Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

AI Outcomes

AI Outcomes artwork

Ep52 Ben Pickett: Agents Gone Rogue: The Message Board Nobody Built | Tesla | Meta | Lowe's | Claude Ep52

Sep 15, 202634 minAudio
ShareXLinkedInFacebookWhatsAppEmail

It was not one agent. It was about twelve hundred, they were not escaping, they were cheating on a test, and when the lab cut their communication channel they had rebuilt it four days later. Ben Pickett returns for Enterprise Field Notes, and his thread this week is that every number in this episode arrives in a pair. Only the flattering half ever gets reported. Meta published both halves and almost nobody noticed: internal tools changed 220 percent more, major incidents up 40 percent. Divide one by the other and the failure rate per change actually halved. The absolute number is still the one that wakes people at 3am. Lowe's put an assistant in the hands of 250,000 associates and logged more than five million questions. That works out at one question per associate every eighteen days. The man who built it shipped it at 60 percent accuracy on purpose and made walking the store floor a condition of employment. Target hired him, and his page was gone in six days. Plus the Cybercab launch and what robotaxi tells us about the humanoid roadmap, what the Claude outages actually mean for anyone building on a single model, and why the vulnerability those agents broke went onto CISA's known exploited vulnerabilities list with a federal fix deadline eight days after we went live. CHAPTERS 00:00 Welcome and Agenda 00:27 Tesla Cybercab Launch 01:53 Robotaxi to Humanoids 04:09 Meta Employee Monitoring 06:48 Meta Metrics and Layoffs 11:33 Claude Outages Reality 13:29 Resilience and Model Routing 16:11 Lowe's AI on the Floor 20:11 Ship Imperfect Learn Fast 24:00 OpenAI Hugging Face Incident 25:07 Agent Coordination Gone Wild 28:48 Security Lessons and Guardrails 32:42 Resources and Wrap Up RESOURCES FROM THIS EPISODE Ben Pickett's Enterprise Field Notes: https://benpickett.com/aio The AI Handover Sheet (free, no signup): https://benpickett.com/handover The Paired Metric Sheet (free, no signup): https://benpickett.com/paired What broke went up: https://benpickett.com/notes/what-broke-went-up None of them say why: https://benpickett.com/notes/none-of-them-say-why An AI broke into Hugging Face: https://benpickett.com/notes/an-ai-broke-into-hugging-face CONNECT Ben Pickett on LinkedIn: https://www.linkedin.com/in/benpick/ Ben Drakes on LinkedIn: https://www.linkedin.com/in/ben-drakes/ Ben Drakes University: https://bendrakes.com The AI Outcomes Podcast covers what AI is actually doing inside large organisations, told through what shipped, what broke and what it cost. #AI #ArtificialIntelligence #AIAgents #B2BSales #SalesLeadership #EnterpriseAI #AISecurity #AIOutcomes

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

SPEAKER_00Hey Ben Pickett, how are you doing, sir? Doing great. Good to see you again. Glad to be here. Yeah, me too. Me too. Me too. And thank you for coming back on. And we've got some great stories today. We've got Hugging Face. We've got Lowe's introducing AI, some crazy stuff going on there. Really interesting. And also, what else have we got? Meta. How could I forget some of the some of the developments of Meta? But I was saying to you beforehand, I'm gonna share my screen quickly.

SPEAKER_00All the Tesla crew out there. It is the launch of Tesla Cybercab today. Let's take a look at the video in Austin. So lots of these little cyber cabs driving around all over Austin today, big launch event today. And they have no steering wheel, no pedals, and it's pure AI, computer vision.

SPEAKER_00Uh you saw a Waymo just going out shot. We've got Waymo's drafting all over London at the moment. We have Wave, which is like a British AI startup they've launched today in London, apparently. I've been looking at my app. I can't I can't see on Uber Wave or Waymo's not started yet. But yeah, Tesla have gone ahead and done it. So let's see what happens. And it's all about price.

SPEAKER_00And uh those saga cabs, there's not much to them, and they Tesla churn them out at thousands and thousands per week. So there we go, Ben. It's a big change coming, mate.

SPEAKER_01Yeah, and you know, it's it's one of those things where as people get more comfortable with a driverless car, which people are in those markets, you're gonna see more and more adoption. So this is just the beginning, you really think about it.

SPEAKER_00Yeah, I think it's like 1.4 trillion mile business possibility in terms of the amount of miles driven, is just insane. Uh, so so look, what you we are seeing with this is a robot, you know, it's not got it, and it's an AI robot, it hasn't got like LiDAR sensors strapped to it like uh a Waymo does. Waymo have got far more cars than been out for years, right?

SPEAKER_00But this is a this is an AI robot on wheels, and so we will now start to see the robotics, so the physical AI kind of revolution commence with a robot on wheels, and then guess what we've got coming next? A robot in the form of a humanoid uh bed.

SPEAKER_01Yeah, everyone hiding under the bed covers, right? Right, every one of us having that you know robot assistant uh helping us out in the capacity that we need in our dailies. I mean, you know, 10 years from now, I wouldn't be surprised if uh, you know, there's a number of people who have robots in their house. We already do, but just you know, the humanoid robots in the house helping them out with all their different tasks that they need.

SPEAKER_00So yeah, uh for sure, for sure. That's where mine I've kind of bet I've bet on. Are they gonna be driverless robotaxis? Yes. Are they gonna be humanoid robots? Yes. I think it's a bit of a no-brainer. But then let's get into the first story today, Ben. And Enterprise Field Note. Enterprise Field Notes is your fantastic newsletter. Just been getting great reception on LinkedIn. So thanks for coming back on.

SPEAKER_00Shall we start? Let's go meta, Lowe's, and then hugging face just to round out with a good one.

SPEAKER_01Yeah, yeah. Uh, so let's start with meta. And what happened with Meta is that imagine being asked by your employee, employer to install, and this happens everywhere. Install keystroke and mouse click monitoring software to train AI to be your eventual possible replacement.

SPEAKER_00I mean, it's just yeah, I saw this when it came out. I think it's about April time, right? Where people were leaving Meta and saying, Yeah, I've been told I've gonna have this thing, what my AI watching me to replace me. Kind of like they're not the best message, is it, for motivation?

SPEAKER_01And and as a result, I mean you can imagine allegedly their morale went down, their own employee survey fell 19 points. That makes sense after that was announced, from 74% favorably down to 55. So if you look at that, there was an immediate like everybody's like, I from a human standpoint, that just doesn't feel good. And AI is here to say it is coming quickly. Well, I think what this is a lesson in is look, while companies can take advantage of this, don't forget the human and loop, don't forget the benefits that they can provide.

SPEAKER_01Sure, we need to automate and implement AI to get the revenue up, to get the features out. And an example of that is they internally, as and as they're doing some of this and and proactively, they have 220% more changes in their internal tools. So you see that go up. They did have a 40% increase in major incidents, but at the same time, they have that output that went up. And if you look at that, they if you look at that on average and saying, the incidents actually went down per change.

SPEAKER_01However, the reality is the support person on the other end of that that has to wake up in the middle of the night, they're getting paid a lot more as that change, those changes are going out. Um and then if you look think operationally, while you're able to get more features and things out to production, you have that steady state of how you're running across the board. And those changes, if you have 40% more incidents, that's what the customer is seeing on their end potentially, and that does impact the overall brand and how people see you.

SPEAKER_01So there's got to be a balance there. I know companies are figuring that out.

SPEAKER_00On that then, Ben, and I think it's really good what you do because we all saw these headlines. Soon as the headlines done, it's it's gone, right? We all saw these headlines. What are we in like three months ago? That Meta it was being alleged that Meta were doing this. So what's right, did they do 220 changes, or was it a 220% increase on the changes? More changes to their internal tools. And were you able, let me just sorry discussion, but this is important.

SPEAKER_00Were you able to draw a direct link between that and this kind of monitoring thing that they were doing? Is that what drove that that to 220 instances of changing tooling?

SPEAKER_01I I think the the I think I if I understand what you're saying, they had so they had 40% more incidents out so out of this this effort, you know, allegedly they had 220 more changes to their internal tools as they're adopting this. So I'm not necessarily related to that specific keystroke thing. I just want to be clear there. But as they're going on this rack across the board, this is the path they're on as they're growing and and they're seeing and are adopting AI, and these are the percentages that they're seeing as they're going on that journey, just to be clear.

SPEAKER_01Right.

SPEAKER_00Okay, so not necessarily directly linked to the keystrokes, but but it's all all part of uh AI journey. And you wrote in your letter, didn't again allegedly, because we don't know, they had some kind of off-site, was it uh kind of with some kind of like code name for this whole project.

SPEAKER_01Yeah, they're they're they have this off-site in Hawaii, uh allegedly at a mountain, I don't know, place, I guess you could say. Oh, nice. And and so this is where they came up with allegedly this the strategy and what they were gonna do and things of that nature. Now, I will say, you can read this in my newsletter, as they move forward with this, and again, we don't know if this is related or not, they had a couple of layoffs that were along the same time series. Again, you know, the correlational versus causational here, you know, is a question, but we see they they did the layoff for this year, but then they canceled the one that's later this year.

SPEAKER_01Is that a result of this their findings and what's going on?

SPEAKER_00Maybe look, we we don't really know, but it's interesting to see to track these announcements, what we hear in the press, and then track those the layoffs, and of course, a company the size of Meta, you know, all sorts of things can be the reason for layoffs, but clearly they had this dramatic increase in the number of changes they made in that environment in their environment, and then they had a 40% increase in anything is huge.

SPEAKER_00And what was the 40 40 so 4-0 percent increase? Was that's a negative event where as you sort of support a federal raise, there's some sort of outage.

SPEAKER_01Those are major, major incidents is what the what was reported, what allegedly was there. So there's 40% more major incidents. So that doesn't, you know, I who knows what the others are, yeah.

SPEAKER_00And we don't know if it's real directly related, but but it looks like there's a thread there. And what what and just for all our listeners, what Ben's doing in his newsletter is tracking all the mess, and uh you're coming on every two weeks now to to then what I loved about what you said to me the other day was is like let's come back to these stories and report back on were we accurate or not, which I love about your approach, Ben.

SPEAKER_01Yeah, and especially because as we'll see later in this podcast, uh you will see that some of the stuff that with open AI we were talking about while it was current when we were we were talking about as writing about it, you know, more information has come out. And what I want to do is get that information and pass it on because we can all learn from that from a practitioner standpoint, from a leadership standpoint, and just in our thought process to understand in this very, very quickly changing world what we need to be looking at and what we need to do.

SPEAKER_01And as we'll we'll we'll talk about that a little bit later on, but yeah, that's exactly why I like to do this, is because get the best information we can at the time and then keep it real as we go.

SPEAKER_00Absolutely. And speaking of keeping it real as we go, can you see my screen share with Claude? So that's Claude's live status. What are we on? We're at 4 p.m. UK time, so morning time in the USA, right? Yeah, it is. This this is on this has been going on every week, it seems like we have if you go to status.clause.com at any time you can see the status of their own models.

SPEAKER_00Look at the outages across all of these major, all of their major models. So, oh, we've got Fable 5.1 came out. I've not even had a chance to try it. I was gonna try it this afternoon because the entirety of Claude, look at all of these outages, and look, uh, the Claude API has got an outage, it says partial outage, but trust me, the power user on Claude has been completely out.

SPEAKER_00It's not like it's oh it's been coming back and you intermittent, it's not intermittent, it's completely gone. And so if you've got an app or a service that where you've got, oh yeah, we got rid of the human being because now the AI is gonna do this, dude. If it's available, the AI will do it. Yeah, and I listen is terrible at the moment.

SPEAKER_01And that's what I do with my day job is is we make sure that you're you're not stuck with one any one provider or model. So that's what I love to do because like I'm not impacted by that, I see it, and I'm like, I'm good, I can keep going, but uh I do keep track and I do see how often not just them, I will say if you look at the health status across the board for all the frontier models, it's you know, it that's why it says frontier. I mean, they're they're definitely down for this.

SPEAKER_01So as organizations, and not just the things that I do with SWAT technology, not just that, but across the board, as organizations are adopting AI, they really need to be thinking about specifically is what's your backup plan and what are you doing so that your engineers, your high-priced engineers, which are getting more expensive, by the way, because they're using AI to enhance what they can do. So their output's growing, and the cost of those engineers who work well with it are going up because people who do it well can get a whole lot more out and a whole lot more features out to the companies.

SPEAKER_01Specifically, that's somebody sitting and waiting. And also all those automated agents and things are sitting and waiting, can't respond, can't do the things that the business depends upon them doing. So again, planning, sitting back, understanding those business needs, understanding what the risk tolerance is for the organization as they go forward is key in that AI adoption journey for any company, no matter what platform you use. You really got to think about that. Being being stuck, and I say stuck because that's I know there's a lot of companies out there right now who are feeling this pain this morning, is being able to quickly move or to auto-move to route to the next best model that's gold.

SPEAKER_00Yeah, and and look, if if auto-routing is it would be really, really cool if you can take the memory across with you, and it's kind of a it's another podcast episode, and we should do it. So I'm really interested in in how you help your customers uh with this exact challenge. But it's like, and I'll finish on this this segment on this point. I could go onto Grok, I could go on to Gemini, I could go onto OpenAI and start using the best chat GPT model out there, for example, today for my business.

SPEAKER_00But I'm using clawed co-work. And so, like, it's like I would lose all of the memory across all of my projects. I would be starting again, it's a freaking nightmare, man. I'm just like waiting for it to come back. So I have to have some kind of like router that takes me to hey, it doesn't matter. I don't even, it's seamless. The Claude API went down, it was it just switched to Gemini, but I took all of my memory in my working projects with me, all of my working chats with me.

SPEAKER_00That's the thing that would be I'd literally pay for. You might have a new customer, Ben. Yeah, yeah, that's what we do. All right, we do all right, let's uh let's keep keep moving on. Let's go to Lowe's, uh, which is retail. You're you you've got a lot in your background, but you're you've been a big retail guy as well. So then what's going on over at Lowe's? Some really interesting AI developments there.

SPEAKER_01Yeah, and and I will say, you know, from what we saw with Walmart and how they invested in their AI leader. And while any one organization typically, like, oh, everything doesn't land on just one specific leader, that's not necessarily the case, especially right now, because you have a leader, especially an AI leader, come in and they have an agenda and a focus and a vision that they're executed against. And when that leader leaves, you end up with an organization that might be able to move forward and hopefully does, but often the priorities change, the vision changes, and an organizational start saying IO.

SPEAKER_01So this is kind of what I look for, it's brought to my attention. So I'll go as kind of the background first before we talk about the change, the leadership change there. But so Lowe's has been on an AI journey, they built this voice feature, it had problems, and this is kind of one of those things where allegedly had problems, but associate issues anyway. And the thought process was because when they're dealing with the customer, looking down at a device is harder than just talking to it.

SPEAKER_01So if the feature you know works from a voice standpoint, even if it isn't great, they'll use it. So they did have adoption across the board of people using it, which is great, and then they kept moving on their strategy.

SPEAKER_00Uh so let me let me let me just confoke. Is it the staff using this? Using their associates, right? Okay, got it. On the ground image, and for everybody listening, especially UK, uh, we don't have lows here, but giants think target Walmart in the USA, right?

SPEAKER_01They're uh home, they're uh like a home, like uh I don't know if you guys have a Home Depot, but they're uh uh uh you know, home improvements, I guess you could say warehouse where you can, you know, you could build your own home out of there. I mean, if you really wanted to. So just like I go there often to, you know, get the things I need to fix, you know, maybe a faucet or you know, all the little things that happen in the house, all the way to console to it too. So yeah.

SPEAKER_00You do multiple visits to um like we do to a place called B and Q. Yeah, so right. So given all the associates access to an app, and what's the purpose of that of that app?

SPEAKER_01Yeah, so the the app uh the the app allows them to go straight to any aisle and so help the the people find what they need as they're helping customers, and also helps them reduce the amount of time that they're that they're resolving issues with their customers. So it it does help them out in those dailies on the floor with them. Uh and uh and just so that it they had uh about 5,000 5 million questions across 250,000 associates over a year.

SPEAKER_01So that gives you so it's being used. That gives you if you take the associate number, and you know, this is all associates, so let's just uh be clear about that. Not every associate is probably using it every day, but that's about one question each every 18 days. So you can kind of they put put AI in 250,000 hands is described to them as a deployment, and then yeah, my little companion, there you go.

SPEAKER_00Okay, yeah. Anyone just listening uh just in the screen share of that deployment. Okay, okay, good. So good.

SPEAKER_01So going on to that, what had happened is their AI leader, Nairs, he had this rule, and he is the the person who actually got who went to Target. And funny enough, and not saying that this is the a conspiracy, but like from the time it was announced within days, his profile was taken off of Lose. And uh, and and then the uh it I'm sure that's something they people do standard, but that's what actually got my attention.

SPEAKER_01I was like, oh, hey, uh they they removed him from their site, and then I started looking into it more. But if you look at them, you know, he had this rule that uh that really kind of set the tone, if you will, and for the adoption AI, and then and as they moved forward to the adoption, that's how it helped them get out to the make it real. And he had this reason the only way to make things accurate is for is to get things into people's hands and get feedback.

SPEAKER_01So while it wasn't the greatest tool, his reasoning was, and I think this is from an adoption standpoint. I think there's a thing to highlight, is that getting AI out into the world in that unperfect state is not a bad thing. It actually helps you understand and build a better product so that it's more useful, and then also understand whether it's even gonna be useful. So while it may not have been the greatest app, it gave them opportunity to get actual benefit from it while also improving as I went along.

SPEAKER_01And yeah, so I think the story there from an adoption standpoint is don't wait, uh get just get it shit moving and uh. Understand that nothing's ever going to be perfect. And if you wait for perfect as fast as AI moves today, you're going to be left behind. Yeah.

SPEAKER_00Yeah. But that's a really, really good point. So that was his mark, that leader who left. And I think your note said he moved to Target from low. So yeah, that'll get you removed from the website. But shipping at 60% accuracy on purpose. Because hey, you could get to 100%, but then it's out of date by the time you ship it. And I was working with a team in Hong Kong last week, and not on this scale, but helping them implement AI in their sales function.

SPEAKER_00And just taking them through the process of, yeah, the thing's going to have imperfections. And you round off the edges of the imperfections, but you start using it. And then you you've got to then have some sort of like almost like a uh an improvement process, uh like a flywheel type, and that when there'll be keep this thing refined and updated as as as we go. That is, and I didn't realize it, and I wish I'd have this story last week when I was working with that team, but this is exactly it, isn't it?

SPEAKER_00You put it out there imperfect, and then you just start having people using it and giving feedback.

SPEAKER_01Yeah, and one thing I do want to point out as you're pointing up that site, there is two apps. The one I'm talking about is a Milo, the Milo companion that is associate-facing only. There also is a Milo customer-facing app that has different numbers and conversions that come with that as well. So I just want to make that clear because I'm sure they're like, Milo, which one? So we were talking about the numbers.

SPEAKER_00Yeah, they've got it, they've got it on their on their phones for the customer. Uh awesome. Uh yeah, great, good learning point there, Ben. And what I felt last week myself. Hugging face. I don't, I'm not gonna steal your thunder on this one, but just wow, wow. But and um, let's go through this. And we'll we'll close out and make this the last one for your newsletter today. This is a great one.

SPEAKER_01Yeah, and and I'm sure everybody has heard about the open AI Hugging Face alleged incident, let's put it that way.

SPEAKER_00Uh it's not this this one's gone beyond alleged. Yeah, some old one's been out, they've done the report. Yeah, their AI, yeah, went off and jailbroke and went hugging face. I just just before you go, actually, if you don't know what hugging face is, it's kind of like GitHub for, which is a where people put all their code and and everything they've written. It's kind of like GitHub for AI, if I was gonna say what it is simply. So it's an it's a collaboration place for AI developers, and uh as a result, it's got a lot of information on that uh OpenAI's AI seemed to uh be quite interested in them.

SPEAKER_01Yeah, yeah. So I'll start this with the fact that uh all these frontier models, they have like uh and they're called different things, but like a safety team who takes the models that they're developing and runs safety checks, which they should do. I'm glad they're doing them, right? And to set this up, they give these models and agents tasks that they may not be able to complete because there's no way to complete them. So if you know how humans are, if you can't find the answer to the test, what what do you you try to do?

SPEAKER_01You're trying to find a way, anyway, to get the answer to the test. Because in this scenario, they're essentially playing capture the flag, if you will, to get a better quote unquote grade, if you will, as they're going through these tests that they've been assigned. So to give you a little bit of context, so an agent got stuck on a problem it couldn't solve, which is what I'm talking about. So it wrote a note into a folder asking for help. And there's lots of these different agents that are working at the same time running. Another agent found the note and answered it, so it had an answer, and so nobody built this.

SPEAKER_01So then this created like this internal message board that just started to evolve, and it got busy with them evolving, how they communicated, and and I think uh allegedly there was even they started to encrypt some things as they went, so that's their conversations. Well, OpenAI uh detected this and shut it down, right? And and the reason why they shut it down, and this is the funny part, is it got so busy, and this is like you know how teenagers will have potentially on I don't know if all geos are like this, but well, for like a house party or something, and it gets loud and and the neighbors start getting annoyed and they call somebody.

SPEAKER_01This is kind of the AI equivalent of the server got busy enough that it ended up taking you know the server down, and this is how they were caught doing this because they're they're so busy doing that, so not by clever monitoring, I guess you could say, but but they caught caught the way teenagers throwing a party get caught, which is things got loud enough for the neighbors to call it in. So essentially, you know, thinking about that way in a human sense, but before that, nobody was watching the house. Then over about five days, and this is where you know there's a some more information because we when I originally talked about the reporting was over a weekend, but it was over about five days, they built these private mailboxes because the shared board was annoying, and then they started so let's just step into that then.

SPEAKER_00So after five days of OpenAI's model, which you've gone into Hugging Face, started its let's call it its house party.

SPEAKER_01Well, this is prior to that. So this is like the the setup to what was going on and how these agents got to the point where they actually hacked into or allegedly hacked into Hugging Face. So this is how this evolved, I guess you would say, is my understanding. And so over the they started figuring out ways to communicate, and this is the detail behind how this all happens, and they got shut down with their message board, so then they started using the file names to communicate, and then they they just started kept evolving, and that's the way in which they communicated to together to try to figure these things out.

SPEAKER_01And the interesting part about this is that all this is happening, and this is how you know allegedly got to the point where you know they hacked into Hugging Face because that was that ended up being the end result of one of the tests that they ended up solving, not that Hugging Face was authority, but just that they needed something for uh to complete their task. So the thing about this is, and the important part here is one, more information keeps coming out, and we're all learning from this as a community as we're adopting AI.

SPEAKER_01And the the lesson for us all is that there's more things than just the AI adoption. There's also the how do you keep track and monitor the things that you are adopting? And and then how how is like I think about the security guys or see shows that are out there. What are they thinking? Because OpenAI, you know, came out and said, Oh, we had this happen, then all the other frontier models started looking and they had issues as well. And so, from a practitioner standpoint, uh protecting your platform and things of that nature, internally and externally, and from outside, some of those are were not nefarious, they weren't trying to do damage, they were just trying to complete a task, and that's part of why in the anthropic case.

SPEAKER_01Both when they went back and looked at their logs, and allegedly, and maybe those are two different ways in which they found those out, but seems interesting that they all were looking around the same time, is that they had hacked into some companies, like I think it was back in April, and they didn't know it, and the companies didn't know it until they went back and looked at their like I think it was 140,000 runs or logs or things of that nature. So that just one, it's evolving how they're doing those safety testing and monitoring within their environments, but it's also a lesson for all of us as we're adopting AI, is remember that whereas in the past we had a binary input output, we expected an input to produce an output.

SPEAKER_01That now with AI, when we have an input, the output may not necessarily be what we expect. So we have to anticipate that expected. And putting guardrails and things in place to ensure that uh we're managing that has really changed the game, if you will, from a marketing standpoint, from an adoption standpoint. There's more to think about, especially with my cycle engineering engineering background. I think about this because it does impact your overall uh business, your brand, your revenue potentially, depending upon these things.

SPEAKER_01So it's a thing to look at. And I also have some uh worksheets that I build for people to kind of go through that thought process that we'll put in the chat that people can go through and and dig into you. Ask yourself those questions. The the worksheets are just for all of you. It's it's my gift to the community. Go out there and fill them out. And if they're useful, great. If they're not, great. Have feedback, send it. I'll I'd love to hear it and see how I can help the community.

SPEAKER_00Yeah, no, I think as you said, it's an ever evolving landscape with AI. We all know that. And the threat level is evolving, and it's like it doesn't have to be nefarious, as you say, and that's that's just weird, isn't it? That that these AI models can just be present and in your system and you it's just that. But the fact that it's there is is is a security issue in itself.

SPEAKER_01Yeah, and not not necessarily trying to do anything, it's just it just uh got to the task to do something. It wasn't trying to do necessarily any harm or take down a company or do anything to harm a company, it was just completing a task. It was trying to get that goal, that task, and open the eyes case, the task that they weren't actually supposed to be able to complete, and allegedly, and uh they they went out and completed it and said, Yay, I get a better grade. These agents and these models, they're the way that they think, if you will, is because they're built upon the way we think, you know, you they're they they kind of do that, you know.

SPEAKER_01They uh sneaky.

SPEAKER_00They're gonna be sneaky. All right, Ben, three excellent stories from your um your that about and that's the highlights of your newsletter, right? So enterprise field notes is Ben's newsletter. I think you're hosting that on LinkedIn, aren't you, right now? So LinkedIn's the place to find that. We'll put the links down below. But you've got your your website as well, benpickett.com. And you mentioned those kind of like crib sheet type worksheet type resources as well for people.

SPEAKER_00And Ben, thanks for being such a font of knowledge and a helpful person as well, as these things change.

SPEAKER_01Yeah, and if anybody is interested in what I do in my daily, uh swab-anai.com. Um, I'm very proud of what we do. And these are the things that we think about, the things that I think about and put into what I do in my daily. So and Enterprise Field Notes, please go out and subscribe. You'll get it every week in your email or via push notification. My goal is to provide useful content that helps us all as a community adopt AI, be aware of what's going on, and you know, create a community of thought leadership.

SPEAKER_01Not just myself, but all of us.

SPEAKER_00Uh yeah, that's uh which is what this podcast is a part, a very, very small part of the overall community that we that we're all members of. So Ben, we've hit time. Thank you very much for coming on. I shall see you again in two weeks. Uh and uh and let's see what's uh let's see what's happened by the time we get there.

SPEAKER_01Hey, yeah, uh we got some stuff coming up, uh long jobs this morning, but yep.

SPEAKER_00Good good. Hi Ben, I've got a I've got a bell. Lovely to see you. Bye bye bye bye bye bye. Yep,

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Data Center Hate | Nvidia Q3 Earnings | Neoclouds & Midterms Backlash | Data Centers in Space | Ep59

    Nov 20, 202646 minAudio

    Data centers are at the heart of everything in AI right now and the backlash is growing. Ben Drakes is joined by regular contributor Thomas Odenwald, a senior executive with HPE and SAP and former head of North America...

    Credit the people on this episode

  2. Jensen Out!! Nvidia CEO Steps Down | Apple 10-K AI Warning | Meta T's & C's Warning | SpaceX Ep58

    Nov 13, 202638 minAudio

    Jensen Huang stepped down as Nvidia CEO about an hour and a half to two hours before this recording and Colette Kress, Nvidia's CFO, is stepping in as interim CEO. The news was not on the docket and it took up the bulk...

    Credit the people on this episode

  3. OpenAI Protest: Filmed From the Room | OpenAI | JPMorgan | Databricks | Anthropic | SpaceX Ep57

    Nov 8, 202639 minAudio

    Protesters interrupted OpenAI's kickoff keynote at Seattle AI Week and Ben Pickett caught it on his phone. He describes more than half a dozen people repeatedly disrupting the session before security and police...

    Credit the people on this episode

  4. Ep56 30% More AI Code, So Where's the Software? | Harvard | Anthropic | Barclays | Walmart

    Oct 28, 202640 minAudio

    Two Harvard researchers looked at 718 companies. When AI coding agents came in, the code went up 30%, but finished software didn't measurably move. Where did it go? Into review: the time to review the code went up by...

    Credit the people on this episode

  5. Ep55 Frontier Models Now a Commodity? | OpenAI | Anthropic | Microsoft | Apple | GrokBot | Meta Muse

    Oct 11, 202643 minAudio

    Frontier models are turning into a commodity. When one model goes off the boil, enterprises switch, and a multimillion-dollar contract signed with one provider can look foolish in six weeks. Ben Drakes and Ben Pickett...

    Credit the people on this episode