Discover more from Future News & Media
Everything Is Early Access Now
Future News 158: From Conan to AI, open-betas are the meme.
Conan dominated a small corner of the internet around 15 years ago.
The Great Recession was looming, mass lay-offs were on the horizon and economic and social depression hung in the air. But the widespread roll-out of broadband meant more and more people were spending their time in virtual worlds.
They were able to reinvent themselves, sheltering away from increasingly troublesome offline realities. Players could make new friends, embrace a new culture and sport a new look. It was a welcome restart channelled through fantastical avatars.
There was one main title which popularised this new outlook and brought massively multiplayer online role-playing games (MMORPGs) into the mainstream – World of Warcraft (WoW).
By 2008, Blizzard’s break-out PC hit reached 11 million subscribers. That meant WoW’s player-base began to rival the populations of small countries like Greece.
Though WoW’s subscriber count would peak at 12 million in 2010, some hardcore MMORPG players were already bored of the title in 2008 and were seeking something new. Enter The Age of Conan.
Already twice delayed, with some doubts in the gaming press around the project, developers Funcom had to build much-needed hype around their title.
They had already tried the old tricks – releasing concept art and screenshots, whilst providing sporadic updates on the game. But it was a new trick from the savvy Norwegians which would change the world we live in today.
They encouraged prospective players to buy into and build the hype by joining the Age of Conan’s forums. Another, crucial, incentive was provided by Funcom – Community members would also be given early access to a beta version of the game.
The bored and disaffected MMORPG players were keen to play in a new virtual world, so they lapped-up Funcom’s offerings. 500,000 people signed-up to the fan club.
Age of Conan would officially release later that year, quickly amassing more than a million sales, making it the best-selling MMORPG since WoW at the time.
Though Age of Conan is now a forgotten success story, the video games industry would continue to copy its hype-building tactics for more than a decade.
At first, the early access move was popular with independent and MMORPG developers. Minecraft did it and so did AION. But in 2009 one of the first AAA titles, Left 4 Dead 2, also offered a playable behind-the-scenes look into its game.
Early access had gone mainstream thanks to the game’s developer, Valve. By 2013, the hype and FOMO-building mechanism was so popular that Valve, which also owned and operated video game marketplace Steam, fully embraced the concept.
A new early access section was added to the Steam website, with the now popular military simulator Arma 3 amongst the first batch of titles.
The same year another genre of PC games, the multiplayer online battle arena (MOBA), began to flourish. Valve’s Dota 2 provided an early access bundle to players.
Following a two-year open-beta testing period, the title launched in July 2013. It is still one of the world’s most popular games, reaching a peak of almost 1.3 million concurrent players online in 2016. Dota 2 attracted casual players, converts from MMORPGs and a new breed of hardcore MOBA players.
The offline world had become more optimistic and fewer people were interested in virtual worlds. The time had proven right for Dota 2, which attracted casual players, converts from MMORPGs and a new breed of hardcore MOBA players.
The game also attracted another community, a relatively small cadre of scientists who concentrated on a niche strand of machine learning, deep learning. Ilya Sutskever and a team from OpenAI took on professional Dota 2 players.
The five neural networks they deployed were taking part in the latest bout in a man-versus-machine saga which was first popularised when IBM’s Deep Blue eventually beat chess master Garry Kasparov in 1997.
OpenAI spent three years on the project. Like Deep Blue, they initially lost, and they eventually conquered the Dota 2 world champions. The project proved a type of machine learning technique, self-play reinforcement learning, could achieve “superhuman performance” on a difficult task.
Another form of this learning mechanism, based on goal-orientated algorithms which are rewarded or punished as they seek out the correct answer or decision through trial-and-error, is at the heart of fine- tuning ChatGPT. Sutskever calls it ‘Reinforcement Learning from Human Feedback’.
“We can say that in the pre-training process, you want to learn everything about the world. With reinforcement learning from human feedback, we care about the outputs,” he told EyeOnAI in March.
“We say, anytime the output is inappropriate, don't do this again. Every time the output does not make sense, don't do this again.”
The problem is that human feedback is expensive. And what if your product is unfinished, still effectively in beta stage and suffering from “hallucinations”, where it lies or misleads its users? Well, welcome to our new world, ruled over by early access.
“I'm quite hopeful that by simply improving this subsequent reinforcement learning from human feedback step, we can teach it to not hallucinate. Now you could say is it really going to learn? My answer is, let's find out.”
News I Found Interesting 🤔