[MikeB]

https://venturebeat.com/games/inworld-ai-raises-new-round-at-500m-valuation-for-ai-game-characters/

With over $100 million in total funding, Inworld AI is now the best-funded startup at the intersection of AI and gaming, said Kylan Gibbs, cofounder and chief product officer of Inworld AI, in an interview with GamesBeat. That’s a lot of money for a company built around making non-player characters (NPCs). The twist is that these NPCs promise to be as smart as people.

The company will use the funding to accelerate research and development, hire top talent, invest in infrastructure, and launch an open-source version of its character engine.

Costs:

Inworld AI trains its own models and those aren’t cheap. Companies like OpenAI have spent more than $100 million developing the large language models (LLMs) that bring intelligence to machines. A lot of the cost is in data preparation and collection, which means figuring out the kind of knowledge that you’re going to feed into the AI. There are also inferencing tasks, and all of this means spending a lot of money on graphics processing units (GPUs) and dataset preparation.

Gibbs said that his company has the benefit of being able to serve results on a large scale. So, it may turn out to cost something like 0.2 cents per request, which Gibbs said is dramatically cheaper than anything else. That’s also pretty good on costs compared to a year ago. With such costs, it’s easier to figure out how to serve AI answers to millions of players, who might interact 20 minutes a day for a month with characters at the cost of maybe $1.

Investors:

The full slate of investors to date also includes Section 32, Intel Capital, Founders Fund, the Disney Accelerator, Bitkraft Ventures, The Venture Reality Fund, Kleiner Perkins, CRV, Meta, Micron Ventures, NTT Docomo Ventures, and SK Telecom Venture Capital.

I wasn’t sure why they didn’t release the total funding in this round.

“We’re focused on the top level, completely focused on the valuation,” Gibbs said. “Increasingly, it is very clear that game studios are taking on AI as the kind of next frontier. And it is increasingly the case that while we experiment early on, we have an increasing cohort coming to us. And the story we basically have is that this character engine will sit alongside other parts of game development as a key component to the future.”

[MikeB]

I've been watching this space for a while as both my NLP & speech recognition work revolve around this.

However while all of my work is attempting to do the opposite... in Response Speed, Lipsync quality, Character definitions, Character knowledge trust, Power usage, Cost, Data Security....

All of this AI work remains the same. Large power consumption, cost, guesswork characters, guesswork lipsync, 1-3 second response speed, unknown data security.

(Note: Nvidias lipsync in the last few weeks has become quite good: https://youtu.be/DGEIRuXP8hQ?list=PL3jK4xNnlCVfIJU10iQ_dd8J_5T_e0YTJ&t=141)

Today all of this is now considered "borderline acceptable" and declared a $500m asset... but the data is actually the asset, not the novelty & quality of talking NPC's in games/experiences.

Even in 5 years if characters and lipsync are perfect, power consumption is still extreme, and data security is unknown... I don't get it

[infurl]

There are other groups working on increasing the efficiency of these things. Whether or not it will be possible to make the current forms efficient enough in the long run remains to be seen, but we know that the human brain runs on about 25 watts so it must be possible somehow.

[MikeB]

The underlying mechanism is statisical analysis (guessing)... the brain doesn't search every interaction it's ever had to deal with everything.

The story of speech recognition is engineers who gave up and started using prediction models

Now data funds more prediction instead of solving the underlying engineering problems...

[WriterOfMinds]

So it's going to be subscription based - the game has to contact a remote server to run the characters?

... it’s easier to figure out how to serve AI answers to millions of players, who might interact 20 minutes a day for a month with characters at the cost of maybe $1.

Then if the company runs out of money and turns off the service, all the NPCs in your game go brain-dead. That seems like a risk, especially for non-MMO games that are intended as static pieces of art. And obviously this removes any ability to play without an internet connection.

[infurl]

So it's going to be subscription based - the game has to contact a remote server to run the characters?

That's pretty much it. While the risk of the company going out of business is a grave one it hasn't stopped hundreds of millions of people, myself included, from enjoying games like Runescape, World of Warcraft, and Eve Online which have lasted for twenty years or more and continue with largely undiminished popularity.

In fact, the behavior of some of the other players in Eve Online is such that I joke that they are just NPCs who have been tricked into paying a subscription in order to exist!