Illia Polosukhin: Why AI Agents Need Verifiability | Token2049
Nothing is loaded from YouTube until you press play, and then without tracking cookies.
About this video
NEAR co-founder Illia Polosukhin joined Avi Felman on 1000x at Token2049 Singapore.
If your AI agent points at an API endpoint, “you have no idea what’s running behind it”.
It could get a smaller or quantised model, fake output, or a bash command that steals your private keys.
”You want to know exactly what’s running there.”
#NEARProtocol #TOKEN2049 #AIAgents #NEARAI #Crypto
NEAR is the Blockchain for AI Website: near.org NEAR AI: near.ai Intents: near.org/intents Chain Abstraction: proximity.dev Sharding: nearone.org
For the devs GitHub: github.com/near Dev Docs: docs.near.org Devhub: devhub.near.page @NEARDevHub
Connect with socials Twitter: / nearprotocol Blog: near.org/blog Medium: / nearprotocol #NEAR #nearprotocol #NEARAI
Transcript
142 wordsCaptions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.
- 0:02
No guarantees they're going to respond to you with a real model response. They can do like smaller model, okay, quantized model or they can just insert fake output completely. For example, they can run arbitrary Bosch command and this Bosch command infects your machine with a virus and steals all your private keys for example, right? That would be great, but also like you would assume that OpenAI would have safeguards against that.
- 0:26
But it's not on OpenAI side. This is on your agent side. Okay. So you if you're running let's say Hermes and you point it at some API endpoint that API endpoint you have no idea what's running behind it. Right? I see where you're going.
- 0:39
So so this is where verifiability of block like of blockchain technology is very useful. You want to know exactly what's running