{"id":"ynSwlkqyVeg","title":"Explain NEAR AI Like I'm 5","description":"NEAR is the unified commerce layer for assets and agents. \n\nWebsite: https://near.org/\nNEAR AI: https://near.ai/\nIntents: https://near.org/intents\nChain Abstraction: https://www.proximity.dev/\nSharding: https://nearone.org/\n\nFor the devs\nGitHub: https://github.com/near \nDev Docs: https://docs.near.org/\nDevhub: https://devhub.near.page/ @NEARDevHub\n\nConnect with socials\nTwitter:   / nearprotocol  \nBlog: https://near.org/blog/\nMedium:   / nearprotocol  \n#NEAR #nearprotocol #NEARAI","channel_name":"NEAR Protocol","published_at":"2026-03-15T00:00:00.000Z","thumbnail_url":"https://img.youtube.com/vi/ynSwlkqyVeg/hqdefault.jpg","transcript_text":"Let's do a quick Eli 5 around some statements made about Near AI. Near AI uses hardwarebased addistation, Nvidia confidential compute to cryptographically prove that inference ran exactly as expected. The Eli 5 is that the GPU has a tamperproof lock on it. When it runs AI, it signs a receipt saying I ran exactly this. Nobody peaked. Nobody changed anything. You can verify that receipt. Is it real? ZK is not ZK circuits. Those aren't computationally feasible for LM yet. It's TE at a station. The Eli 5 here is that real ZK math proofs for AI would take years to compute. That's like asking a calculator to prove it did a billion math problems. Instead, we use a secure box that can't lie about what happened inside it. You know how your phone has a secure chip for like Face ID that even Apple or Android can't look into? Same idea, but for AI. The chip does the work in a locked room and hands you a signed note proving what happened. When it comes to trusting hardware, you're trusting Nvidia silicon. The add a station is cryptographically signed by the hardware. The Eli 5 here is that it's like a notary stamp baked into the chip at the factory. Nvidia can't fake it. The cloud provider can't fake it. Nobody can. The chip literally cannot produce a valid stamp for work it did not do. Pure ZKML is two to three years out. Nearai ships verifiable inference today. The ELI 5 here is that everyone else is saying we'll have provable AI someday. But Near AI is working on it now. When that fancy math for ZK catches up, we can add it in. But right now, we're not waiting to see what happens. And we're using TE add astation to prove this today. That's a quick [snorts] Eli five on some things on near AI. Follow and subscribe for more of this.","transcript_synced_at":"2026-11-03T04:33:39.328Z","segments":[{"start_seconds":3,"text":"Let's do a quick Eli 5 around some statements made about Near AI. Near AI uses hardwarebased addistation, Nvidia","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=3s"},{"start_seconds":11,"text":"confidential compute to cryptographically prove that inference ran exactly as expected. The Eli 5 is that the GPU has a tamperproof lock on","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=11s"},{"start_seconds":20,"text":"it. When it runs AI, it signs a receipt saying I ran exactly this. Nobody peaked. Nobody changed anything. You can","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=20s"},{"start_seconds":27,"text":"verify that receipt. Is it real? ZK is not ZK circuits. Those aren't computationally feasible for LM yet.","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=27s"},{"start_seconds":34,"text":"It's TE at a station. The Eli 5 here is that real ZK math proofs for AI would","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=34s"},{"start_seconds":40,"text":"take years to compute. That's like asking a calculator to prove it did a billion math problems. Instead, we use a","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=40s"},{"start_seconds":49,"text":"secure box that can't lie about what happened inside it. You know how your phone has a secure chip for like Face ID that even Apple or Android can't look","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=49s"},{"start_seconds":57,"text":"into? Same idea, but for AI. The chip does the work in a locked room and hands you a signed note proving what happened.","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=57s"},{"start_seconds":66,"text":"When it comes to trusting hardware, you're trusting Nvidia silicon. The add a station is cryptographically signed by the hardware. The Eli 5 here is that","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=66s"},{"start_seconds":75,"text":"it's like a notary stamp baked into the chip at the factory. Nvidia can't fake it. The cloud provider can't fake it.","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=75s"},{"start_seconds":82,"text":"Nobody can. The chip literally cannot produce a valid stamp for work it did not do. Pure ZKML is two to three years","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=82s"},{"start_seconds":90,"text":"out. Nearai ships verifiable inference today. The ELI 5 here is that everyone else is saying we'll have provable AI","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=90s"},{"start_seconds":99,"text":"someday. But Near AI is working on it now. When that fancy math for ZK catches","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=99s"},{"start_seconds":105,"text":"up, we can add it in. But right now, we're not waiting to see what happens.","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=105s"},{"start_seconds":110,"text":"And we're using TE add astation to prove this today.","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=110s"},{"start_seconds":116,"text":"That's a quick [snorts] Eli five on some things on near AI. Follow and subscribe for more of this.","url":"https://www.youtube.com/watch?v=ynSwlkqyVeg&t=116s"}],"youtube_url":"https://youtube.com/watch?v=ynSwlkqyVeg","url":"https://near-pedia.xyz/videos/ynSwlkqyVeg"}