Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
OpenAI acquires Global Illumination (openai.com)
120 points by davidbarker on Aug 16, 2023 | hide | past | favorite | 55 comments


Global Illumination appears to have been working on a browser-based multiplayer Minecraft clone (https://www.biomes.gg/), not anything explicitly related to AI content generation.

Odd acquisition choice unless OpenAI is going to do an unexpected pivot.


Their statement is pretty clear about what they see as the value add:

> OpenAI has acquired the team at Global Illumination, a company founded by Thomas Dimson, Taylor Gordon, and Joey Flynn. The entire team has joined OpenAI to work on our core products including ChatGPT. Global Illumination is a company that has been leveraging AI to build creative tools, infrastructure, and digital experiences. The team previously designed and built products early on at Instagram and Facebook and have also made significant contributions at YouTube, Google, Pixar, Riot Games, and other notable companies. We’re very excited for the impact they’ll have here at OpenAI.

Seems like a very standard acquihire. There's no indication that any of Global Illumination's projects will continue, or that OpenAI will get into the gaming business.


> Seems like a very standard acquihire. There's no indication that any of Global Illumination's projects will continue, or that OpenAI will get into the gaming business.

Deploying AI agents to 3D worlds seems like a huge potential market. Not just for gaming, but as a test bed for real world 3D SLAM.

Rendering scenes in 3D with AI seems like another huge potential market, at the intersection of movies and gaming.

This is both at once. I think it'd be a mistake to put them to work on ChatGPT.


I saw a demo of a Skyrim mod on Youtube [1], where they wired the dialog screen with NPCs in the town of Windhelm up with the OpenAI API.

Each character had all their Bethesda-written dialog fed to the LLM as a starting prompt, so each one communicated with the player within the bounds of their authored identity, and knew what that character was written to know.

The human player used freeform in-game text-to-LLM as input to talk to the people in the town square. The latency was large, but not more than a few seconds.

He was able to walk up to someone and say things like "Where do you live? Who is your father?" .... latency pause ... "Oh me? I live in the Cloud District. It's where all of us Battleborn clan live, of course!"

The real kicker of this demo was that they passed the LLM-generated response dialog in to a deep-fake model trained on each character's original voice actors (since the corpus in Skyrim is vast, this ended up quite accurate).

All of this was a major generational shift for gaming as a whole. I had just witnessed the first time a video game player could actually TALK to an NPC, in a fully unstructured and unscripted fashion, while each character remained as in-character as the original game's authors intended.

So maybe this purchase by OpenAI is about something like that. Building NPCs for Biomes.gg that you can actually talk to.

[1] https://www.youtube.com/watch?v=d6sVWEu9HWU


>The real kicker of this demo was that they passed the LLM-generated response dialog in to a deep-fake model trained on each character's original voice actors (since the corpus in Skyrim is vast, this ended up quite accurate).

The voice is added in post through editing. The voice generation wasn't performed in real time. Still, a very interesting project.


Well.. but it's relatively easy to do that. It's combining the OpenAI API, Whisper, and Eleven Labs (or similar). Not trivial, but you don't need to hire a whole team.


Maybe not, but frequently hiring a team that has accomplished a significant percentage of a goal is a better choice than trying to retask existing folks, or to grow a new product set outside of core competency.


yEAH THIS IS A 5 YEAR PLAY.

They are going to put AI to voxel 3d spatial dev... "EnvironGPT" '3DGPT'


???

Sorry, I don't see any of that written there, not "pretty clear" to me.


Scroll down.


Scroll down to where? It's a 4 line comment.


Scroll down on the post, not the comments, and you'll see where the parent pulled their quote from, which you said you "don't see any of that written there," well, it's written there.


Nonsense.

What's the point of the quote, then?


I'm honestly at a loss for what you want from this conversation. What "nonsense" are you even talking about? You said you don't see the quote written there (I presume on the website) so I told you to scroll down and you'll see the quote. That's all I was saying, but now you're asking what the point of the quote was, like, how should I know? Ask the parent.


You're hilariously confused.

Someone posts this:

>Their statement is pretty clear about what they see as the value add

And add a quote to support that.

I argue that I do not see what the value added is from that quote, (and I still don't, btw).

You tell me to scroll down ... ??? What? I don't need to scroll down to read a 4-5 sentence quote.

You then start talking about the article, which is not what I'm not referring to, I'm talking about the quote.

I ask "what's the point of the quote then?", as in, if GP wanted to point its argument at something else in TFA why would it choose to quote this instead? And you also didn't grasp that.

Anyway, hope it's all clear now.


I worked with these guys a long time ago at FB.

Their founder and CEO created The Algorithm at Instagram. Rest of the team is super strong too. Makes a lot of sense for OpenAI to acquire a team at the intersection of AI and consumer products.


I think they have to decide if they want to be an API provider or a product-maker.

Businesses don't tend to do well when they directly compete with their clients.

At the very least spin one off into a subsidiary whose affiliation isn't widely known.


Seems pretty obvious they decided to be a product maker. This is them doubling down.

They started out obviously doing API stuff. chatGPT was some sort of proof-of-concept or whatever but once it went viral, the obvious pivot is to be a product maker. The margins on ChatGPT+ is way better than an API, as every ChatGPT clone will tell you. A viral product is really hard to make, never mind make and throw away to focus on lower margin corporate customers.

They seem to have a great vision now - make a good product (WIP, but rapidly iterating), and sell the underlying models to Microsoft to offer as an API for the residual value they can’t capture directly. This should make it clear that if you’re selling a Chatbot off of their APIs, they’re planning to compete with you.


100% Yep. I like OpenAI but I think their peril will be due to lack of vision more than anything else. Other players are catching up. IMHO they should stick to providing AI infrastructure.

That being said, considering Microsoft's investment, the most likely outcome is that OpenAI will move into the consumer business while being the research arm for Azure.


>Businesses don't tend to do well when they directly compete with their clients.

True in general, but then we have Amazon.


People like Walmart still refuse to let any customer data near AWS.


you wrote people. did you mean corporations?

don’t let those darn things trick you into thinking they are people. legally it gets murky, but they aren’t people. it is all just a ploy for emotional connection for the sake of taking advantage of you.

like cats.


> I think they have to decide if they want to be an API provider or a product-maker. > Businesses don't tend to do well when they directly compete with their clients.

Isn't this what MSFT and GOOG do?


IG's algorithm is one of the worst on the planet. Though not as bad as Spotify's. Not really a selling point.


They are devoting enormous resources to studying LLM safety - my assumption is they’re trying to push the limits of what’s possible with autonomous agents before someone else does.

If you can build a bot to beat an open world game as complex as Minecraft, you’re pretty close to automating an enormous amount of jobs and potentially doing some destructive stuff. This seems like an aquihire to kickstart that initiative.

Full disclosure this is a complete assumption, could just as well be on the whims of Sama. I’m skeptical of the other answers here - OpenAI is perfectly capable of hiring some of the best PM’s in the world without having to buy a company.


Part of Minecraft is gathering resources, building weapons, and out-manoeuvering and killing enemies. Forget jobs, those implications keep me up.


There was that Sims post the other day from a16z as well. Seems like a plausible next area for them to take on.

Was it them or someone else who trained their bots on Dota a few years back?


It was them. https://openai.com/research/openai-five

They've since abandoned all their reinforcement learning projects. (aside from RLHF for training ChatGPT)


Thanks for the link!

That's a shame, I was really impressed with their research in that space.


Did they make a public announcement to that effect? Or do you have insider knowledge?


A co-founder announced they disbanded their robotics team a couple years ago: https://venturebeat.com/business/openai-disbands-its-robotic...

That was the same time they depreciated OpenAI Gym: https://github.com/openai/gym


If it is the people but not the company that were interesting, OpenAI could make an offer to each of the persons working there.

So obviously yes it's an odd decision to buy an open-source MMORPG, but there are tons of possibilities.

Plenty of times, you can see large companies buying non-sense, or companies at outrageous price (we just don't know the price in the case here).

Sometimes even to fire the people behind, and throw away the product.

It can be for many reasons:

- Either they really have a solid business use which justifies the acquisition

or

- Perhaps they have a common investor/VC, and that investor is influent enough in OpenAI to "convince" the board to buy the company.

or

- Some investors give a kickback on the acquisition (I don't think so)

or

- Microsoft wanted to early kill a competitor to Minecraft.

etc.

Probably it'll be difficult to know.


People have been training ML to play Minecraft via text commands for a while. Having access to the guts of a clone makes perfect sense given their past and present projects.


Just speculating here but... More than just _playing_ a game, I could see OpenAI wanting to go after the 3D assets market by letting players generate levels, items, and perhaps even NPCs, simply by describing them. It might be easier to generate voxel-based assets first though, which would make Minecraft-esque games a logical first target (perhaps just behind 2D games, but I suspect the market isn't as big there).


I agree, let’s presume that we will expect some more functionality like Text to 3D generated content. CAD is big market


I'd be really curious how this training works. Is it done on a real-time minecraft application, or did they build a custom game engine that mimics minecraft but without any movement delay and without any graphics so the speed of training could be thousands of times faster?


Maybe...if they didn't already have direct access to the original via Microsoft.


That's a good point. Looks like there's more to the story in the announcement.


"Conspiracy" theory:

Couldn't it be the opposite ?

Let's imagine, that Microsoft, one of the main investor of OpenAI, wants to get rid of a solid open-source Minecraft clone before it gets popular ?

https://www.youtube.com/watch?v=vPHEtewFm3M

Not saying it's true, but that's totally reasonable to spend xxM USD if you have to defend your multi-billion game license.


I would totally buy this. This way Microsoft doesn’t look bad in an era where big tech is coming under scrutiny for acquisitions.


The market is littered with clones of minecraft and has been since before it was even done being made. None of them have even been remotely a threat to Minecraft. Turns out you need way more than just a bunch of blocks.


Read somewhere recently about OpenAI bleeding senior talent. Could be in response to that.


Where did you read this, and what was the reasoning?


Open AI voxel-based AI dynamic navigable 3d environs.


seems like they gunning for deepmind rl based ideas.


Roblox for AI


I guess no one else is allowed to have indirect lighting anymore (/s)

As someone not well versed in how startups work, why do they need to acquire the entire company to have the team work for them, as opposed to negotiating with the employees individually? Is it essentially a lump sum payment to stop what they were currently working on.


a quick overview of the startup economy

1) have an idea, negotiate seed funding. Someone will give you a bunch of money to make an MVP. the "angel" or "seeder" will exchange money for a percentage of the company. They will have a say in how its run

2) grow the company according to the metrics that make you look valuable (market share is normally the common one)

3) growing costs money, so raise a "series a" same deal as before, either seeder/angels are bought out, or their share of the company is diluted in exchange for more monoey from the funders, typically a VC.

4) repeat until either bankrupt, IPO, Profitable, or boughtout.

Now. As an early employee, you are given shares in the company. Typically <1%. the more rounds you have, the more diluted that percentage gets.

when the company is bought out, your shares are exchanged for a price. however as you didn't put the money in, you are a way down the pecking order for getting the cash. Not only this, but a "buyout" price is rarely the headline price. For example CTRL-Labs[1] was "bought out" for $500 mill. However thats not the price they paid for the company. thats the price paid for the company and the "golden handcuffs" to keep employees.

of that 500mill some of is say <75m will go to physically buy the company, the rest will go towards golden handcuffs to keep the employees working. After all, if you give most of the company >$2m upfront, they'll say "work for facebook? fuck that, I'm retiring"

So to combat that, you'll get your 0.0001% of $75m, then offered a contract with the new company that will give you shares that will be released to you gradually after a certain cool off period, say one year (Fb and google give you 6.25% every 3 months.) That means that to get your $2m you need to work for the new company for four years.

[1]https://www.businessinsider.com/facebook-acquired-brain-comp...


> As someone not well versed in how startups work, why do they need to acquire the entire company to have the team work for them, as opposed to negotiating with the employees individually?

Because otherwise, the team would have the option of continuing their current employment.

Take away the alternative they are demonstrably already choosing over you, and you have a better chance of getting and keeping them.


It also usually triggers a liquidity event, so the team gets paid (or their equity gets transferred to OpenAI) in a way that provides a bit of a bonus and incentive to stick around.


Just dropping this here since it looks like a lot of people haven't heard of it.

Voyager: An Open-Ended Embodied Agent with Large Language Models

https://arxiv.org/abs/2305.16291


The fact that they built a Minecraft clone for the web is probably not the main point. It's just one of many systems they have built.

(There have actually been multiple Minecraft-like web codebases around for years. Nothing as complete as Biomes that I know of though. I built a basic one 11 years ago. https://vimeo.com/50111926/comments)

They obviously already have lots of machine learning expertise and wanted top-level application engineering. This is a way to acquire a whole integrated team.


They started with games with RL libraries like gym, so probably they may try to optimize their llms with simulated agents. Or they are planning for a great marketing strategy with improved algorithms


Reinforcement learning with better graphics?


Well a metaverse or character.ai like product is an obvious application.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: