Connect with us
Dota Dota

News

OpenAI reveals self-play information after successful Dota 2 test

Source: OpenAI

Published

on

OpenAI is revealing more information about its bot system after its win streak in the game Dota 2 last Friday, and it’s another step forward in the world of artificial intelligence.

The graph below was released this morning by OpenAI, which depicts the rate at which the bot improved in playing Dota 2.

Dota

(Source: OpenAI)

“Our Dota 2 result shows that self-play can catapult the performance of machine learning systems from far below human level to superhuman, given sufficient compute,” OpenAI wrote in a post. “In the span of a month, our system went from barely matching a high-ranked player to beating the top pros and has continued to improve since then. Supervised deep learning systems can only be as good as their training datasets, but in self-play systems, the available data improves automatically as the agent gets better.”

OpenAI’s Dota 2 project started in March of this year, starting the bot off with simple tasks.

For background, Dota 2 is a free multiplayer battlefield game on Steam, a gaming streaming site. The game prides itself on not imposing limits on its players.

The bot went on an impressive winning streak starting on August 7, beating a notable Dota 2 player named Blitz. That same day, the bot beat two more high-ranking players. The following day, the bot beat Arteezy, another respected player in the game. All four of the players the bot defeated said that fellow player SumaiL could defeat the bot.

Advertisement

SumaiL did not fare as well as his comrades thought he would.

Finally, the bot took on Dota 2’s former world champion, Dendi. The bot beat Dendi 2-0.

As for how the bot learned to play the came, OpenAI had the following explanation, “The bot learned the game from scratch by self-play, and does not use imitation learning or tree search. This is a step towards building AI systems which accomplish well-defined goals in messy, complicated situations involving real humans.”

Some of the skills that the bot picked up were the ability to predict where players will move, to improvise in response to new situations, and how to influence the other players.

Advertisement

In between battles, OpenAI workers combined some “coaching” with self-play, which helped the bot continuously improve.

OpenAI is a non-profit AI research company, founded by Tesla CEO Elon Musk, that researches safe artificial intelligence.

The Musk company has previously trained bots to successfully compete a task after watching it on virtual reality just once, and developed bots that created their own language.

Musk’s involvement in AI should come as no surprise as he has reiterated time and time again how important, and potentially dangerous, AI can be.

Advertisement

“I have exposure to the most cutting edge AI and I think people should be really concerned about it,” Musk has previously said. “I keep sounding the alarm bell, but until people see robots going down the street killing people, they don’t know how to react because it seems so theorial.”

Aside from OpenAI, Musk is also the CEO of Neuralink, a brain-computer interface company.

The developments of the bot in these varied complex scenarios is a fairly large step forward in discovering the power of artificial intelligence. With limited input form the OpenAI engineers, the bot went from a decent player to being essentially unstoppable. This holds a promising future for further AI developments, if they can be contained.

Advertisement
Comments

News

Tesla launches new loaner program that owners will love

Tesla is now giving owners the opportunity to rent a vehicle from them, and it includes a few very attractive features that will have you second-guessing another loaner from insurance.

Published

on

tesla service tech
Credit: Tesla

Tesla has launched a new loaner program that owners will love, as it resolves some concerns over a replacement vehicle while it is being repaired.

Earlier this week, Tesla launched the option to rent a Tesla loaner vehicle for just $45 per day if your vehicle is in Collision Repair. Collision repairs did not formerly warrant the issuance of loaner vehicles, as the insurance provider of the car owner would provide transportation arrangements.

Tesla is now giving owners the opportunity to rent a vehicle from them, and it includes a few very attractive features that will have you second-guessing another loaner from insurance.

The Tesla you rent while your car is in collision repair will come with free Full Self-Driving, free Supercharging, and free toll coverage, no small print included.

All things considered, this is a great deal for those who require a car for transportation while their car is being repaired.

The cost of Supercharging and Full Self-Driving alone would warrant the $45 per day price tag. Add in the tolls for those who commute on turnpikes for work or are planning an extensive trip that would require it, and it truly becomes an even more attractive deal.

Tesla has done a good job at improving its Service division over the past few years, and it truly needed it. In hopes of launching an F1-style service experience, Tesla started doing away with some of its perks, including loaner vehicles for single-day visits and even Uber credits.

Advertisement

Tesla’s ‘F1’ Service strategy eliminates same-day loaner vehicles, Uber credits

However, it has listened to the complaints of its owners and tried to cater an experience that is more advantageous and less of a hassle. It’s already made tremendous steps in the past few years, and this is the icing on the cake.

Continue Reading

Elon Musk

SpaceX Starship Flight 10: What to expect

SpaceX implemented hardware and operational changes aimed at improving Starship’s reliability.

Published

on

Credit: SpaceX

SpaceX is preparing to launch the tenth test flight of its Starship vehicle as early as Sunday, August 24, with the launch window opening at 6:30 p.m. CT. 

The mission follows investigations into anomalies from earlier flights, including the loss of Starship on its ninth test and a Ship 36 static fire issue. SpaceX has since implemented hardware and operational changes aimed at improving Starship’s reliability.

Booster landing burns and flight experiments

The upcoming Starship Flight 10 will expand Super Heavy’s flight envelope with multiple landing burn trials. Following stage separation, the booster will attempt a controlled flip and boostback burn before heading to an offshore splashdown in the Gulf of America. One of the three center engines typically used for landing will be intentionally disabled, allowing engineers to evaluate whether a backup engine can complete the maneuver, according to a post from SpaceX.

The booster will also transition to a two-engine configuration for the final phase, hovering briefly above the water before shutdown and drop. These experiments are designed to simulate off-nominal scenarios and generate real-world data on performance under varying conditions, while maximizing propellant use during ascent to enable heavier payloads.

Starship upper stage reentry tests

The Starship upper stage will attempt multiple in-space objectives, including deployment of eight Starlink simulators and a planned Raptor engine relight. SpaceX will also continue testing reentry systems with several modifications. A section of thermal protection tiles has been removed to expose vulnerable areas, while new metallic tile designs, including one with active cooling, will be trialed.

Advertisement

Catch fittings have been installed to evaluate their thermal and structural performance, and adjustments to the tile line will address hot spots observed on Flight 6. The reentry profile is expected to push the structural limits of Starship’s rear flaps at maximum entry pressure.

SpaceX says lessons from these tests are critical to refining the next-generation Starship and Super Heavy vehicles. With Starfactory production ramping in Texas and new launch infrastructure under development in Florida, the company is pushing to hit its goal of achieving a fully reusable orbital launch system.

Continue Reading

Elon Musk

Elon Musk takes aim at Bill Gates’ Microsoft with new AI venture “Macrohard”

It is quite an appropriate name for a company that’s designed to rival Microsoft.

Published

on

Credit: xAI/X

Elon Musk has set his sights on Microsoft with a new company called “Macrohard,” a software venture tied to his AI startup, xAI. 

Musk described the project as a “purely AI software company” that’s designed to generate hundreds of specialized coding and generative AI agents that could one day simulate products from companies like Microsoft entirely through artificial intelligence.

Macrohard‘s Purpose

Musk announced Macrohard on Friday, though xAI had already registered the trademark with the US Patent Office a few weeks ago, as noted in a PC Mag report. Interestingly enough, this is not the first time that Musk has mentioned such an initiative.

Just last month, he stated that xAI was “creating a multi-agent AI software company, where Grok spawns hundreds of specialized coding and image/video generation/understanding agents all working together and then emulates humans interacting with the software in virtual machines until the result is excellent.”

At the time, Musk stated that “This is a macro challenge and a hard problem with stiff competition,” hinting at the venture’s “Macrohard” moniker. A few years ago, Musk also posted “Macrohard >> Microsoft” on X. 

Advertisement

Powered by xAI and Colossus

Macrohard appears to be closely linked to xAI’s Colossus 2 supercomputer project in Memphis. Musk has confirmed plans to acquire millions of Nvidia GPUs, joining rivals such as OpenAI and Meta in a high-stakes race for AI computing power. Colossus is already one of the most powerful supercomputer clusters in the world, and it is still being expanded.

xAI is only a couple of years old, having been founded in March 2023. During its Engineering Open House event in San Francisco, Elon Musk highlighted that the company’s speed will be its primary competitive edge. “No SR-71 Blackbird was ever shot down and it only had one strategy: to accelerate,” Musk said.

Continue Reading

Trending