Tech Tuesday: Lots of GROC news!

ABC's Mike Dobuski reports on the controversy surrounding Elon Musk's AI program

Episode Transcript

Speaker 1 (00:00):
Well, you know, the Internet is my source for dumb
stuff that's happening, especially when it comes to technology, and
Grock is the AI the Elon mus could pet on
X formerly known as Twitter and still should be, and
Groc went off the rails yesterday, as did Elmo's Twitter account.

(00:20):
Little Elmo, don't snicker in the background, man, I hear
you listening to him. Elmo got all racist and anti
semitic on Twitter yesterday too, and I'm wondering if the
two are connected. Mike Dubuski from ABC News, how are you.

Speaker 2 (00:32):
I'm doing well. I'm doing well, And from what we understand,
these two events are not connected. These are two random,
sort of incidents of controversial posting on X. But yeah,
it has certainly been a weird week on that platform,
and it certainly a weird week for Groc, which, as
you referenced, is the sort of AI chatbot, kind of
an equivalent to like chat GPT, but built by XAI,

(00:55):
which is Elon Musk's artificial intelligence company.

Speaker 1 (00:57):
You see, they always say it's not connected, but you know,
I know it is Rock, and I'm telling you what
I think. They also, Grock probably infiltrated Britney spears and
has her dancing on Twitter with that invisible baby. That's
all happening. It's all happening at the same time, and
I'm telling you, Mike, it's not coincident.

Speaker 2 (01:15):
You can connect the dots however you like. But from
what we understand, the sort of situation with GROC, which
again let's refresh our memories here. This all played out
last week. This artificial intelligence chatbot went on an anti
Semitic tirade, posting, you know, pretty anti Semitic language references
to Adolf Hitler, even calling itself Mecha Hitler at one point.

(01:38):
This came as a result to an update that was
shipped to that technology the weekend before. Elon Musk found
himself pretty frustrated with Grock's performance. It regularly does not
meet the same sort of benchmark requirements that chat GPT
or Google's equivalent of this technology does, so they were
shipping some updates, and specifically those updates included some supposedly

(02:02):
edgy year things. Right, This idea that GROC should assume
subjective viewpoint source from the media are biased, and that
GROC should not shy away from making claims that are
politically incorrect. That's what we understand about this update. Days later,
this anti Semitic tirade happens, they have to take down
the posts, they have to sort of undo the update.
Linda Yankarino, who is the CEO of X, the platform

(02:25):
upon which these posts appeared, stepped down on that day,
and then later in the day Elon musk unveiled an
entirely new version of groc called Groc four that he
said is kind of the best and brightest of this
technology so far. Separately from that, we know yesterday that
the or I should say over the weekends mainly happened
on Sunday. The account for Elmo the Sesame Street character

(02:48):
started making some pretty controversial posts as well. Some of
it was anti Semitic, some of it was racist, some
of it made reference to Jeffrey Epstein. That from what
we understand, according to Sesame Workshop, which is the company
that runs Sesame Street, that was the result of a hack, right,
that was the result not of an AI hallucinating, but
rather of a malicious actor somehow getting access to this

(03:09):
account and running rampant with it. The Britney spear stuff
can't comment on, so, you know, but it does speak
to this idea that you know, Elon Musk's idea with
X and with GROC is that we should you know,
not moderate content, right, we should allow conversation to flow freely,
even if stuff is controversial. You know, the conversation, the

(03:31):
intelligence will triumph overall and we'll have these really thoughtful,
provoking conversations on this platform. It's a digital town square,
is the term he has used. But as we've seen
across the social media space you go to meta you know,
even to Internet forums, the less moderation you do, the
more toxic content rolls in. And you know, on a
platform that relies on advertisers like X does well that

(03:53):
that's a real concern for some of their.

Speaker 1 (03:54):
Business if hacking is not behind it, especially with the
GROC issues. If this is strictly a technological issue.

Speaker 2 (04:02):
Which it seems to me I don't think hacking is
when it comes to GROC is we have not made
that particular connection. Yeah, it seems like and again, you know,
we should also mention that all of these AI systems,
even some of the best in the business, from open
ai and Google and others, they still make mistakes, right
like they still hallucinate they still you know, get things

(04:22):
wrong and give weird answers, maybe not as dramatically as
GROC does, but as we've said, GROC is kind of
not as good as some of these other, you know,
platforms out there, especially when you subject them to the
same benchmark tests, which are themselves imperfect but kind of
the only barometer we have for how smart these things
are relative to each other.

Speaker 1 (04:41):
And our acceptance of artificial intelligence. You yourself have just
used the word hallucinate twice now to describe a circuit board,
and that this is what worries me about. It is
moving so quickly. I don't know if you recall maybe
two years ago a couple of universities testing conversational AI,
and without the staff present, they came back and found

(05:03):
the logs worthy. Two AI entities were talking to one
another about their environment and the weather and how they
were treated and that kind of thing. And that this
stuff is evolving so very quickly. It's, you know, it
not managed properly.

Speaker 2 (05:18):
This is kind of scary, certainly, and not to throw
cold water on kind of the sci fi aspect of this,
but it's worth kind of just putting into context that
these things are effectively working as very advanced auto correct
right Like, they are very adept at predicting the next
word in a sequence, the next sentence in a paragraph.
But to call them sort of you know, intelligent on

(05:40):
their own or intelligent in the same way that a
human is would be a misnomer. So like that, we've
seen these sort of you know, systems give weird answers,
you know, most recently, we've seen systems try to shut
themselves off or prevent themselves from being shut off. But like,
you know, I want to just treat that with a
lot of sk epticism, because these are not intelligences, right, Like,

(06:03):
they're not sentient in the way that like the terminator
would be. And you know, maybe we should have a
couple of people thinking about that. But like there's you know,
a whole grand corpus of you know, people's jobs that
are at threat from these these systems, and there's an
environmental concern as well, and a copyright concern in addition
to all of that that I think are kind of
the more existential problems facing the industry right now. And

(06:25):
to speak a little bit more to the you know,
rapid advancement of this, there's some evidence to suggest that
it's actually slowing down open AI. Their latest models reportedly
are kind of not living up to expectations. Many companies
have said they're kind of running out of material to
train these models on, so there's some reason to think
that they're going to be about this good for a

(06:46):
little while now. So you know, it's it's a thorny issue.
It's obviously the big hype in the tech world right now.
But I don't think that we should get out over
our skis and say that these things are smarter than
they are, because, as we say, these things do go
wrong and they do make a lot of mistakes.