0:00

Welcome to the prompt engineering podcast,

0:03

where we teach you the art of writing effective

0:05

prompts for AI systems like chat,

0:07

GPT, mid journey, Dolly,

0:10

and more. Here's your host, Greg

0:12

Schwartz. Hello,

0:15

everyone. I am back.

0:18

So I appreciate the time

0:21

that you were all able to give me to take off.

0:24

It was very helpful. I'm

0:26

on two new medications and they're

0:28

definitely helping. Still working

0:30

on dosage and a lot of other stuff,

0:32

but it has made a huge difference. So

0:34

I'm going to be publishing again.

0:37

And this is the first one. So

0:39

wanted to share. I just got back from

0:41

AI four, which is an AI

0:44

for enterprise conference and

0:46

had a good time there, got to see a bunch

0:48

of different people and see a lot of different

0:51

talks. In fact saw a ton

0:53

of different stuff. Lots of talks from

0:55

people selling things to enterprise.

0:57

A lot of non-technical talks as well, and

0:59

a few technical ones. Unfortunately,

1:02

there were Quite a few talks

1:04

and products that were really

1:07

a solution in search of a problem. You've

1:10

probably seen that before. And frankly, you've probably

1:12

seen this. Comic illustrating

1:14

that before. And by the way people who

1:16

are just listening. This is particularly

1:19

an episode that you might want to go to YouTube

1:21

to watch because I'm showing some

1:23

different clips from the conference. But

1:26

you'll still be able to follow along on audio. If

1:28

you don't know, the YouTube channel is youtube.com/@PromptEngineeringPodcast.

1:36

You can also go to PromptEngineeringPodcast.com

1:39

to get the link to YouTube and to

1:41

watch this episode. Did

1:44

want to share. I saw a few particularly

1:47

really good talks and a

1:49

really interesting discussions. So

1:52

the first one was A company called

1:54

vital. Which is founded

1:56

by the former founder of mint.com.

2:00

They created a tool to translate.

2:03

Doctor's notes and all the jargon and

2:05

medical terminology into plain English.

2:08

Now I interviewed them. So

2:11

I'll save more discussion on that for later.

2:13

But you're seeing at the moment. An

2:16

example of me running some information through

2:18

it and you can see it. Turns it

2:20

into some actual normal English

2:22

words. I also saw two talks on

2:24

weather. Those actually

2:27

had no chatbot stuff in the middle. They

2:29

were like run of the mill machine

2:31

learning and they were really cool. Two

2:34

interesting uses that I had not

2:36

heard of. One of them

2:38

was able to predict wildfires.

2:42

And also paths

2:44

of hurricanes. And then there was another

2:46

system Google's weather prediction

2:48

system, which also was able

2:50

to do. Hurricane prediction Next,

2:53

there was a talk on bias.

2:55

And you know how to deal with biases.

2:58

In large language models. So

3:01

examples of that would be things like. Biasing

3:05

based on gender or race

3:07

or religion or political ideology

3:10

or profession and all these different things.

3:13

And then They also,

3:15

the speaker also talked about

3:17

some of the ways that you can actually remove

3:20

bias. Particularly that most

3:23

ways of doing it are just, humans looking at

3:25

it and going, hang on. That's biased. But

3:28

people are starting to play with. Actually

3:31

asking a model and an

3:33

LLM two. Say

3:36

is this biased, but there's some interesting

3:39

technical challenges with that. And

3:41

then there are examples of anthropic

3:43

and other companies who are looking at.

3:45

Creating what's called the constitutional AI. And

3:48

I'm actually looking for a speaker to come on and talk

3:50

more about that. So if that's something you know about, please

3:52

let me know. And then my favorite talk

3:54

was one given by the founder

3:57

of the Khan academy. They've

3:59

built a chat bot that doesn't simply give

4:01

you the answer the way, you can ask chat GPT

4:03

to solve problems or do your homework.

4:06

Instead, it helps you to

4:08

make better decisions

4:10

and basically learn. The

4:12

example that's being shown here is

4:15

actually giving you feedback on how to

4:17

write a better essay. But

4:19

they also showed some examples

4:21

of, for example, with a math problem,

4:24

say. Y

4:26

equals two. Times

4:29

parentheses X plus seven close parentheses.

4:33

The example shown was then incorrectly

4:35

applying the times two to the

4:37

pieces in the parentheses. And

4:40

it then talked through how

4:42

to actually do it right. And showed you the next

4:44

step and said, could you explain how

4:46

you got this answer? Because that's not what

4:48

I got. Which I thought was an interesting

4:50

nuance. It doesn't say that's wrong. It

4:52

just says "that's not what I got, could you tell me how you

4:54

got there?" One

4:56

of the things that was really interesting about the chat bot was

4:59

you could turn on a

5:01

different mode. That was the teacher mode. And

5:04

then it would, for example, show you answers.

5:06

You could say, what's the answer to this question. It would say the

5:08

answers, selection B. And

5:10

then it would say. Do you want to brainstorm ways

5:13

of getting your students excited about this topic

5:15

or of discussing this

5:18

question with your students? Things like that.

5:21

So had some pretty cool uses for

5:23

trying to help teachers. So

5:27

I am back. This is just a very

5:29

quick, fast episode because I wanted

5:31

to just put something out there and say, hi,

5:33

I am still around. And

5:35

look forward to, I've got

5:37

interviews coming up with like I said,

5:40

vital, which is the doctor.

5:43

Jargon to patient translator.

5:45

Plus they do a bunch of other stuff around.

5:48

Machine learning around medical records and things like

5:50

that. Also have a

5:52

discussion with me and a bunch of other

5:54

podcasters at AI4

5:56

that we'll be publishing sometime soon. And

5:59

then also going to be talking

6:01

to a bunch of other people that I met at AIA

6:04

four and other places. One last

6:06

thing. If you are familiar

6:08

with tree of thought.

6:11

I would love to chat with you. There's a contact form

6:14

on the website, PromptEngineeringPodcast.com

6:18

or email me, reply in the YouTube

6:20

comments, I would love to bring someone on

6:22

to discuss those. Oh, and

6:24

also if anyone is working

6:26

on constitutional AI or is

6:28

familiar with it, I would love to bring someone on

6:30

to talk about that because I am very

6:32

curious about it. Thanks. Hope

6:35

you're having a great week and I will talk

6:37

to you soon. Bye. Thanks

6:40

for coming to the prompt engineering podcasts podcast

6:43

dedicated helping you be a better

6:46

prompt engineer Episodes are released

6:48

every Wednesday I also

6:50

host masterminds where you can collaborate

6:52

with me and 50 other people live

6:55

on zoom to improve your prompts

6:58

Join us at promptengineeringmastermind.

7:00

com for the schedule of the upcoming masterminds.

7:03

Finally, please remember to like

7:06

and subscribe. See you next week.