!) You will also get to see a Pam Minion and a Jim Minion suck face for a second, in case that was something you needed to fill out your personal sexual awakening bingo card this afternoon.

The much-delayed Minions: The Rise Of Gru is supposedly due out on July 1, 2022; meanwhile, were holding out hopes to see the Minions address Carells turn as Vices Donald Rumsfeld, next.

Originally posted here:

Brand synergy eats its own tail as The Office recreates its intro with Minions - Yahoo Entertainment