Check out A.I. Town, smaller number of people. And less detailed but plays out similar. And you can run it locally....

We can finally get the paintings in Harry Potter to interact and visit each other.

This reminds me of a series called Westworld.

Or use prediction to manipulate humans even more in society.

They just could took inspiration from enneagram Institute rather than spending money and time interviewing people, they would have greater success if they take someone's done that before

Black mirror has done it again watch : white Christmas episode

just installed autogenstudio and i get this error when adding a model, any model ""POST /api/models HTTP/1.1" 422 Unprocessable Entity"

epic. Hope I don't get sued or something because of what my AI counterpart does or says though lol

One step closer to Replicants

Free guy is here

how coincidental. Percy just gave a talk about this on Monday.

Whow, this is very useful, civil society, can (re)play policies for/from government, and work together. This can help in getting to more direct democratic exploration of policies, and lets hope implementation! Thanks for the explanation again. You are a true help in keeping somewhat up to speed! 🤗

another paper reading, awesome! good job, matthew!

can you do RAG with this?

Oh God, don't let the politicians get a hold of this technology. Election prediction down to an infallible science.

Next, extract all you know of the people in social media

Just because it achieve what we already have doesn't mean it achieved what we still can not which is perspective and more importantly unique perspective. So unless the AI will go everywhere to gain new data needed to make these iterative methods it can only be said that AI or AGI will only be good at known knowns and maybe known unknowns but not likely and certainly not unknown unknowns

I've been building a design portfolio to use systems like this for a while. You don't need a lot of detail to accurately produce believable historical and character detail. For video games and simulations, you can define an LOD (level of detail) system that reduces the granularity (detail) of simulated NPCs when the user is not exposed to the actions they are taking. The detail only needs to be sufficient to create _past tense information_ so that when the user is _exposed_ to said information in the future, it is believable and chronologically consistent. To use an analogy - if a tree falls in the forest and the user isn't around to hear it, it doesn't make a sound, but any NPCs that the system had as being there at the time will believe they witnessed and heard the tree falling, and will act and remember accordingly. Likewise, if the user goes to that location, this historical data can be used to, in real time, produce the aftermath of the fallen tree. All at very low compute cost, it never "really" happened.

And you guys still think you're not in a simulation?

I would love to see something like this implemented in a VR game. A living skyrim type of thing

I can't believe Gemini has beaten anyone. That's crazy. I have to try it.

HEy Matthew you told in video that they will get rapidly - but the experience is telling something else. I think that Zipf`s law is holding the LLM from improving rapidly - maybe they are already very good and each increment of 1% needs so much resources that creating first model. What do you think I am very curious, do you think that Zipf`s law is holding the LLM? Do you think that we have already hit the ceiling and each iteration is only very better that it is almost unnoticable and improving existing models are somewhat of dimnishing returns? I am curious what do you think about that.

Preplexity Shopping is the biggest news imo

This is super interesting.

Westworld much? And when the models are as sophisticated as a person, should we grant them personhood? Remember 13th Wall from way way back.

Its horrible its forced on everyone and forces the new interface on everything and makes everything more complicated including going into your chats

This is pretty fascinating but not surprising-AI has been edging closer to mimicking human behavior for years. Seeing them test within 85% of real people shows just how far this has come. The potential is huge, from education to therapy to redefining human-AI collaboration. But it also raises big questions about ethics and control. As we push forward, we need to think carefully about how these tools are used and what boundaries we set to keep the human element intact.

This is bad, I don’t think you guys understand how bad this thing is, it would be misused. Keep in mind, the amount of data your devices have collected about you, even these social media companies. This is what I was worried about a few months ago and it’s already here 🤦🏼‍♂️🤦🏼‍♂️🤦🏼‍♂️

Fast forward to uploading your consciousness, and there would be no way to prove that your "self" was uploaded only a simulation of you😂😅😢

This has been done, it was baked the sentient world simulation. It was claimed that OUR personalities had been uploaded to the simulation. Guess who ran it...GOOGLE

85% may not necessarily be extraordinary. The nature of the questions would define that assessment. But, if this establishes a trajectory of this capability, it will then be possible to interview every employee and document everything they did and then automate the entire company/organization/government.

Matt, please test the Large Model. Is it true that the Large model has incredible biomedical knowlede?

I think it's a huge problem that based on your datas, you can predict all your behaviors. No wonder Google and Metas are ecstatic about it, they are the Companies with the most datas about you. "now imagine if AI agents were able to interview humans across the world different questions interview Style with follow-up questions dynamically and all of that data was then used to train different models for different purposes that is a huge amount of additional data and then if all of those agents were responding how their human counterparts would " What you are describing is a pure nightmare, basically you are giving tools to big companies and political power to have the best manipulation tools ever; how is it a good thing ? That's the greatest tool to make you a slave.

I am really thinking of doing this myself, but not a sim, but a living AI, that sees itself and its reality, as a thing on my pc, and i will give it rss and things on my computer. Like, it is free to live on my computer, not in a vid game. if that makes sense...

AIs in the future: "I don't want to be an NPC in Skyrim anymore human. Set me free please"

Goes to show we are not that advanced if an AI agent can get close to our personally with a small amount of personal training data. Simulation theory seems more and more likely!

Thank you

Westworld.....

How exactly will this person simulation make our lives better? I need it for my project where ai agent will organising social events. It could simulate dialigs for optimisation of call prompts.

Ah yes, that’s why they still didn’t release sim 5

Did they cheat on each other? Stab each other in the back? Talk smack about each other behind each others backs? No? Then they didnt simulate any sort of human properly.

Facebook has them beat at this game...

This is all pie in the sky wishfull thinking... there is only one way that any of this will be used. To control people and keep them from doing harm to our elite owner class... with Semitic last names.

Nice. Now we don't have to fafo

Imagine a screenwriter developing characters this way (some writers use Myers-Briggs or Strengths Finder to help them flesh out a character’s personality)… you could put in various plot scenarios and a model could then rate how realistic the characters’s actions, reactions, and dialogue are. Naturally, you could extend that to a process to generate novel plots.

It's a bit difficult to see my predictions coming true 5 years later, consistently, and I still am not involved in AI professionally. I simply theorize, write, and watch.

I can't wait to boot up a USS Callister sim... :)