Latest / Newsradio WGAN / Tech Talk: Craig Peterson

Transcript

Tech Talk: Craig Peterson · Newsradio WGAN

• 0:00
  The moment, this is the time that Craig Peterson joins the
• 0:03
  program every week for a conversation about the world of
• 0:06
  technology. Craig Peterson, how are things
• 0:09
  this morning, Sir? Things are going pretty well,
• 0:12
  although the Russian hackers are at it again.
• 0:16
  We just got a warning here the last couple of days.
• 0:19
  Oh, tell me more about Russian hackers.
• 0:22
  I always enjoy a good Russian hacker story.
• 0:23
  That's that's you're familiar with the term of poisoning when
• 0:28
  it comes to Google, right? Yes, I'm.
• 0:30
  Not sure the audience is, but you you might want.
• 0:32
  To exactly that's that's where people will go ahead and poison
• 0:37
  the Google search results, usually by putting the same data
• 0:42
  on a lot of websites. And so Google picks it up as
• 0:45
  truth. And of course, AI can be
• 0:48
  poisoned. And that's one of the big
• 0:49
  concerns right now. In fact, in my newsletter, I was
• 0:52
  talking about how Google's bragging that they're going to
• 0:55
  use AI to help with your e-mail. And then on that same the same
• 1:00
  press release, it's got a warning.
• 1:02
  And and we're really worried that AI is going to be poisoned
• 1:06
  here. So you might lose all of your
• 1:08
  data that's in Gmail. But besides that, what the
• 1:11
  Russians are doing right now is they are poisoning Google's ads.
• 1:17
  Now a lot of people are interested in LLMS. You know,
• 1:20
  the whole AI thing. If you're a programmer, almost
• 1:24
  certainly you're using a Mac, you're not using Windows.
• 1:27
  And so a lot of the stuff they've been using to attack
• 1:29
  Windows isn't getting them what they want.
• 1:32
  So what did you've done? Now this is this is crazy and
• 1:36
  it's a lot deeper than I, I really have time to go into
• 1:39
  right now. But Google has something called
• 1:42
  AdWords, which is an advertising network.
• 1:46
  If you have a website and you want to want to get make some
• 1:49
  money from it, you can go ahead and sign up and Google will give
• 1:53
  you the ads to display on your website.
• 1:56
  And so people will pay, get paid based on you'll get paid based
• 2:00
  on how many people might have seen the ad or how many people
• 2:02
  clicked and various other things.
• 2:04
  So all of that said, what Russia is doing right now is they are
• 2:10
  putting up ads for people who are looking to download
• 2:15
  Anthropic Clod onto for a Mac. So if you click on that, because
• 2:23
  again, they know people are using Anthropic Clot is the
• 2:27
  number one tool if you want some programming support or other
• 2:30
  things, they also know that Macs are what people are using to do
• 2:35
  software development. So the ad comes up, you click on
• 2:39
  it thinking you're downloading Clawed from Anthropic, this
• 2:43
  large language model programming tool, and and guess what, you
• 2:48
  get some malware you're going to.
• 2:50
  Be not clawed, right? On your Mac.
• 2:54
  And so there, there's been warnings over the last couple of
• 2:57
  days about this because apparently a lot of people have
• 3:00
  been have been hit by it. Now I've got a real easy
• 3:04
  shortcut when it comes to Russian malware that most people
• 3:09
  aren't aware of. You know, Uncle Vlad, he doesn't
• 3:14
  like it if the Russian oligarchs get malware, right?
• 3:18
  And you know, these guys go to prison.
• 3:20
  They take take, they take hacking very seriously in
• 3:23
  Russia. And So what happens is these
• 3:26
  Russian hackers, when they write their malware, the first thing
• 3:30
  their software does is look to see if there is a Russian
• 3:35
  Cyrillic keyboard on the computer.
• 3:39
  And if there is, then it immediately exits, doesn't do
• 3:42
  anything else. So what you can do to protect
• 3:46
  yourself against most Russian malware is a little trick that
• 3:50
  works on Windows, because Russians aren't just going after
• 3:54
  Macs here. That works on Windows, that
• 3:56
  works on Macs. It works on anything.
• 3:59
  That is, install a virtual Russian keyboard on your
• 4:03
  computer. So if both in Windows and in
• 4:06
  Mac, you go to your settings, you go to keyboards, and then
• 4:10
  you just install a Russian keyboard.
• 4:13
  Now, that doesn't mean you have to go out and get a Russian
• 4:15
  keyboard and nothing on your keyboard's going to work
• 4:17
  anymore, because it's all in Russian.
• 4:19
  It's a virtual keyboard. So your computer stays the same.
• 4:23
  You have the same keyboard every, the layout's the same,
• 4:25
  everything's the same, but they're in the background
• 4:28
  waiting for you is a Russian keyboard.
• 4:30
  So what they found is when when they have dissected a lot of
• 4:35
  this malware, apparently including this one, is that the
• 4:39
  first thing to check for is a Russian keyboard that they see
• 4:41
  at the exit. And it can just be a virtual
• 4:44
  keyboard. So if you do that, you're
• 4:46
  protected against the vast majority of Russian malware
• 4:49
  right there. Simple trick.
• 4:52
  Craig Peterson giving you one simple trick to make sure that
• 4:56
  you protect yourself from malware.
• 4:57
  It's almost like it's a little listicle.
• 5:00
  Craig Peterson. I was going through my tech news
• 5:02
  of the day here and I saw a couple things that were sort of
• 5:04
  interesting to me on the AI front.
• 5:06
  One of which is that Google is now unveiling what they call
• 5:08
  Gemini Intelligence, which bundles a lot of the existing
• 5:12
  Gemini features together and task automation apps, you know,
• 5:15
  stuff like that. It got me thinking to a question
• 5:19
  I ask every once in a while to you.
• 5:21
  Maybe you can call it the state of AI address from Craig
• 5:24
  Peterson. Because, you know, one of the
• 5:26
  things I've noted in using these tools quite a bit over the last
• 5:32
  several years has been that depending on the week I use it
• 5:36
  or the month I use it, they all seem to be better at things, you
• 5:41
  know, and, and sort of rising and falling in, in their quality
• 5:46
  rather significantly and changing a lot over the course
• 5:50
  of even just a few months here and there.
• 5:52
  You know, I at one point, you know, Grok would be better than
• 5:55
  ChatGPT at this thing. And then I would do it again a
• 5:58
  month later. And now it's pretty terrible.
• 6:00
  And the other one's better. And, you know, you're starting
• 6:02
  to see that kind of thing. And I know you've been an
• 6:03
  advocate of using some of these tools that actually use multiple
• 6:07
  AI, you know, bots or whatever, across many spaces or whatnot.
• 6:10
  But can you give me your state of AI address here?
• 6:13
  Like when you're looking at these various companies, the
• 6:14
  Geminis, you know, the Anthropics, obviously the chat
• 6:17
  GPTS and, and open AI stuff and all the others, Are you starting
• 6:22
  to see long term trends develop? Are you starting to see some
• 6:26
  separation in the pack re emergence of some that had kind
• 6:30
  of fallen back, etcetera? Or what do you think?
• 6:34
  It's interesting to look at the numbers and when you're talking
• 6:36
  about adoption of AI, the number one country is the United Arab
• 6:41
  Emirates. And according to some surveys
• 6:43
  out there, it's huge. 70% of people are using it.
• 6:47
  Singapore, UAE, is it? That's just crazy.
• 6:52
  But other countries like Norway, Ireland, France, they've,
• 6:55
  they've all gotten into it, as have the US.
• 6:58
  Here's where things are going. The current state of AI.
• 7:02
  You might have seen some ads online from Google Ads, not the
• 7:06
  Russian ones that tell you that you can make money by helping to
• 7:12
  train an AI. Now, we know that Open AI, for
• 7:15
  instance, is alleged to have stolen all kinds of data from
• 7:21
  the Internet, right? So they scrape that, they scrape
• 7:23
  YouTube videos, they grab all of that.
• 7:25
  And so some of the very first AIS were pretty dumb.
• 7:29
  So what they decided to do is they were going to hire people
• 7:33
  and they did. They hired over 5000 people in
• 7:36
  sub-Saharan Africa to sit there all day long and go through and
• 7:42
  pretty much every question that people have asked and manually
• 7:47
  help to come up with answers for it.
• 7:50
  So that's what they did. And then we got ChatGPT.
• 7:54
  That's what people were using. Now these ads are calling for
• 7:59
  people. I know a couple of people that
• 8:00
  have done this, calling for people who have knowledge in
• 8:04
  specific industries and they're having them train the AI in that
• 8:10
  industry. So what'll happen is again, the,
• 8:13
  the, let's say it's a lawyer. In fact, one of my friends who's
• 8:16
  a lawyer was looking into this. He, he.
• 8:20
  So anyways, it'll come up with a legal question and the AI will
• 8:24
  give kind of its answers as to what it thinks might be the
• 8:27
  right answer. And then the looking at it,
• 8:32
  we'll say this quest, this answer is better or that answer
• 8:34
  is better. So a lot of these different AI's
• 8:39
  out there have been fed the entire Internet from 2023, and
• 8:45
  they don't really look at things that are that current because so
• 8:49
  much of it's been poisoned by AI slop that's out there, but
• 8:53
  they're all trying to improve it.
• 8:55
  So what you get is one of the AI companies is really focused on
• 8:59
  programming that is Anthropic, by the way.
• 9:03
  So they have programmers who are sitting there and helping to
• 9:09
  train the AI. For instance, I use Anthropic
• 9:12
  for a lot of the code that I'm writing and it asks me, hey, we
• 9:16
  will double your usage for no extra charge if you don't mind
• 9:21
  answering some questions from time to time.
• 9:24
  So what will happen in the background is Anthropic will
• 9:28
  have a new version of an AI they want to test and they don't tell
• 9:32
  you that they're testing it, which is what you saw, right?
• 9:36
  It gets worse and then it'll say, OK, so did I do bad, good
• 9:41
  or great on these answers? So that's how anthropics
• 9:46
  training it with for regular people.
• 9:49
  So I didn't sign up to be a trainer.
• 9:51
  I kind of did, I guess I said, yeah, give me double the usage
• 9:55
  and I'll be glad to answer these questions.
• 9:57
  So they, they're all trying to focus on different things.
• 10:00
  This new one that you just mentioned that is coming out
• 10:04
  here pretty soon, it looks like is, is a bit of a concern for a
• 10:10
  few reasons. All of these agentic things,
• 10:13
  agentic AI is where you have an AI agent that performs a
• 10:18
  function for you. And there are a lot of those out
• 10:21
  there and some of even hit the news.
• 10:24
  But there's a great example. Did you see the one about this
• 10:26
  Stockholm cafe? I did not know that.
• 10:30
  Sounds like a fun story. Yeah, it is.
• 10:32
  So here's what happened. They, they had what they called
• 10:37
  Mona, which is a Gemini, Google Gemini powered agent.
• 10:41
  And they gave this, this smart small startup coffee shop a
• 10:45
  $21,000 budget of this new agent, right?
• 10:51
  And the idea was, hey, let's go ahead and let's have it do
• 10:55
  everything, hire the staff, place the orders for goods,
• 10:58
  maintain the inventory for the cafe in Meet Humans obviously
• 11:03
  did the actual work of catering, receiving the AI overlords
• 11:07
  commands, right. All all hailed the AI overlords.
• 11:12
  But since launching in mid-april, it brought only
• 11:16
  seven, $5500 in sales and burned through $16,000.
• 11:22
  And it apparently made some, the AI made some terrible, terrible
• 11:27
  decisions. And it was given a simple set of
• 11:30
  instructions. It, it ordered like 3000 rubber
• 11:34
  gloves for first aid kit, 6000 napkins, canned tomatoes, which
• 11:39
  aren't used in any of the dishes in the cafe.
• 11:42
  Right, right. So people are counting on these
• 11:45
  things to do specific work. All of these different AI
• 11:49
  companies are focusing on training in different areas.
• 11:52
  Open AI is now trying to compete with Anthropic on the
• 11:56
  programming side, and Tropic is competing meeting with Open AI
• 12:00
  on the legal side. By the way, Anthropics Opus is
• 12:03
  pretty darn good in legal stuff, but you're going to see that
• 12:07
  juggling a lot as time goes on and, you know, clear winners.
• 12:12
  There aren't any and it changes all the time.
• 12:15
  That's the big take away, isn't it?
• 12:17
  Well, we're going to have to end there, Greg Peterson.
• 12:19
  Sadly, I have to stop. We are out of time.
• 12:21
  Talk to you again next Wednesday, Sir.