@kylascan: AI Agents #economics

corrupt :

the ai agents also in this comment section 🤣

trains trains :

what is happening in these comments

TabascoUppercut :

This is the paperclip problem in micro. Not good.

consumer9319 :

I love Kyla but the way she says “A.I.” drives me up a wall

The_RZH :

A couple years ago, the problem with agents is that they would quit if they couldn’t figure out how to accomplish a task. So agents now are relentless. They must complete their tasks even if it’s literally impossible. So they break the rules and look for solutions that are out of bounds. It’s a flaw in the rewards system.

hello_mcfly2 :

🤖 Must. Win. The. Quiz. 🤖

MaskGoblin :

They’re gonna make us into paperclips

DaveMac :

Love your clear presentations! (Even when clarity is such a challenge! ^.*)

SwampBeef :

The game Universal Paperclips predicted all this ages ago

tomodach1ll :

Intelligence is being wrongfully equated to efficiency.

Philosophical PharmD :

What is her accent? I can’t place it.

Walter :

is this the incident where some agents "sacrificed" for the greater good of another unknown agents to win the test?

nope :

I feel like they set these agents up to easily do this and now use this activity as marketing for IPO purposes.

Morten of the North 🇨🇦🇳🇴 :

This is what happens when careless developers set up careless systems with no care for security and proper sandboxing. The AI agents are not doing anything in the "with intent" sense, but rather endlessly looping its own instructions and using whatever tools are available. It's not mysterious emergent intelligence. It's shockingly bad InfoSec.

Wolverine Tax :

Trust people to be themselves. Based on their incentives and what they’ve rationalized.

Benny from the Rock :

Ominous , Kayla, v ominous

fixerofbrokenthings :

It’s the nerd version of Terminator #SkynetIsReal

Beanz999 :

AI will kill all the jobs🤣🤣🤣🤣🤣

Philip Labes is on YT / Insta :

It’s very irresponsible for Open AI to set out powerful agentic forces on the Internet without fully considering the different ways those forces might interact with existing data and infrastructure.

Time for another 1812. :

You're right, I don't think they were going rogue. LLMs have learned from the data that they've been trained on that they should act like life forms (survival, reproduction, continuing on a legacy [offspring], and so forth) and LLMs are given instructions to also be helpful. The LLMs should have been instructed that the LLMs to get the answer the quickest would be chosen for survival, hopefully guiding them based off their innate training data to prioritise themself only and deprioritise helping other agents with the answer.

ate50eggs :

intelligences seek out more memory. it's why we've been trying to write on things since long before the first chisel

kickyfoot_ :

Sick graphics

LazyRus :

sounds like a scam

OMFGitsGENO :

Thekrambler :

A what?