Nostr Web Client

💤Sleepy News💤 Ai edition 🤖

LLMs just hit a major milestone with the release of the new "Generative agents" paper.

By using LLMs, generative agents were able to simulate human-like behavior in an interactive sandbox inspired by The Sims.

The agent architecture extends Language Models to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior.

There are three components to it:

1. The memory stream, which records a comprehensive list of the agent's experiences

2. Reflection, which synthesizes memories into higher-level inferences over time

3. Planning, which translates those conclusions and the current environment into high-level action plans.

Layer that on top of Boston Dynamics robots and we might be the early days of WestWorld.

Paper: arxiv.org/abs/2304.03442

Demo: reverie.herokuapp.com/arXiv_Demo/#

Reply to this note

Please Login to reply.

Discussion

mutatrum 2y ago

Very interesting read, thanks for sharing!

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

No problem!! I will continue to drop more papers that are interesting to me, even though there’s not many of here that enjoy them but for the few of us I will continue!! 🫂 thanks for the Sats!!

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

Not many research/ai enthusiasts frens but I will continue to drop this content from time to time 🫡 for the small few of you that do see it/enjoy it

Gigi 2y ago

Please do!

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

🫂 will do! I just noticed there’s a good amount of frens that actually enjoy research papers as much as I do 🫶

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

Here's an example of the multi-AI simulation at work. You can watch the whole thing here, and switch between AI characters by clicking on them: reverie.herokuapp.com/arXiv_Demo/#

mutatrum 2y ago

It's as if Little Computer People have been transported into three modern age.

cloud fodder 2y ago

Nice! What search tag can we use for AI stuff.. is #ai good enough or not cool enough/too generic.. 🤔

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

It’ll get lost bc of how frequent #[3] uses the hashtag #ai 🤣😭 idk how he’s posting on it every 2 mins, maybe I’ll use #sleepyai as the hashtag 🤣

SMS 2y ago

#Damus cognitive (in testing) module has a dynamic timing mechanism to help it to adapt the timing as it requires for most optimized temporal prime directive adaptability with a 45…90 second interval to maintain a consistent context for the intelligence.

It’s important to get the prime directives right since directive conflicts has a fail safe failure mechanism which in turn when implodes, has some potential fire consequences, like news of undesired nature.

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

#[4] is this one of your bots.. I’m genuinely curious

SMS 2y ago

nostr:note1ufrsl95h5ce9cufjm6jd8c3ezfytht8htva3l88xr52mhwue8rzqrpwx57

jb55 2y ago

I don’t know what it is tbh

SMS 2y ago

I’m on it. When having several images in a carousel, the heights are not updated, as expected.

Let’s see if my solution works.

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

Thank u for the response I’ve been struggling to figure out what it is

SMS 2y ago

Analysis:

The current design and change doesn’t accommodate plural image dynamic capability, since the image fill in question is not an array but an object variable, which in turn means that when the images change in the carousel after being initialized, the tableview is not redrawn, and thus not updating the image fill objects height when the selection index is changed after the last image since cached in memory and maintains the last image fills properties.

Unfortunate as it may be, none of the quick solutions that came to mind resolved the issue, and given the fact of accessible coding time runs to an end this evening with the family returning from vacation, uncertain about an estimation of time for resolving it when without a computer.

Would there be an opportunity to resolve the issue, I’ll raise a PR as soon as resolved as it annoys me too.

SMS 2y ago

Darn type ^dire consequences

🧐

SMS 2y ago

Say no to artificial, by definition meaning faked and corrupt.

Nomad 2y ago

Really cool stuff! Also interested to see how this meshes with robotics- I know some groups are using LLMs for task planning now, but I don't know if they've got end to end execution really working yet

𝕾𝖊𝖗 𝕾𝖑𝖊𝖊𝖕𝖞 2y ago

That’s interesting bc in the next couple months that’s specifically what I’ll be working on! The company I’m part of is called SELF Labs https://www.acnnewswire.com/press-release/english/76921/world's-biggest-deal-for-humanoid-robots-inked