SHOW 5 min
0:00 cloudflare just launched a new security service powered by openai's gpt 5.6 cyber it's not just scanning code it's watching live internet traffic to find and fix vulnerabilities as they're being exploited in episode one sixty two we talked about the long road from a new model to a real world tool today that road just got a lot shorter0:21 for security teams here's what else is moving first google deepmind just shipped weathernext three this isn't another research paper it's an ai model trained on raw satellite data that is now live in google search and maps it delivers hourly forecasts at five kilometer resolution and boosts precipitation accuracy by up to 50% in regions where forecasts used to0:42 be a coin toss this is one of the clearest cases yet of an ai model flat out beating traditional physics based simulations in the real world second the hardware is catching up to the software lenovo just announced a new line of ai capable desktops the thinkcentre x ultra built around amd's latest ryzen ai processors these machines can1:02 run large language models and multi agent systems locally with up to 128 gigabytes of unified memory the goal is to move complex ai work off the cloud and onto your desk tackling data privacy and latency in one move it's a bet that the future of enterprise ai is hybrid and third the big reality check while 82% of1:23 small businesses now say they've invested in ai the story for larger companies is complicated studies are showing enterprise ai project failure rates anywhere from thirty percent to over eighty percent and in health care despite the fda authorizing over 1,600 ai enabled medical devices a new survey shows only 14% of us hospitals have actually deployed them into routine1:43 clinical use everyone is piloting almost no one is in production so let's dig into that gap why are some of these ai tools landing with incredible impact while thousands of others stall in pilot programs the answer isn't the model it's the data and the workflow let's go back to cloudflare their new service is called vulnerability discovery and2:02 remediation what makes it different isn't just using a powerful model like gpt 5.6 cyber the secret is that cloudflare feeds that model its own unique proprietary data live telemetry from trillions of network requests hitting its servers every single day think about that most security tools scan your code and tell you about theoretical vulnerabilities they use a generic2:25 severity score cloudflare's ai does that but then it cross references those flaws against its live view of the entire internet it can see if attackers are actually using a specific exploit right now that allows it to prioritize threats based on real world risk not a theoretical score it separates the urgent from the merely important as ceo matthew2:48 prince put it if your security team is manually fighting ai driven attacks you're not just burning them out you're losing the service then offers two paths for immediate protection it can deploy a custom firewall rule at the network edge for a permanent fix it generates a code patch for your engineers to review and approve the human is3:07 still in the loop but the ai does the discovery the prioritization and the drafting that is a tool that fits directly into a security team's workflow and solves their biggest problem noise now contrast that with the situation in hospitals 100% of surveyed hospitals are piloting ai but only 14% have fully deployed why 77% blame tool immaturity 47%3:33 cite financial concerns the peterson health technology institute looked at ambient scribes ai tools that listen to doctor patient conversations and write notes even in hospitals where they're used the financial return on roi remains completely unclear the ai is available it's even fda authorized but it doesn't fit the economic or organizational reality of the hospital this is the3:55 pilot to production gap a report from intuition labs nails it the problem is measurable and largely organizational not primarily a technical one it's not that the ai can't do the task it's that the organization can't figure out how to absorb it how to pay for it and how to measure its value the difference between a successful ai4:15 deployment and a failed one is becoming crystal clear it has nothing to do with the number of parameters in the model it has everything to do with whether the ai is integrated with a unique data source that gives it an unfair advantage and whether it slots directly into a workflow to solve a problem that someone will actually4:31 pay to fix cloudflare is connecting a new model to its data moat google is connecting a new weather model to products you use every day they aren't just shipping ai they are delivering a complete solution
