AI Breakthroughs: Supercharging Cybersecurity with Reinforcement Learning #shorts

Now we see models performing at 80 90%. And to understand the progress and the excitement right now around geni which is not really around geni it's about reinforcement learning I'd love quickly to introduce uh the differentiation between supervised and reinforcement learning. But what's important is in the past in many cases in cyber security we've tried really to label the data and that's our biggest problem. We don't have a lot of labelled data because the complexity of cyber security is just so bad. But this thesis is everything where it's easy to verify solution but hard to generate machine learning is going to solve. And indeed we saw this progress across many of the use cases in the

past. So to summarize that the you know what this man model teaches us is that the ease of training AI to solve a task is proportional to how verifiable a task is. Which means all tasks that are possible to solve, all tasks which are possible to solve in an easy way will be solved by AI or can be solved by II. And I think this is really really important statement for us to understand how this can be applied to cyber security going forward because just looking at all the other benchmarks and what is a benchmark a benchmark is a mechanical verifier where you can very easily verify something. All the recent big benchmarks which we saw around mass

um you know medicine health whatever they've been all solved by machine learning literally within a couple of years and now if we progress to cyber security we're seeing a similar progress now we see models performing at 80 90% %