Paul Stenhouse: AI companies battle with the US Department of Defense

The AI companies are in a battle with the US Department of Defense/War 

The US government has plans to use AI and wants to work with Anthropic but anthropic isn't so sure. The government wants unrestricted access, no guardrails, no safety protection that is usually built into the Claude chatbot to prevent misuse. Anthropic is especially concerned about using AI to survey the public, and carry out lethal strikes without human involvement. 

The DOD/W is threatening to invoke the Defence Production Act - which would force the company to make its product available - or label it a 'supply chain risk' to prevent military suppliers from using it. Anthropic has pointed out that it's a little contradictory to say that it is both essential and not to be used in the same threat. 

Open AI is backing Anthropic. 

 

Open AI raised more money - more circular money merry-go-rounds 

This time Amazon is throwing them tens of billions of dollars in exchange to use their in-house made chips and the AWS products. Bloomberg created a graphic showing all the AI deals and investments and it's wild just how many arrows are pointing every which way. Some analysts are nervous that if the AI boom turns to bust, there will be massive ripple effects. 

LISTEN ABOVE

See omnystudio.com/listener for privacy information.

Episode Transcript

Speaker 1 (00:07):
You're listening to the Saturday Morning with Jack Team podcast
from News Talks at B.

Speaker 2 (00:12):
I was saying just after nine o'clock this morning that
I've had a bit of a wow moment once again
with AI. I'm not a total AI hyper I'm not
like going whoo for AI everything, but I've had a
few moments over the last couple of years where using
the AI tools that are available, I thought, oh, my goodness,
this is really impressive. So the first time I use

(00:33):
chat GPT, I thought that. The first time I saw
some of the video generation, I thought, okay, this is
pretty impressive. But now, for the first time in my life,
I have created a website, I've created an app. I've
created a few other bits and pieces via Vibe coding,
so zero programming experience, zero coding skills, but talking to

(00:54):
Clawed code, which is one of the big AI systems,
I've been able to just use English prompts to make
some of these things, which has been quite a remarkable,
a magic thing to experience. Honestly, our textbook, Paul Steinhouse
is a little more skillful and a little more ticked
adapt than I tick adept than I am, and he's

(01:15):
with us now to Paul.

Speaker 3 (01:17):
Yeah, morning Jack, So you went straight for Claude code.
That's pretty impressive because you claud code you have to
use in the terminal. Did you use it in the
term you don't have to use it on the screen.

Speaker 2 (01:26):
Nope, No, you don't have the desktop app. I use
the desktop app to do it. So yeah, yeah, no,
I tried. I over to terminal. Yes, I've connected to
get hub.

Speaker 1 (01:36):
There you go.

Speaker 2 (01:37):
See, I got called. I said, Claude, treat me like
an idiot. I talked to the chatbot. First of all,
I said, treat me like an idiot clad and it said,
don't worry, Jack, we know you're an idiot. And I said,
you tell me, like, talk me through and really basic
steps what I need to do. And I said, okay,
you need to get a GitHub account, you need to
use terminal. I did that and then it was like, oh,
the easier way to do this is this on the
desktop app. So that's what I did. But I mean

(01:59):
it is. I mean it is. It's hard to distinguish
the AI hype from reality, right and to work out
how warranted it is. But in seeing what it could
do with my basic English language prompt and seeing how
quickly it could do it, it's hard not to imagine
that people who are a little more skillful than me
when it comes to computer programming and coding and stuff

(02:21):
are able to build really impressive things. Oh for sure.

Speaker 3 (02:25):
I think what's really fascinating about the AI type stuff
is that if you're a really good communicator and you
know the right terminology and the words in particular industries,
you can go really far. And the thing in AI
is that every single word matters, So when you're thinking
about how you're prompting it and what you're saying, it

(02:46):
just gets better and better, the more specific and the
more if you're using the correct terminology, it just does things.
It doesn't have to guess as much. It doesn't make
quite as many mistakes because what it is really is
autocomplete on steroids.

Speaker 2 (02:58):
Yeah right.

Speaker 3 (02:59):
It's looking at this massive amount of information, old projects
and all this sort of stuff, and it's literally just
like compare ones and twos ones and two ones and zeros,
i should say, And it just literally is looking to
see where the patterns are. And that's but you can
create with it. It's it is, it's remarkable.

Speaker 2 (03:17):
But the thing that's interesting is it can work backwards. Right,
So if you say to it, you know, I want
a rhyming sonnet, and you give it the first line,
so make me a rhyming like like or give me
a ryming couplet, right, So so I want I want
you to come up with a second line that makes
sense and rhymes with this.

Speaker 3 (03:35):
It will.

Speaker 2 (03:36):
If it were just autocomplete on steroids, then it would
just that The obvious thing would is that it would
go to the next word, and then the next word,
and then and then then the next word. But it knows,
because you want a rhyming couplet, that it's got to
start with the rhyming point and then work backwards. So
it's smart enough to do. I mean, that's quite a
basic example, but it's smart enough to be able to
kind of contextualize requests in that way.

Speaker 1 (04:00):
Yeah. I mean.

Speaker 2 (04:00):
The thing that's amazing about these new systems too, is
that obviously a lot of the code for the systems
themselves is being written by the system, which is it's hating.

Speaker 3 (04:10):
It's incredible, wasn't it. It's crazily circular. Okay, so can
we go? Can we go slightly even more dystopian? So
there's Jack Tame, you know, New Zealand's sexiest personal live
turned software engineer.

Speaker 2 (04:21):
What was it?

Speaker 3 (04:21):
Wasn't it the award you wont wasn't.

Speaker 2 (04:23):
It a few years ago?

Speaker 1 (04:24):
People? Yes?

Speaker 2 (04:26):
I mean you were the one who bought it up.
But you know, we need to get lost in ancient history.

Speaker 3 (04:31):
Okay, So imagine yourself. You're now working at the US
Department of War and you you just create a little
app and it decides that it wants to launch an
attack from an overhead drone. Is that a good idea
or a bad idea? Because it's happening. Well, the conversation

(04:51):
about that exact thing is happening right now. And while
I was waiting to talk to you, I got an
alert from CNN saying that because of this rift between
anthropic who makes flawed code and clawed the chat box,
they've got some hard lines in the stand. They work
with the Department of Defense slash War at the moment,
and they've said, you're not going to be using it

(05:12):
for anything like that. We do not want like, you
can't just start launching missiles that AI directed, and AI
decided no, thank you. It also said you won't be
using this to help surveil the public. So I guess,
looking at vast quantities of data and figuring out things
that are going on, the Department of Defense slash war said, well,

(05:33):
we want to use it, and we don't want you
to say what we can and can't do. And so
they have until it's currently four forty three here in
New York at the moment. They had until five one
pm to reach an agreement with the Department of Defense.
But President Trump has just tweeted a truth social I

(05:54):
should say, and said that every government department now has
six months to get off everything Anthropic, and the Department
of Defense is now deciding whether or not they'll be
invoking the Defense Production Act. Well they were previously. Who
knows what they're doing now, but the talk was they'd be.
The threat was they were saying they'd invoke the Defense

(06:14):
Production Act, which would force Anthropic to make its product available,
or they're going to go down a different road and
label it a supply chain risk to prevent every military supplier,
which there are obviously a number, from using it. So
crazy kind of situation at the moment. And you've got Anthropic,
whose biggest competitor is open AI behind chat GPT that

(06:38):
has backed Anthropic and said, yeah, those guardrails and things
sound you know, probably pretty good. So it sounds like
it's going to be the Department of War battling the
AI companies.

Speaker 1 (06:51):
Yeah.

Speaker 2 (06:51):
Man, it's going to be amazing to see how this
scends up. And it certainly sounds like what you've advised
us to read. The president's response means that it's only
going to get more and more feisty from here. Thank you,
paul A Textbert Paul Stenhouse. They're going to be one
to watch over the next few well minutes, hours, weeks, months,
et cetera.

Speaker 1 (07:08):
For more from Saturday Morning with Jack Tame, listen live
to News Talks ed B from nine am Saturday, or
follow the podcast on iHeartRadio.