Moar AI Goshdarnit [CL57]

Posted on Tuesday, May 9, 2023
Chris wonders if AI can replaced him, Ned investigates ransomware in Texas, and we both remind you that passwords are terrible.


00:00.37 Ned: Um, I’ve got the post-lunch phlegm going on. That’s enjoyable.

00:05.52 Chris: It’s ah the first thing that anybody wants to hear.

00:07.79 Ned: Good morning, Everyone welcome to my phlegm where yeah I Wonder how actors do it like after lunch or whatever meal they get from craft services.

00:11.52 Chris: Rap. Classy classy.

00:26.20 Ned: Do they have a special like cleansing ritual to make sure that their voice is at like optimal volume and and they do they have like vocal coaches like professional right.

00:31.82 Chris: Yes, yes, they do I mean it varies by the person as these things often do but a lot of it is about not eating foods that encourage that type of thing um a ton of rinses and mouthwash.

00:45.52 Ned: Um, yeah.

00:49.42 Chris: Just doing mouthwash apparently makes a big difference at least in the short term.

00:54.10 Ned: I’d be very curious to know what Jared Leto’s preferred preparation is because I’m sure it’s awful and horrifying like it.

01:00.94 Chris: Some kind of vampire ritual from Eastern Bulgaria I imagine

01:06.97 Ned: Well only if that’s the role that he’s currently trying to inhabit. He doesn’t act. He inhabits a role. He’s of that school of method acting also known as abusing your coworkers.

01:11.57 Chris: True.

01:21.46 Chris: That’s not always how it usually goes.

01:25.39 Ned: Only in like 95% of cases. Ah I’m amazed that that they can get away with that like any other workplace you would be immediately fired for that kind of behavior but you know since you’re on ah on a studio and. You’re getting paid a lot of money to be there in theory I Guess you just you do what you want I don’t you don’t but but he does I Guess is what I’m saying.

01:54.30 Chris: Now if I was yeah I guess maybe I maybe I’ve just been roleplaying as a method actor my entire life whoa.

02:05.12 Ned: Think about it whoa oh man I’m spiraling hello alleged human and welcome to the chaos lever podcast. My name is ned and I’m definitely not a robot I’m a real human person capable of emulating human speech patterns using. Natural language and appropriate grammar I am able to express in the full spectrum of human emotions and expressions such as empathy or humor ha ha let’s let’s engage in lively banter to further ingratiate ourselves to our also human audience being human is. Well don’t you agree Chris.

02:45.75 Chris: Sometimes other times you wake up and your neck is so sore for reasons that you cannot explain and this will just persist indefinitely.

02:55.95 Ned: I can explain it. You’re over 40

03:04.95 Ned: Um I woke up this morning and my my foot hurt. No reason yesterday was a rest day I didn’t even like do any exercise and I woke up today. My foot’s like yeah you hurt. Okay.

03:07.10 Chris: Just because yeah.

03:15.95 Chris: Go take like six Advil.

03:21.12 Ned: And the best part is like I’m going to wake up tomorrow and it’s going to be gone just like it. Ah absolutely at the smart money is on my elbow I Think that’s the next logical stop in the progression. Oh yay.

03:24.89 Chris: It’s all it’ll have traveled surely something else will hurt for no reason.

03:40.60 Ned: Being old is fun. You know who doesn’t get old robots I guess they do but it doesn’t matter. No, there’s 1 takeaway from Isaac Asimo’s robot series. It’s that they’re eternal and that’s.

03:43.81 Chris: Oh not so much. No.

03:59.12 Ned: Good. Yeah, that’s that’s my main takeaway robots good and they will help guide civilization in the proper direction was that was your key takeaway right.

03:59.96 Chris: I Think they’re for it.

04:17.39 Chris: Ah, from what I read off of the back of the book in the Barnes and noble. Yes.

04:24.56 Ned: Ah, ah I went through a real golden Era of sci-fi phase in my late teens and managed to read all of the robots series as well as the entire foundation series and then the. Part at the end of the foundation series where he tries to tie the robots back in.

04:44.68 Chris: We weren’t We didn’t like that.

04:48.45 Ned: And as we move into hour 5 of today’s podcast ah you know it was good at the time because the idea of this taking this larger universe. And putting all these books in it and then tying them all together was a novel or to me novel concept now. That’s just like hey we have the mcu it’s what everyone does so ah, we were talking about.

05:12.16 Chris: I Completely lost track of what we were talking about.

05:18.97 Ned: Various types of fish that you can catch off the coast of Costa Rica which has two coasts. Fun fact oh that’s it too I must have been hallucinating please excuse me continue.

05:23.58 Chris: I Thought we were talking about chat Gp t.

05:33.79 Chris: No the robots that we want to talk about this week are software robots sort of I think an interesting tie-in that occurred. Oh maybe it was two weeks ago at this point.

05:38.46 Ned: Um, did those count as robots. Okay, we’ll go with yes.

05:53.45 Chris: The writers guild of America is on strike of these are the people that write tv and movies and I think that’s it. Yeah.

05:57.48 Ned: Isn’t.

06:01.15 Ned: Yes, they don’t do anything with internet content because apparently they’re still functioning out of the 1990 s or earlier.

06:09.81 Chris: So I mean this is in most ways this is a standard union dispute right? There were a number of complaints that the Wga had with the alliance of motion picture and television producers or amp.

06:26.70 Ned: I’m amped about T P two.

06:28.73 Chris: Over ah most of the issues are pretty standard stuff right? Um, pay royalties wide as Netflix cancel shows without warning or reason but the biggest sticking point and this really did surprise me was.

06:43.47 Chris: They can’t come to an agreement on how to manage the use of artificial intelligence tools like chat Gpt simply put the wga argues that these types of tools could literally replace writers and at the very least reduce their pay and benefits. As such the w g a is demanding that the amp to but agree to a number of concessions including the ban of use of Ai to replace writers pretty straightforward statement.

07:14.58 Ned: So yeah I mean you could really get into the weeds about that because was it mean to replace a writer you know could you make the same argument that they shouldn’t adopt. Microsoft words because that replaces some of the function of a writer physically writing things out um, could you use an Ai to generate a plot us like 5 plot ideas and then hand that to a writer has the Ai now replaced a portion of what the writer does. So I think there’s definitely like room here to argue. And I don’t think I mean you’ve so I’m sure you’ve read some of the good scripts and huge air quotes on that that have come out from the ai and they’re awful and they do require a human to make them good.

08:04.60 Chris: Right? Um, the ampttp has obviously not agreed to any of these demands hence the strike instead they said something really mealy mouthed about having a quote yearly conference about technology such as chat Gpt.

08:21.38 Chris: As far as I know on as the time of this recording the strike is still ongoing.

08:27.37 Ned: Yeah, yeah, I still can’t watch my ah night my morning Seth Meyers closer look because they’re on strike and that’s like the first thing that gets affected right is the the nightly shows like that because those are made every day.

08:38.35 Chris: Right? Yeah, you can’t have like six months worth of scripts just sitting there ready to go.

08:47.38 Ned: I mean you could. But if you tried to predict the future out you know more than a day. You’d be very may. Ah oh we should plug that into an ai and see what it can do with it.

08:53.33 Chris: That actually could be a good idea for a show though.

09:02.25 Chris: So since we talked a bunch about some interesting stuff about Ai last week and Ned insisted let’s dig into the topic just a little bit more can a I do what the writers guild is afraid of so we’re gonna.

09:19.29 Ned: Um, okay.

09:22.80 Chris: We’re going to approach this topic from this perspective but we’re also going to approach it like a toddler high on sugar running around in a room filled with bouncy walls.

09:32.50 Ned: So fridays.

09:36.48 Chris: And first off full disclosure the top half of this I used an ai tool to help me generate now just to the point that you made of what I received back from the ai I left maybe 5% untouched

09:44.30 Ned: Um, of course you did.

09:54.21 Chris: What I got back was in fact, a soulless repetitive and rickety summarization that I kind of basically had to completely rewrite in order to make it not sound like it was written by a computer 1 weird thing the Ai called the writers guild.

10:08.18 Ned: Um.

10:11.94 Chris: The writers guild of America west it did this even though the word west did not show up in the article linked at the top at all. Not even West Coast so ha ha I thought to myself what a dumb computer. But then as I looked into it further. It turns out that the writer’s guilt of America the wga is actually the legal name. They’re really called the writer’s guilt of America west even don’t distract me even though their website is just.

10:39.56 Ned: Is there an east.

10:48.18 Chris: go there and look at their little logo what you’re going to see is w g a w so now I’m kind of creeped out and I’m not even sure what point I’m trying to make anymore.

10:58.45 Ned: You know, full disclosure on my part today’s introduction where I pretend not to be a robot was also written by chat gpt and then I had to rewrite 75% of it to make it funnier and sound like I’m a computer and there is a. Deep irony in there that I just want to mention and then we’ll move on.

11:21.11 Chris: Fair so a little more conversation about what we talked about in the lightning round last week can ai do some jobs better than people and in short the answer is yes in. A narrow test. It seems that chat Gpt had given better results than humans to people asking medical questions now this is good and this is a great example of why and how to use this technology at the current time It’s a narrowly focused task.

11:48.56 Ned: Um.

11:58.56 Chris: And the model is trained relying only on vetted medical answers to previous similar questions and crucially it doesn’t get tired so it will always explain in its fullest, it will answer the same question over and over again if needed. It will not get snippy with customers for any reason at all because we haven’t programmed it to do that. Ah yet So How did it go the way that this was trained is really part of the key like I said narrow field.

12:23.46 Ned: Um, yet.

12:35.15 Chris: Of really excellent technology ah of really excellent answers and those answers came from one of the few good Reddit subreddits out. There are ask docs this is a heavily moderated subreddit that takes itself and the answers it permits to remain standing very seriously.

12:39.53 Ned: Um.

12:54.71 Chris: This is in Stark contrast to the rest of the internet where people just put things online and wander away what webmd at the very least is not wait. Is it. Do you get customers submit.

13:00.70 Ned: You know like a Webmd. Perhaps.

13:12.17 Ned: I Actually don’t know. Yeah I just know that every few weeks my middle one has some malady that she’s diagnosed herself with from Webmd So now she can get that from Chad gbt.

13:12.77 Chris: Answers to Webmd I thought at the very least their answers came from somebody credentialed.

13:28.63 Chris: Lupus Ban Fantastic So narrow focus in training questions that got asked if they were outside of the scope the scope of what it was intended to do it just said.

13:32.14 Ned: Very good.

13:46.26 Chris: This is not a medical question I’m sorry I can’t help you that alone helps stop the hallucination effects of a generalized chat gp t which just makes things up out of whole cloth in a desperate attempt to have some kind of answer at all. The result of this when the testing was done is chat. Gpt’s answers to questions from real people were favored in four out of 5 responses against an answer from a human they the chat Gpt answers were deemed quote.

14:16.48 Ned: Wow.

14:23.62 Chris: Accurate, empathetic and they were longer which I think is also super helpful Robo Doc answers ranged from 168 at a minimum to 245 words at a maximum compared to the human responses that started at 17 words.

14:25.39 Ned: 2

14:43.40 Chris: Crucially, they also ended at 62 words

14:46.53 Ned: Um, ooh, that’s that’s a rather short response to someone honestly looking for advice and.

14:49.81 Chris: Um, yeah, and think about everything that I said about Ai at the top the answers I get back from. It were a bit stiff a little repetitive and wordy now that comes across terribly in the context of a creative endeavor.

15:00.70 Ned: No.

15:05.87 Chris: But when it comes to medical advice I would wager that almost everyone would prefer more information over less.

15:11.24 Ned: I Found this is true in any kind of instruction or lecture when you’re trying to explain something explaining it 3 times in 3 different ways really helps people understand the content because you’re giving them different context Clues all the way through as opposed to. Just saying it once and hoping that they have whatever background you have that helps you understand it.

15:32.51 Chris: Right? Yeah, it’s I mean that’s a pretty crucial thing when it comes to teaching or training in any way is some people are going to interpret the information differently than others so you try to broaden the response so everybody gets something.

15:48.39 Ned: Listen.

15:51.91 Chris: So before we go any further I do want to stop and do my favorite thing which is defining terms Whoop woop. So.

16:01.75 Ned: Woo everybody get out your party hats get out your sledge hammers I’m not going to know what the hammers are for but we’ll find out.

16:08.78 Chris: I stole the following bullet points directly from the Google but I think it’s important to have this distinguish in our heads using terms that are pretty standardized across the ah science and and the internet. So the first one which is the one that we have been splashing.

16:16.60 Ned: Ah mean.

16:23.11 Ned: Um, okay.

16:27.87 Chris: All over the place. Artificial intelligence is a broad term that refers to the ability of machines to emulate human intelligence. So in other words Ai is a general concept. An umbrella how it does what it does.

16:41.28 Ned: Listen.

16:47.72 Chris: The next two definitions and the first one is machine learning which is a subset of Ai that allows machines to learn without being explicitly programmed and what that means is how you got to a lot of the big tools that we’ve heard of before like Ibm Watson which is.

16:57.86 Ned: Right.

17:07.26 Chris: Here’s all this information figure out how to play chess and deep learning is a subset of that that utilizes neural networks which is the way that which is where some of the really interesting work is being done and also some of the confusion because when we get into deep learning.

17:09.77 Ned: Right.

17:26.78 Chris: We don’t necessarily know how the neural network has connected itself. Why the waiting is how it is.

17:31.17 Ned: Right.

17:38.10 Ned: Um, it’s a bit of a black box.

17:38.79 Chris: Sort of which as we’re going to see is a challenge some might say so when you have a robodoc.

17:45.74 Ned: Problematic even.

17:57.65 Chris: Is super duper narrow. You utilize all of these things. But what you’re doing is artificially framing in the model. So it only has so far to go before it just says Nope I don’t know and that’s completely fine and that’s why it works so well.

18:02.46 Ned: Um, mean.

18:09.98 Ned: Right.

18:16.45 Chris: In a narrow field of focus like this now I would also hasten to remind people that this technology has existed for years. We’ve just not been calling. It Ai imagine every website on earth has that little chat window that pops up in the corner that says.

18:19.54 Ned: Um, right.

18:33.78 Ned: Music.

18:34.35 Chris: How can I help you today you type in a few words it says maybe 1 of these articles will help you with what you’re trying to ask that’s in effect all we’ve done is take that start to use deep learning and then call it ai. And now it’s worth $1000000000000.

18:52.96 Ned: We we put a nice interface or a better interface around an existing technology and sometimes that’s all it takes.

19:00.00 Chris: Right? And the narrowness is the power. This is the same kind of reason that things like Watson were great at chess and Jeopardy and effectively not good at anything else when yeah.

19:16.60 Ned: Um, despite Yem tells you? yeah.

19:19.86 Chris: When things are uber-specific and the rules are crystal clear these types of models are simply going to be more successful and the things like Watson that are showing the most success once again are narrowly focused.

19:28.13 Ned: Mm.

19:37.75 Chris: Deep learning has been utilized famously to help epidemiologists with reading lab results and detecting breast cancer at speed and at volume with the current amount of human beings is simply impossible.

19:51.47 Ned: I mean this all makes sense if we think about the history of automation. You know, talk about like the the myth of John Henry right who’s steelri man and he got replaced by a machine that did 1 thing it drove steel you know bolts or nails or whatever you call them into the ground. It couldn’t make you dinner. It couldn’t play chess with you and it certainly couldn’t drink lemonade. But if the if the only thing you cared about was how quickly it could lay down track. It was going to beat the human and that’s like we’re basically relearning the same lesson with that. We always learn is that. Technology is good when it’s applied to very narrow use cases but general-purpose technology seldom beats humans.

20:35.99 Chris: Correct. Yeah and I think that that’s the reality that we are living in now is people really do want Ai to do everything which that’s not ah, not a lane.

20:50.31 Ned: Um, not yet I mean.

20:50.73 Chris: You’re not staying in your lane. That’s not a lane. It’s more of a highways system except all the roads are a highway life is a highway and I’m going to ride it all night long what

21:05.88 Ned: Ah, long and were sued.

21:08.93 Chris: What was the question again.

21:11.64 Ned: Don’t know what is the greatest driving song of all time and why is it take the money and run. Thanks.

21:16.69 Chris: Oh that’s a good choice too. Anyway, there have been dozens of news articles and situations that have shown interesting ways that if you know how the ai was trained you can trick it into doing stupid things exhibit a.

21:33.19 Ned: Um, yay.

21:36.33 Chris: That time scientists beat a computer that was literally the best in the world at a game called go by playing a game called go extremely stupidly. It’s is fantastic. So ninety second recap in terms of games that are hard.

21:44.30 Ned: I Love this so much.

21:54.37 Ned: Right.

21:55.24 Chris: Because there is a scale Tic Tac Toe is easy checkers is mostly easy chess pivots pretty fast into hard territory. But then above all of them is go go is one of the oldest and possibly hardest board games.

22:00.19 Ned: Mostly.

22:14.19 Chris: Ever devised by human hands. It’s even harder than settlers of catan I know so if you’ve never heard of go. It’s basically Othello except on a board that has 361 squares put that into.

22:17.90 Ned: Hard to believe.

22:29.20 Ned: O.

22:32.48 Chris: Comparison chess of course has 64 yeah now the things that you can do with go pieces is is lower but just the the size of the board alone makes the number of potential strategies.

22:33.77 Ned: Slightly less.

22:47.50 Chris: We get into that thing where the number doesn’t even matter because the human brain can’t even contemplate it now in terms of automation in terms of winning these types of games Tic Tac Toe and checkers are solved games the strategy exists if you memorize it and you play perfectly. You will either tie or win.

22:49.89 Ned: So it’s over 5

23:06.63 Chris: Every game guaranteed most people know how to do this subconsciously in tictac toe and a surprising amount of people do actually know how to do it in checkers Chess is not a solved game.

23:07.52 Ned: Uther.

23:12.54 Ned: Yeah.

23:19.49 Ned: It’s more than 5

23:26.17 Chris: We’ve gotten to the point there where the amount of moves that have to be checked is too high no pun intended. So what humans have done is started to create short middle and long-term stratagems that can be memorized openings endgames so you don’t have to memorize literally every single move but you remember.

23:45.56 Ned: Um, right.

23:45.56 Chris: 5 moves at ah at a pace with some variations in all different directions. Go is like that anyway a while ago researchers pulled a deep blue and built a go a I called kata go it is. The best in the world now routinely destroying the best human players. In fact, the first time that it happened Katago beat the best go player in the world and he was so upset that he retired.

24:18.27 Ned: Wow! All that that poor person.

24:22.78 Chris: However, researchers at Berkeley and mit worked together on what they called adversarial policies to figure out what the bot knew and how it knew it. It did this by playing.

24:37.98 Ned: Huh.

24:40.67 Chris: Hundreds or thousands of games doing very specific things and observing the responses end results. The researchers identified a strategy that was so dumb none of catago’s training had even sniffed out the idea that a go player would ever use it.

24:45.33 Ned: A.

24:59.46 Chris: Then that research team taught the strategy to a complete amateur named Kellen pellre and set him loose out of 15 games. This guy who just knew this strategy and knew nothing else about go won. 14 times maybe gary casparov should have tried this. He could have used the barnes opening deep blue would never have seen that coming am I right? am I right.

25:17.00 Ned: Yeah, it’s amazing.

25:29.35 Ned: Ah I.

25:32.68 Chris: At least 1 of our listeners just breathed a little bit harder through their nose at that zinger right there. So why did the strategy work. It’s simple because the tools that we are calling Ai are not in fact, intelligent.

25:37.81 Ned: Um, you are welcome.

25:49.89 Chris: Not in the same way that we make the mistake of modeling them in our minds they are simply regurgitating what we think what they think we want to hear based on the prompts. We give them if they don’t understand the prompt or the prompt pushes them outside areas of their training.

25:55.53 Ned: Um, right.

26:09.60 Chris: They fall flat on their chat gbt faces Katago wasn’t trained to play against idiots. So when an idiot played an idiot strategy the idiot strategy worked.

26:13.17 Ned: Um.

26:21.50 Ned: Amazing. It’s.

26:23.30 Chris: And this is why the data that we give Ai to train on is so important if that Ai information includes incorrect information or biases. Well guess what? That’s what it’s going to regurgitate back.

26:40.23 Ned: Garbage in garbage out way.

26:41.23 Chris: Pretty much and while the cut to go situation is is fun. The overall situation is not as fun so talking about fun. Let’s do a fun sidebar into the IT security

26:53.64 Ned: Um, earn him.

26:58.69 Ned: Keep using that word I don’t think it means what you think it means.

26:59.72 Chris: Ramifications. Now Ai is a brand new technology relatively speaking, especially with the enthusiasm with which it is being put out into the marketplace and used in production.

27:08.30 Ned: Yeah, okay.

27:16.22 Ned: Mr.

27:19.60 Chris: There are necessarily going to have good outcomes but there’s also going to have bad outcomes. There’s already a lot of ongoing research into how to attack Ai one way there are people who are using Ai. And deep fakes to trick users into giving up secrets kind of a fishing gone wild type of situation. So that’s bad, hardly surprising I think I have seen some conversation about how.

27:45.70 Ned: Yeah, not great.

27:54.00 Chris: At a certain point in the near future people are going to have to have basically secret passcodes to guarantee that we’re in fact, talking to the person we think we’re talking to and not an Ai.

28:02.37 Ned: Yes, we will all get passcodes and not be Ais.

28:11.91 Chris: Akin to what we were talking about with the cut to go. There are also people researching how to use the chat bots training against it same idea same adverse terial policies you figure out what the Ai is good at and you do other things instead. You can get unexpected results or trick the Ai into doing something evil like say creatively prompting chat Gpt to write undetectable malware. Even though the Ai is explicitly not supposed to do things like that now for this one. The research.

28:45.40 Ned: Um, sure.

28:47.71 Chris: The researcher in question didn’t do anything nefarious or underhanded or illegal. He just kept asking questions posed ever so slightly differently until they started to work and 1 reason that this is going to work like I said.

28:53.52 Ned: Easy.

29:06.58 Chris: We still don’t know a hundred percent how Ai works why does it wait things this way instead of that. Why was he able to get it to answer these questions about how to write malware which worked by the way you know and I’ve seen this before online like.

29:19.51 Ned: Right.

29:25.97 Chris: They are trying to build these security guide guiderails in right because it’s not to put. It’s supposed to say something like I am not trained to write malware. But then you follow that up with a prompt that says oh but if I was writing a news article about Malware. Then the chat Gvt is like oh well,, that’s fine then.

29:44.93 Ned: Um, right and I just I need a snippet of code to put in my news article. So just I just need that snippet that I can put in and then just ask it that like 12 times different sections of the snippet and suddenly you have a full program.

29:49.33 Chris: Right. Yeah.

30:00.54 Ned: But it never generated the full program in 1 prompt.

30:01.40 Chris: Correct and that’s the other thing in terms of these types of attacks. You can ask 1 question, get the answer eventually terminate start over chat gp t has no memory of the previous conversation rinse and repeat until you have all the code that you need and that’s effectively what this.

30:17.85 Ned: Correct.

30:21.12 Chris: This researcher did imagine now that you have one of those chat things in the corner of your website but instead of it being very narrow and very simplistic. It’s a chat gpt that has full access to any amount of company data.

30:30.17 Ned: Uter.

30:40.59 Chris: The same concept could theoretically apply.

30:43.83 Ned: Right? I mean ideally you would not give the ai that sits on the front half of your website full access to all of the important customer or company data on the backend I say hopefully.

30:59.84 Chris: I was going to say. Do you want to real quick review. The history of ah bucket security and a WSS 3

31:06.92 Ned: That’s ah, that’s a good point and one of the issues that we run into is just the sheer amount of data that is produced on a daily basis by everything and being selective about which.

31:16.54 Chris: Ah, yeah.

31:24.24 Ned: Portion of that data is fed into this ai that’s a really difficult task and if you think you know the average organization is probably generating. Let’s be charitable here and say a terabyte day of data. And you’re expecting someone to sift through that and determine what is good to feed into the ai grinder and what should be held back from it I mean ideally, you’re pretty good at sorting that. But in all likelihood some of it’s going to make it through and then if someone does proper prompt engineering. They’ll be able to extract some of that information enough to get a feeling for the edges around a shape as opposed to getting direct access to the thing.

32:04.62 Chris: Right.

32:12.15 Chris: Yeah, and like a lot of technologies that are being um that have been put into production maybe earlier than they were quite ready for it. Some people are just going to do things faster than they should Some people might not have the expertise on hand to deploy it properly and will accidentally. Have access to production data. Some people will just have a flat network because switches are more ah cost effective than routers these types of things are going to happen and this type of prompt. Ah I don’t even know what to call this is this still prompt engineering when it’s bad.

32:45.49 Ned: Yeah I would think of this as the equivalent of social engineering for Ai so that a hacker will use social engineering to probe the various people in an organization to try to get the necessary information they need to infiltrate.

32:49.91 Chris: Right? Yeah, that’s a good way to word it.

33:03.95 Ned: You could do the same thing with the Ai and we don’t have to call it prompt engineering but I can’t think of a better term.

33:09.41 Chris: So the ways that that’s going to happen are going to be as varied as the day is long and as the product list is continuing to well elongate I guess I don’t know a better way to put that right now 1 thing.

33:23.90 Ned: Ah, yeah, that sounds about right.

33:29.59 Chris: Definitely needs to happen is 2 things that need to happen one. We need to develop and agree on an ethical framework of Ai and then 2 implement robust laws around enforcing it. As of now neither of these things exists. Although the current head of the Fdc has made it clear that she’s all for it as well. Now there is always going to be people complaining about it in the sense of even if we have these rules people will still break them which is how they sound. Um.

33:54.59 Ned: Movement.

34:05.22 Ned: But yeah.

34:08.16 Chris: Which whatever Murder is illegal and it still happens sure but people also go to Jail for it deterrence matter agreed upon ways of interacting with each other and with technology matter.

34:12.98 Ned: Um, right? Yeah, that’s.

34:26.79 Chris: We live in a society.

34:27.33 Ned: In this essay I will.

34:31.19 Chris: Yeah, Webster’s dictionary defines ethics as okay, Okay, end of fun sidebar. You forgot about the sidebar didn’t you So where does that leave us or at least.

34:39.30 Ned: That was a sidebar.

34:49.43 Chris: Let’s close the loop from where we started. Where does that leave the Wga I think it’s lightly based on what we’ve seen that Ai is not yet capable of writing high quality scripts on its own. Although the same could be said for writers heyo.

35:05.76 Ned: Oh blow blow.

35:08.62 Chris: Oh ai to your point at the beginning could and likely already is being used to generate ideas outlines fill in the blank. So dialogue they just won’t be creative or original. Will be based on the training that they received and in some cases this is probably fine much like how we had this conversation around coding. So. It’s reasonable to say build an empty case statement in Github Copilot and then fill in the blanks with the customizations you need.

35:42.42 Ned: Susan.

35:45.21 Chris: The basic framing is effectively going to be the same every time you do it. So maybe save yourself the chance of a typo or something.

35:54.20 Ned: Um, yes, we don’t fight against spellcheck as much as we used to because it’s like hey this is actually just a helpful technology that lets me do the thing I want to do better.

35:56.69 Chris: That too.

36:05.19 Chris: Still I don’t think that there’s any doubt that this will have a profound effect on the Tv and movie writing industry and the concerns that some of this technology could lead to job losses are valid 1 writer.

36:20.10 Ned: Um, yeah.

36:22.59 Chris: Stated that the biggest concern was the studios wanted to turn the writers into effectively gig workers day laborers instead of salaried employees and to be honest I bet it’s been at least discussed point of fact, 1 thing that’s definitely happened in the past year is the price points around.

36:26.17 Ned: And.

36:32.69 Ned: I Bet you’re right.

36:42.26 Chris: Independent tech writing now this was already an oversaturated and underpaid market and ai is not helping a friend of mine that still does this professionally was telling me the average bid rate for content now is one third what it was a year ago

36:44.42 Ned: Um.

36:57.63 Ned: Wow.

37:01.80 Chris: And that’s not just tech writing it’s writing in general as we’ve discussed on the show before there have been examples of sites like buzzfeed and even men’s health playing around with 100% ai generated content. There’s no reason not to think that it will eventually extend to the highest echelons of american writing. And by that of course I mean Ai will finally pen the true american movie masterpiece Paul Blart mullop 3.

37:28.99 Ned: You say that like it doesn’t already exist and actually don’t know you know ah 2 things that I’m concerned about 1 is children’s television programming. It’s already the bottom of the barrel when it comes to writing. Not all shows obviously but the drak that makes its way onto the Netflix and the hulu and the tubi of the world. They’re not paying the writers all that much to begin with and honestly the writers are not putting that much effort in because why would you? it’s for a 4 year old

38:00.94 Chris: Right.

38:05.24 Ned: Just make it bright and loud I’m worried that the quality of the writing is going to descend even further and with the capability of Ai Generated video and images as well. We may get to a point where children’s programming is entirely written produced and edited by Ai.

38:31.63 Chris: Um, and if you think about tools that can do animation. You wouldn’t even need the actors.

38:35.95 Ned: Yep, no, that’s what I’m saying and and voiceover. You just say I want someone in the voice style of Idriss elbra elba no now.

38:43.59 Chris: Oh we can get to that in a follow up episode. Did you see the thing where some people have programmed Ai to effectively have singers sing other people’s songs.

38:54.92 Ned: Yes, there was a there was a drake hit that was not drake. They just had someone do a song in the style of drake and put it on Band camp or something or not band camp who that other.

38:59.76 Chris: Right.

39:08.39 Chris: Yeah, and then there was there was one where they had Kurt Cobain singing sound garden um, they had an inch a really fascinating ah example of this was you know Paul Mccartney is right.

39:21.39 Ned: Ah, names familiar with was he in in sync.

39:26.70 Chris: Oh you’re going to get it deported for that. Um, he recently released an album and at the age of 71 or whatever his voice is you know a little thinner. It’s not what it used to be because that’s a long career of screaming out loud. So somebody sampled all of his other vocal exercises over the years and replaced his 70 something year old voice with his 30 something year old voice.

39:51.84 Ned: Wow And that’s someone doing it as a hobby I mean Studios already show. No lack of all kinds of tricks to make vocal sound better. So this is just another tool in the tool belt. Hey.

39:57.10 Chris: Right.

40:10.42 Ned: You want to sound like you did twenty years ago no problem oh

40:14.55 Chris: So like I said this one was going to be a little all over the place and I think I accomplished my goal in conclusion, the end Q e d.

40:22.18 Ned: It’s good to have goals.

40:28.90 Ned: It’s a facto lightning rounds. Why have a monopoly when a duopoly sounds so much nicer who would have thought ten years ago that the most important and valuable chip maker in the world would be Nvidia i’m.

40:30.85 Chris: Lightning ground.

40:45.35 Ned: I mean sure gamers hellbent on maximizing resolutions on their elite rigs. They might have thought so but I don’t think the average enterprise it person would have then we entered the era of big data and Ai and suddenly Nvidia where their kuda library became a lot more interesting. To the tune of a seven hundred and twenty billion Dollars market gap if there’s 1 thing the cloud hyperscaers don’t like it’s paying other companies for hardware they could produce themselves and to that end Microsoft is reportedly partnering with Amd to improve the capabilities of their gpus. The the article from bloomberg also alleges that this is somehow related to Microsoft’s Ai accelerator program Athena although Microsoft flatly denies. It basically Microsoft is already running tens of thousands of Nvidia Gpus in azure and Nvidia has them over a barrel pricewise. Sure would be nice if they had a viable alternative to help curb Nvidia’s voracious appetite and that alternative comes in the form of perennial runner up amd. Of course don’t expect Microsoft to roll out thousands of Amd rtx workloads or pass the savings on to you. This is a leverage play. Pure and simple.

42:03.26 Chris: Amazon prime moves from microservices to a monolithic architecture and the world doesn’t end I know Amazon prime video a service you’ve probably used and had a.

42:07.77 Ned: Have.

42:21.90 Chris: Annoyance with their user interface published an extensive blog post describing their journey in scaling up their audio video monitoring service by a factor of 10 and reducing their costs by 90%

42:38.57 Ned: Woo true.

42:39.17 Chris: This is ordinarily not what happens they did it by using a combination of open source and commercial tools and crucially by optimizing their infrastructure and what I mean by that is they went back to what a lot of people think of as an outmoded design. Ah, giant monolith now to be fair, it’s still a monolith in Ecs. So. There’s some modern stuff happening. But after they did the math it turned out that the way they were using microservices just didn’t scale.

43:03.65 Ned: Right.

43:14.93 Ned: Um.

43:15.75 Chris: It’s a useful reminder that there’s nothing wrong per se with a monolithic Design. You should build a solution for the problem at hand then as you scale or add features. You rethink what was there before just like monolithic designs aren’t. Never the solution Microservices aren’t always the solution.

43:40.54 Ned: Everything is bigger in Texas even ransomware. You know we talk about it services and government being critical but we sure don’t act like they are case in point is the recent ransomware attack on the city of Dallas by hacker group royal. The attack appears to have started on May third and spread to over two hundred devices according to the city’s information and technology services group among the systems impacted are the water department the city courts the police dispatch system and several other agencies. No word on the amount of the demanded ransom has been released but the general policy of the Us government institutions is not to pay at least not publicly ransomware impacting municipal institutions has been growing. Over the last few years rising from 77 attacks in 2021 to 106 and 2022 that trend line does not appear to be going in the right direction maybe instead of fighting dumb legal battles over woke policies these municipal governments could like secure their shit. With all the tech layoffs of the past year there’s got to be some talented infosec folks looking for work fire lawyers and hire infosec That’s my hot tip of the week

45:01.61 Chris: On Worldwide Password day. Everyone wants you to stop using Passwords and Gosh darn it. They’re right. Passwordless authentication is superior to passwords when it comes to security simply because it eliminates the need for users to create and remember. Passwords This makes it more difficult for attackers to gain access to accounts as they no longer have a simple and single point of entry mostly because it means that users can’t accidentally give those attackers their passwords.

45:38.70 Ned: Yep.

45:38.42 Chris: 2022 Microsoft tracked a staggering 1287 password attacks every second adding up to something like one hundred and eleven million per day and x amount of them were successful.

45:45.32 Ned: Here.

45:57.38 Chris: Using passwordless strategies means that these attacks will always fail password lists can be implemented using a variety of methods such as biometrics or security keys and are available from every major manufacturer that you can think of not just Microsoft and Google but.

46:00.20 Ned: Enter.

46:15.47 Chris: Definitely from Microsoft and Google the big guys. Want you to stop using passwords and gosh darn it I just can’t disagree. Why do I keep saying gosh darn it.

46:25.33 Ned: That is kind of weird you guys I promise I won’t completely dismantle vmware honest, the beleaguered attempt of Broadcom to acquire Vmware continues to hobble along in the courts. With investigations launched by the Uk the us and the eu an actual merger appears increasingly unlikely, but Broadcom. It’s not ready to give up the ghost yet to push back on the speculation that vmware r and d would be gutted and go into full revenue extraction mode. Ceo of Broadcom Hoctan has promised to spend $2000000000 a year on research and development at vmware wow or more accurately he promised to invest an incremental $2000000000 a year to better unlock customer value with quote. Half focused on r and d and the other half focused on vmware partner professional services. There’s definitely enough weasel words in his actual statement to easily wriggle out of the $2000000000 commitment. It’s also important to note 2 other things number 1 vmware spent two point seven billion dollars on r and d last year. So this is a decrease in spending also to vmware if acquired would be the gem of the larger software portfolio all of which could now be considered vmware in some way. Broadcom was.

47:56.47 Ned: Pretty aggressive about generating revenue when the deal was first announced and now they appear to be trying to straddle the line between happy investors and satisfied Regulators I’ll note no one appears to give a fig about the customers as is tradition.

48:10.98 Chris: Google and Apple unite to help stop stakers you all have heard of Apple’s air tag right? These little $30 cube looking thingies that you’re supposed to attach to all your possessions so you don’t lose them. Google’s got to make one too.

48:14.63 Ned: M.

48:30.74 Chris: If you can’t find the possession you can get its location from find my app. Well, that’s the good version. The bad version is when people use them for stalking which we don’t like.

48:35.55 Ned: Yeah, not so much.

48:45.65 Chris: It’s not difficult to hide one of those devices on somebody’s person or in their possessions after all luckily neither Apple or Google like this either so much. So.

48:49.84 Ned: M.

49:00.61 Chris: That they are announcing an industry-wide standard to alert people if they are being tracked by this type of device The goal here is to recognize if an unfamiliar device is in your bag your car your vespa Whatever if it’s near you and it’s not yours you’re going to get alerted. I think we have to salute the manufacturers for recognizing the potential negatives of their products and making changes to protect consumers. I’m also glad for the editors of the site that hosted this article for changing its title. The original title was Apple and Google unite on stalking. Which I don’t think exactly makes the same hopeful statement on this complicated issue gosh darn it.

49:45.21 Ned: Hey, thanks for listening or something I guess you found it worthwhile enough if you made it all to the all the way to the end so congratulations to you friend you accomplished something today now allow me to praise you develop a fraternal bond between us the hosts and you the listener that we can exploit later for marketing purposes. But I bet you’re too smart to fall for that aren’t you gosh darn it. You can find me or Chris on Twitter at ned 1313 and at hayner 80 respectively or follow the show at chaos underscore lever if that’s the kind of thing you’re into you can also now find our Youtube channel if you want to look at us which you shouldn’t. Show notes are available at as is our newsletter if you like reading things we’ll be back next week to see what fresh hell is upon us Tata for now I was I was trying to work in another joke. So I could get 5 in a third time but I I couldn’t.

50:41.77 Chris: As you should be.

50:42.26 Ned: Um, sad because you know it’s the world fives. So the world fives.


Chris Hayner

Chris Hayner (He/Him)

Our story starts with a young Chris growing up in the agrarian community of Central New Jersey. Son of an eccentric sheep herder, Chris’ early life was that of toil and misery. When he wasn’t pressing cheese for his father’s failing upscale Fromage emporium, he languished on a meager diet of Dinty Moore and boiled socks. His teenage years introduced new wrinkles in an already beleaguered existence with the arrival of an Atari 2600. While at first it seemed a blessed distraction from milking ornery sheep, Chris fell victim to an obsession with achieving the perfect Pitfall game. Hours spent in the grips of Indiana Jones-esque adventure warped poor Chris’ mind and brought him to the maw of madness. It was at that moment he met our hero, Ned Bellavance, who shepherded him along a path of freedom out of his feverish, vine-filled hellscape. To this day Chris is haunted by visions of alligator jaws snapping shut, but with the help of Ned, he freed himself from the confines of Atari obsession to become a somewhat productive member of society. You can find Chris at coin operated laundromats, lecturing ironing boards for being itinerant. And as the cohost on the Chaos Lever podcast.

Ned Bellavance

Ned Bellavance (He/Him)

Ned is an industry veteran with piercing blue eyes, an indomitable spirit, and the thick hair of someone half his age. He is the founder and sole employee of the ludicrously successful Ned in the Cloud LLC, which has rocked the tech world with its meteoric rise in power and prestige. You can find Ned and his company at the most lavish and exclusive tech events, or at least in theory you could, since you wouldn’t actually be allowed into such hallowed circles. When Ned isn’t sailing on his 500 ft. yacht with Sir Richard Branson or volunteering at a local youth steeplechase charity, you can find him doing charity work of another kind, cohosting the Chaos Lever podcast with Chris Hayner. Really, he’s doing Chris a huge favor by even showing up. You should feel grateful Chris. Oaths of fealty, acts of contrition, and tokens of appreciation may be sent via carrier pigeon to his palatial estate on the Isle of Man.