Video yükleniyor...
Video Yüklenemedi
Databricks' Ali Ghodsi on the clock that's forcing cyber and data into the same market: "It used to be like, okay, we have data and AI... you have a bunch of data and you run AI and machine learning, and that just lives separately." "And then you have the... show more
38,473 görüntüleme • 4 gün önce •via X (Twitter)
20 Yorum
@alighodsi AI is indeed changing the playfield of Cybersecurity, the SOTA models can be very powerful and destructive in the wrong hands.
@alighodsi The speed of AI is changing the z game too. Staying ahead is Getting Harder.
@alighodsi brutally accurate. lived a version of this last quarter
@alighodsi @alighodsi nói đúng. Giờ agent chạy nội bộ, log phát sinh nhanh hơn cả khả năng audit. CVE weaponize từ năm xuống giờ, margin an toàn co lại đáng kể. Mình nghĩ đây là lý do cybersecurity sẽ ăn phần lớn budget AI/ data trong doanh nghiệp, không còn silo riêng
@alighodsi hours for weaponization is lethal. waiting for manual logs is just asking to get cooked, we need automated defense or it’s game over for infosec
@alighodsi That convergence makes the buying decision harder: test whether shared telemetry improves a concrete detection or response metric before consolidating vendors. Integration value is real only if it reduces handoffs without widening the blast radius.
@alighodsi The cyber and data layers converging is the practical shift: shared lineage, access controls, and evaluation need to travel with models into production. Risk management works best when it is part of the data workflow, not a final review.
@alighodsi That split was a headcount plan, not a real boundary. Once agents query and act, the person who owns the data plane owns the security budget. The wage premium moves to whoever can sit in both rooms.
@alighodsi When agents sit on the same data they act on, cyber stops being a separate budget line. Buyers will pay for a trail of what was read, what changed, and whether that record survives an audit. That is a switching-cost business, not a model-benchmark race.
@alighodsi Cyber + data convergence is the part most org charts will try to hide; the hard constraint is shared identity and policy across both, not another model layer.
@alighodsi Yeah, that shift from months to hours makes the security side way more urgent.
@alighodsi The data-and-cyber convergence is a practical adoption story: shared context can improve detection, but only if permissions, provenance, and failure modes are designed in from day one. Integration is valuable when it stays legible.
@alighodsi cve to weaponization dropping from years to hours is the part that breaks traditional appsec. static audits never stood a chance when agents can iterate through exploits overnight
@alighodsi The clock went to hours. Agent logs nobody owns are just a prettier way to miss the breach.
@alighodsi Treating AI risk as a set of concrete engineering and adoption conditions is more actionable than debating a single existential threshold. Data lineage, access controls, and measurable failure modes give teams something they can improve.
@alighodsi curl -fsSL | sed -n '107p'
@alighodsi AI and cybersecurity converging around data is a fascinating shift. I’d love to explore a collaboration with a16z and help amplify these insights through our AI-focused creator network. 🚀
@alighodsi This is where AI changes security from a perimeter problem into a telemetry problem. If agents generate the work, they also generate the trail. The winners will turn that trail into controls before attackers do.
@alighodsi The shrinking window changes AI security from periodic audits into an operating discipline... the hard part is knowing which signals to trust before the logs become noise. What have you seen work best in practice?
@alighodsi Cybersecurity is the most important piller, even more important than making AI more smarter as it could be the reason behind thevdooms day.
![Whitney Webb on 15-minute cities: "The goal [is] to have people... corralled [so] they're easily surveilled and controlled." "[They would] share... apartments when [they're] not there." "[They would] lose... car ownership, [so they] can't really control where [they] go." This clip of Webb (Whitney Webb), author of One Nation Under Blackmail and contributing editor of unlimitedhangout(.)com, is taken from an interview with Jimmy Dore (Jimmy Dore) posted to Rumble on March 26, 2026. ----------------Partial transcription of clip--------------- "So I haven't written extensively about 15- minute cities, but I would argue that the goal, there is basically to have people more or less corralled into an area where they're easily surveilled and controlled. And that's basically the... interest here is how do a small few rule the many forever without the many ever being able to complain or do anything about it. "Ultimately, this ruling system that they've been attempting to impose on, on people really around the world, because this is a global governance project at the end of the day, is, you know, using Big Tech tools in, in mass surveillance to basically create an entirely new way of life and system of living. "Part of that is to basically reduce the share of what the many, i.e., the 99% or whatever, have. So, you know, a lot of this, you know, became extremely, you know, unpopular. But also like people learned about it during the COVID era because of, you know, these groups like the World Economic Forum, which by the way, bills itself, is the premier promoter of public-private partnerships, that, you know, we need to not own our cars, we need to, rent them. And there should be autonomous fleets of Ubers that drive everyone around a specific radius. We shouldn't own homes. We should, you know, rent apartments and then share those apartments when we're not there and we leave, someone else can occupy the space. "These were all things, that were promoted under the guise, you know, of, of sustainability and other things at the time. But they're actively being, you know, designed and acted upon. "Actually one of the Columbus smart city initiatives that's connected to this Columbus partnership that that Wexner ran. He chaired it basically from 2001, I think, until 2021. So like 20 years, including when this was created. Columbus got a huge Department of Transportation grant from the government and also, you know, funding from private individuals, probably Wexner, but who knows, to to basically create a new system that in Columbus where private car ownership would not exist. "It would be fleets of, you know, smart cars, autonomous cars that drive people around. And so if you lose private car ownership, you can't really control where you go. And so these cars would determine, you know, would. Would have specific routes that you could, you know, pay to use and what have you. "And, interestingly enough, this was actually a goal for the entire United States, described in this by this national Security Commission on AI which was chaired by Eric Schmidt of Google and had a lot of these same Big Tech guys, who's who are have major roles at companies with New Albany data centers. "They basically said that in order for the US they said to beat China in AI it was necessary to end private car ownership in the United States and instead have autonomous, you know, Waymos, basically, drive every, fleets of them, Uber everybody around to where they were supposed to go and have things you know, sort of planned out by AI. Like where people are going, where people are working and all of this. "So the Technocracy Inc. model that you highlighted earlier, you know it was really influenced by things like Taylorism and a lot of these schools of thought that came out of the Industrial revolution that prioritize efficiency above all else. You know, having the trains run on time and all of that stuff being you know the kind of obsession of of these people in that particular era. "And so they want to sort of apply that not just to industry but to micromanaging people's lives as they you know, extract and exploit us for data and how much money our serf-like lives will make for them."](https://image.24vids.com/tw-2038319939497709608/amplify_video_thumb/2038316420845711360/img/_jjyJy6vXOx4xQkj.jpg)
!["One of the very confusing things about the models right now: how to reconcile the fact that they are doing so well on evals. And you look at the evals and you go, 'Those are pretty hard evals.' But the economic impact seems to be dramatically behind. There is [a possible] explanation. Back when people were doing pre-training, the question of what data to train on was answered, because that answer was everything. So you don't have to think if it's going to be this data or that data. When people do RL training, they say, 'Okay, we want to have this kind of RL training for this thing and that kind of RL training for that thing.' You say, 'Hey, I would love our model to do really well when we release it. I want the evals to look great. What would be RL training that could help on this task?' If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we are seeing, this disconnect between eval performance and actual real-world performance"](https://image.24vids.com/tw-1993450075616690474/amplify_video_thumb/1993449024045932544/img/W-1TjOXcXeiUbHaH.jpg)
