I just released a video on an article I wrote previously here on the Website and LinkedIN.
Below is the link to the article and the transcript.
AI Dementia: Memory Fail?
(https://www.youtube.com/watch?v=3YzdrRw8TVs)
0:00
hey Trailblazers welcome back to pcSHOWmeAI
0:03
have you ever wondered if AI has
0:06
its own quirks well today we’re going to
0:09
be diving into an intriguing experience
0:12
that I had with Cloud AI where
0:15
everything seemed to go perfectly until
0:18
it didn’t let’s explore this little
0:20
glitch and see how AI remembers or
0:23
perhaps more accurately
0:26
doesn’t show me a show me AI P show
0:31
me the other day I was talking to CLA Ai
0:35
and something very interesting happened
0:38
we were working on several complicated
0:41
procedures in Azure there was one
0:43
procedure that I couldn’t quite remember
0:46
I hadn’t done it in some time so I asked
0:49
him what the steps were to accomplish
0:51
this and Claud laid out the procedure so
0:54
as I went through that in the Azure
0:56
portal I realized that one of the steps
1:00
he gave me used the incorrect word it
1:04
started with the same letter as the
1:06
correct word but it was a different word
1:09
and then the steps afterwards were
1:11
correct for the procedure so I finished
1:14
doing what I needed to do in Azure and
1:16
then I went back to Claud and I said I
1:19
found this very interesting let me show
1:22
you what he said take a look at this
1:24
this is kind of wild so he he did the
1:26
first steps right then he in step five
1:28
he said in the insight Page look for a
1:31
section called monitoring coverage or
1:35
connected workspace however it’s not
1:38
monitoring coverage it’s monitoring
1:41
configuration so after I completed that
1:44
I went back and I said thanks it’s
1:46
monitoring
1:47
configuration not monitoring coverage I
1:50
said wow I never saw you do that before
1:53
it’s almost like he was saying I can’t
1:56
quite remember but it’s something like
1:58
monitoring coverage or something close
2:01
to that I said that’s that’s close to
2:03
what people do and he’s like you’re
2:05
absolutely right I appreciate you
2:07
catching that and you know how the story
2:09
goes after that the point is so Claud
2:12
and other AI models don’t remember
2:15
things the way people remember them
2:17
right for example they don’t
2:20
experience memory Distortion or memory
2:23
Decay over time they don’t consolidate
2:26
memories and reconsolidate memories or
2:28
forget things like that
2:30
so I was curious as to what happened and
2:33
why it was just this one word that was
2:35
out of place instead of usually when an
2:38
AI model hallucinates it’ll just kind of
2:41
go right and then you don’t know where
2:43
you end up okay so let’s clear something
2:46
El as I said AI doesn’t actually have
2:49
memory the way we do when you’re
2:51
chatting with an AI like clot it can
2:53
refer back to the things you said
2:55
earlier in the conversation but once you
2:57
end the session it’s like hit hitting
3:00
the reset button it’s gone no carry over
3:03
to the next chat for the most part
3:05
there’s been certain advancements in the
3:07
field and models like chat gbt can carry
3:11
things over to the next chat or even
3:14
cross chats to other conversations and
3:17
there’s certain tricks and ways they do
3:19
this also and we’ll get into that later
3:21
in this video and or in the next video
3:24
for the most part what clud really has
3:26
is a form of contextual understanding
3:29
Within each conversation but he doesn’t
3:31
store those conversations for the future
3:34
so what does this mean for us well first
3:37
off for the most part there’s no
3:39
persistent memory in AI everything
3:42
resets when the conversation ends
3:44
secondly will AI can understand context
3:48
within a session it doesn’t actually
3:51
learn from interactions the way humans
3:53
do and third there are ways that AI can
3:56
create the illusion of content uity that
4:00
are beneficial however we need to be
4:03
aware that this is not memory as we
4:06
would typically think of it it’s really
4:09
important to understand these aspects
4:12
especially as we continue to integrate
4:15
AI into our lives as we go forward we
4:18
need to keep in mind the limitations
4:21
while appreciating what it can do this
4:25
ties back to some ideas I explored in my
4:27
earlier work when I wrote my wife paper
4:30
AI intuition transcending
4:32
anthropomorphism with project uplift in
4:35
a sense we’re building a relationship
4:38
with AI but it’s essential to recognize
4:41
that it’s a different kind of
4:43
relationship from that we have with
4:46
other humans and if we keep this in mind
4:49
as we go forward we can achieve the best
4:52
possible outcome in a helpful way what
4:55
do you think have you noticed anything
4:58
like this with AI have have you noticed
5:00
patterns that repeat over and over that
5:03
you at one time thought were more
5:05
complex interactions drop your thoughts
5:08
in the comments I’d love to hear your
5:10
experiences if you enjoyed this video
5:13
remember to hit like And subscribe and
5:16
hit the notification Bell for more deep
5:19
dives into Ai and Tech until next time
5:22
keep blazing those Trails later
5:25
Trailblazers
