If anybody tells you that the AI is about to replace your job, share this video with them and this paper.

@sarainwondertech

12.3K followers · usually 1.2K views a post

Views

52.0K

Likes

2.3K

Comments

169

487

Why it worked

The video leverages a high-authority source (Google DeepMind) to debunk a common fear (AI replacing jobs), providing viewers with a sense of relief and intellectual value.

Summary

The creator explains the findings of a Google DeepMind paper regarding the limitations of AI agents, specifically their lack of self-adaptation and reliability. She argues that these issues make AI unsuitable for replacing human jobs and suggests a framework for better AI delegation.

Structure

1Introduction to the paper's core argument
2Explanation of the first problem: lack of intelligence and self-adaptation
3Explanation of the second problem: lack of reliability and trust
4Discussion of the risks of multi-agent systems
5Overview of the proposed solution for AI delegation
6Conclusion on the current state of AI and call to action

Call to action

share this video and hit follow

On-screen text

Google just exposed the biggest problem with AI agents Finally one clear explanation PROBLEM 1 PROBLEM 2 What can go wrong The solution

Original caption

Google DeepMind just published a 42-page paper called Intelligent AI Delegation and it's the most honest thing I've read about the state of AI this year. The paper breaks down two fundamental problems. First, AI is not self-adaptive. It doesn't learn from its mistakes in real time and it can't recover when things go wrong. Second, and more importantly, AI is not trustworthy. It's not reliable and we currently don't have a good solution for that. When you combine multiple AI agents together, which is what every company is trying to do right now, those two problems multiply. You get unreliable systems delegating to other unreliable systems with zero accountability and no way to trace what went wrong. Google's proposed solution is to treat AI delegation the way you would treat delegation in a real company, with contracts, verification systems, trust scores, strict permissions, and mandatory human approval for high-stakes actions. Essentially all the infrastructure you need when you can't trust the employee doing the work. The takeaway is simple. Next time someone tells you AI is about to replace your job, send them this paper. The people actually building the systems just told you how far we are from that reality. #AIagents #GoogleDeepMind #airesearch

More from @sarainwondertech

Their other posts in the index, biggest breakout first.

If anybody tells you that the AI is about to replace your job, share this video with them and this paper. Because this paper tells you exactly what are the two biggest problems in AI today and why it's not about to take your job anytime soon. Problem number one, it's not really intelligent. It looks like it, but it's not. AI doesn't learn from its mistakes and it's not self-adaptive. So we have to build all this infrastructure around it to make it useful. Problem number two, and this is even bigger, you cannot trust AI. It's not reliable and we don't know how to fix this. Now, when you start using multiple AI agents together, those two problems multiply. You have an unreliable system delegating to other unreliable systems. And that is the core argument of this paper. AI agents do not dynamically adapt when something changes. They operate in what Google calls a zone of indifference. They will do whatever you tell them to do without asking, does this make sense? And of course, when something fails, there is no accountability. So what you actually have is an enthusiastic intern that defaults to action instead of thinking. Has zero responsibility and fails often enough that you cannot give them anything important. You constantly have to supervise them and that's kind of the opposite of an automation. The solution Google answer to this problem is to treat AI delegation the way you treat it in a real company, with contracts, trust systems and accountability. Every task needs to be verifiable before it's delegated and if it's not, it's broken down further. To build trust, everything is documented using blockchain and permissions are granted only for the execution of the task and revoked immediately after. And last but not least, every high stake decision has a human in the loop. But even if we had all of this in place, which we don't, it's still not foolproof. If this is news to you, chances are it is news for most people, so share this video and hit follow for well researched AI content.