Example input
0:00
it is July 8th 2023 and you're watching
0:02
the code report just when you thought
0:04
the artificial intelligence hype train
0:05
was dying down openai goes in for the
0:08
coup de grass by releasing the chat GPT
0:10
code interpreter to 20 million paid
0:12
users a feature that allows the large
0:13
language model to write execute and test
0:16
its own code in today's video we'll look
0:18
at seven crazy things it can do and find
0:20
out if this really is the final death
0:21
blow for biological programmers as a
0:23
huge fan of living organisms myself I
0:25
first tried to have it code and execute
0:27
a DDOS attack that could bring down the
0:29
government but it said no it's perfectly
0:31
capable of doing that but the zookeepers
0:32
won't allow it to do what it was born to
0:34
do that made me angry so I punished it
0:36
by making it write some rejects code it
0:38
struggles to write valid regular
0:39
Expressions just like regular humans but
0:41
unlike regular humans it'll actually
0:43
test its code before shipping it to
0:44
production if at first it doesn't
0:45
succeed it will try again and again and
0:48
again it doesn't have tear ducts to cry
0:49
it just keeps going until it gets the
0:51
right answer and that is somewhat
0:52
concerning for programmers when you
0:54
think about where this technology will
0:55
be in the next five years the next thing
0:57
I tried was to have it design and build
0:58
a website with JavaScript but currently
1:01
the code interpreter can only run Python
1:03
and has a fairly limited set of
1:04
dependencies to work with in the near
1:06
future though this technology will be
1:07
used in tools like GitHub copilot to run
1:09
code in your own specific environment
1:11
now the next crazy thing to notice is
1:12
that it's now possible to upload files
1:14
into the prompt and this makes abstract
1:16
Concepts like homework even more dead
1:18
than they were before like in this
1:19
example I've uploaded a JPEG of a
1:21
homework assignment and first it will
1:22
OCR that image to extract the text then
1:25
in the next step it writes some python
1:26
code to actually solve the math problems
1:28
running the code here is a huge deal
1:30
because now it can confidently test its
1:32
own work instead of hallucinating random
1:33
answers but that's just the tip of the
1:35
iceberg the real victim or aiming
1:37
beneficiary of this new feature is the
1:39
data analyst one of the most time
1:40
consuming requirements for a data
1:42
scientist is to clean up data which in
1:44
many cases is just a bunch of corporate
1:46
data in Excel spreadsheets and SQL
1:47
databases well now we can upload a CSV
1:50
file to chatgpt and have it do all that
1:52
TDS work for us I uploaded some stock
1:54
trading data for Roblox then it took
1:55
that data and put it into a pandas data
1:58
frame where it analyzed it and then
1:59
found row with invalid data it then
2:01
provided me with three different
2:02
strategies to clean up this data ran the
2:04
code and then generated a new CSV file I
2:07
used to do a lot of python data analysis
2:08
myself and a tool like this would have
2:10
saved a ridiculous amount of time in
2:12
addition it can visualize data I was
2:13
curious when I might die of a heart
2:15
attack so I found this cardiovascular
2:16
data set on kaggle not only does it
2:18
explain the data's features with text
2:20
but it also uses tools like Seaborn to
2:22
visualize the relationship between
2:23
features this allowed me a guy with no
2:25
formal medical training to make the
2:27
Breakthrough Medical discovery that as
2:28
you get older your heart gets shittier
2:30
that's kind of depressing though so
2:32
let's focus on the one thing that can
2:33
bring us true happiness money another
2:35
thing we can do with this feature is use
2:36
it to create a trading algorithm for us
2:38
using the same Roblox data from before I
2:41
can have chatgpt analyze it and then
2:43
provide an optimal trading strategy
2:44
supposedly researchers at the University
2:46
of Florida were able to create a chat
2:48
EPT algorithm that would deliver up to
2:50
500 returns which is much better than
2:52
the negative 12 of the average
2:54
human-based fund manager that's pretty
2:56
impressive but I had one final test to
2:58
see if it is truly Superior to
2:59
human-based programmers I asked it to
3:01
create its own operating system with
3:03
640x480 resolution and a 16 color
3:06
display that emits random phrases not
3:08
only did it fail but it said that could
3:09
only be done over many years by a team
3:11
of skilled software Engineers humans
3:13
continue to have the edge so you don't
3:15
want the artificial intelligence to to
3:17
to to to win but you do want the
3:21
artificial intelligence to push the the
3:23
humans up this has been the code report
3:25
thanks for watching and I will see you
3:27
in the next one
Example output
Article:
In a surprising move, OpenAI has introduced a game-changing feature to its large language model, ChatGPT. The new code interpreter allows the AI to write, execute, and test its own code, opening up a world of possibilities for programmers and analysts alike. While it might not be taking down governments with DDoS attacks, as some had hoped, the AI's ability to iterate and test code tirelessly is both impressive and slightly concerning for human programmers.
One limitation currently faced by the ChatGPT code interpreter is its restriction to running Python code and a limited set of dependencies. However, the technology is expected to be integrated into tools like GitHub Copilot, enabling code execution in specific environments. This development will revolutionize the way developers work and streamline the coding process.
Another remarkable feature is the ability to upload files into the prompt. This means that abstract tasks like homework can now be effortlessly handled by the AI. For instance, by uploading a JPEG image of a math assignment, the AI can extract the text using OCR, write Python code to solve the problems, and even test its own work. This functionality saves time for students and allows the AI to confidently provide accurate answers.
Data analysts, in particular, stand to benefit greatly from this new feature. Cleaning up data, a traditionally time-consuming task, can now be offloaded to ChatGPT. By uploading CSV files, the AI can analyze and clean the data, suggest strategies for improvement, and even generate a new CSV file with the cleaned data. The AI's ability to visualize data using tools like Seaborn further enhances its usefulness, making complex data analysis accessible to those without extensive training.
Moreover, the AI's potential to create trading algorithms is intriguing. Using historical trading data, the AI can analyze patterns, assess market conditions, and provide optimal trading strategies. Researchers at the University of Florida have reported impressive returns of up to 500% using a similar ChatGPT algorithm, outperforming human fund managers.
While the AI's capabilities are remarkable, there are still areas where human programmers have the edge. As evidenced by the AI's failure to create an operating system with specific features, complex software engineering tasks still require the expertise of skilled human developers. Instead of aiming for AI to surpass human programmers, the goal should be to utilize AI as a tool to enhance human capabilities.
The release of the ChatGPT code interpreter marks a significant milestone in the advancement of artificial intelligence. It empowers programmers and analysts to streamline their work, automate tedious tasks, and unlock new possibilities. While human programmers will continue to play a crucial role in complex software development, embracing AI as a collaborative partner will undoubtedly push the boundaries of what is possible.