Should you still learn to code? (ft. Devin)
Summary
Please replace the link and try again.
Takeaways
- 🤖 脚本讨论了未来工作的变化,特别是AI在编程和数据分析领域的应用。
- 👨💼 CEO认为学习计算机科学并非必要,这与普遍观点相反。
- 🔧 Devon是一个全新的自主AI工具,能够通过简单的提示执行复杂的任务。
- 🛠️ Devon展示了作为AI软件工程师的能力,包括调试代码和数据科学分析。
- 🔑 尽管Devon功能强大,但它在解决实际问题时需要详细的指令和人类指导。
- 📈 脚本中提到了GitHub的新工具,它可以自动修复代码中的漏洞,减少对人类干预的需求。
- 🎓 Corsera提供了Google数据分析师证书课程,覆盖了数据分析的核心技术和工具。
- 📊 Claude作为经济分析师的演示,展示了AI在处理经济数据和预测方面的潜力。
- 📈 Claude在分析世界经济时,使用了大型语言模型框架和并行处理技术。
- 📊 Claude提供了世界主要国家经济预测的图表和总结,尽管图表选择有待改进。
- 🚀 脚本强调了使用编程和大型语言模型结合来解决复杂问题的潜力和未来趋势。
Outlines
Please replace the link and try again.
Highlights
Please replace the link and try again.
Transcripts
that nerds our jobs in the future are
going to be a lot different the CEO of
the third most valuable company made
this bold claim um almost everybody who
sits on a stage like this would tell you
it is vital that your children learn
computer
science um everybody should learn how to
program and in fact it's almost exactly
the opposite so is it even worth your
time learning how to program well in
this I'm going to be exploring a few new
technologies that just came out in order
to answer that question last week the
world was introduced to Devon a fully
autonomous bot equipped with tools like
a coding environment and browser with
this it can basically take over the
world providing it with only a simple
prompt it gets to work putting together
a plan of action breaking up what it
needs to do into simple tasks it starts
by browsing the internet to make you
feel more confident that its answers
aren't going to be hallucinations then
jumps into some light coding to make you
even feel less secure about your
long-term job security and then after it
fixes a quick bug it provides this
groundbreaking analysis that you can
provide to your boss as your own work so
let's break Devon down it's advertised
as the first AI software engineer
there's a lot of problems with that
we'll get to that but let's actually
look at some of the use cases they've
used it for now all these demonstrations
were done by employees of cognition so
to be fair there haven't been a lot of
outside tests of this tool anyway in
this case the engineer wants to fix a
bug in the code he provides some pretty
detailed instructions and Devon gets to
work now one of the impressive things
from this demo was it used as an
iterative approach so Deon here actually
wrote uh actually added a print
statement to debug the outputs uh and
the uh inputs to the failing test reran
the tests and actually found which case
was wrong which is actually a second bug
that Devon found and it then went and
updated the code to fix this second bug
and with that demo you may be like Luke
I'm a data nerd not a software engineer
this thing's just troubleshooting code
bases and not actually performing data
analytics so I have nothing to worry
about with my job well if you recall
from that first example that I showed
Devon did do data analysis additionally
they had a demo showcasing well as they
stated today I'm going to show you an AI
training in AI which is not only meta
but also shows that this is not just
geared towards software Engineers but
also has the potential defect Us in data
science now this followed a similar
approach as we've seen before of
downloading the code and in this case
going through and fine-tuning a model
which after about an hour it's only
about 4% done with training and
conveniently there's no conclusion on
what happened with the training now one
of the most impressive exercises that
demon demonstrated was the ability to
actually make money it was provided with
a problem from upward which side note
how are you going to calculate hourly
rates whenever AIS work almost
instantaneous anyway this thing was
looking at making inferences with a
computer vision model that's fancy talk
in this case as all it really wanted to
do was label potholes on a Road Devon
got to work and the first thing I noted
was some of the packages were out of day
so it updated it which then it found a
bug in the code which wasn't supposed to
be there and once again it used that
print statement approach in order to
find it I'll be honest I don't know why
it's not using a debugger so finally
after this it gets into running the
model and providing a detailed report on
it and even provides some screenshot
examples of it working in action along
with this final write up in a text file
that overviews the work and also the
conclusions that it came to not going to
lie if I received this on upwork I'd be
pretty impressed so there's a Core theme
that Devon is following that I found in
all these examples some human is unhappy
because it doesn't know how to solve a
problem so it all flows that to Devon
who gets to work Devon then in all these
cases went and pulled this GitHub
repository after it found a solution
working in an iterative approach and
reported back to the human I love you
Devon which should no longer be unhappy
and this demonstrates an important point
you still need a human in the mix in
order to guide Devon onto what problems
it needs to be solving oh wait what's
this GitHub introduces a new tool in
order to automatically fix code this new
feature promises that this new system
can remediate more than 2third of the
vulnerabilities that it finds often
without the developers having to edit
any code
themselves okay scratch that on human
intervention all right before we go
further we need to pay some bills and
give a shout out to the sponsor of this
video corsera which is having a special
deal right now now the number one course
that I recommend for aspiring data
analyst is the Google data analytics
certificate this covers not only what
it's like to be a data analyst but also
goes into all the core Technologies you
need to know including SQL programming
languages Vis tools and spreadsheets
it's where I recommend anyone new to
that analytics start and I've made a
number of videos interviewing those that
have taken this to better understand the
value of this certificate now right now
corsera is offering a heck of a deal
where you can receive $100 off your
yearly subscription to corsera plus
which works out to being less than a
dollar a day with this it not only gives
you access to the Google certificate but
also 7,000 other learning programs
including a ton of resources on my
favorite programming language now I'm
not just recommending corsera because it
was a sponsor of this video I've
actually personally paid for a corsera
plus and used it for my learnings as
shown by this receipt more recently I've
been using this to improve my knowledge
on applying AI in data analytics
specifically I've been working through a
lot of different courses and I just
completed this project based course on
using Python's Lang chain for analyzing
your own data which we're going to go
into more detail in a bit of what
Technologies I'm going to be covering
over the next year all right thanks
again to corsera for sponsoring this
video and let's get back to it so how
does Devon actually perform in a
comparative test to other common models
and these results were testing whether
it can resolve real world GitHub issues
Devon got a whopping 14% which you're
probably like Luke that's nowhere near
100% that's like 3 and 20 how the heck
is it going to actually do my job but if
you look at the best model in the market
today it's only at around 2 % so it's a
little bit better personally I think
this graph answers on whether you should
learn coding or not it's not solving all
the issues today or tomorrow but we're
on a positive trajectory to maybe one
day be at 100% but not anytime soon now
the other thing that reassures me from
this is that I don't know if you noticed
this from those videos that I showed
earlier but for complex tasks Devon
takes a fair bit of prompting and by
fair I mean a lot and this case I feel
the engineer had to go into an enormous
amount of detail in order to specify how
it wanted it to solve its problem which
with this level of specifity I think
even the free version of chat gbt could
solve it and that's where I think we are
with this technology today yeah although
they're claiming that Devon is first AI
software engineer Auto GPT which has
been around for almost a year now has
been doing a lot of the same things but
doesn't get as nearly as much virality
as Devon did which coincidentally is
happening almost in tandem when these
type of companies are raising funding
like cognition did last week I want to
be clear I'm not trying to on Devon
and say it's a bad tool in fact I think
quite opposite I think they've done
incredible advancements and we're moving
in the right direction but these type of
Technologies can be overhyped and is
driven by funding now there's another
announcement last week that I feel is
more relevant to us data nerds and it
deals with this model which is only
second to open ai's GPT 4 the team at
anthropic released this video on Claude
working as an economic analyst they
prompted it to look up GDP trends for
the US and write a markdown table of the
estimates which you got to work
transcribing this screenshot of a graph
of the GDP from there they went to
evaluate how accurate those
transcriptions were so it had the model
plot those transcribed values in this
interactive plot and then after having
provided the model the actual results it
plotted them side by side so how
accurate is Claude at using the vision
model for transcription we tried it with
a large sample of madeup GDP graphs and
its transcription accuracy was within
11% on average which not bad but
probably can be improve so then they
moved into having Claude use machine
learning in order to predict GDP in this
case using a Monte Carlo simulation and
just like most people thinks the US
economy is going to be just fine for the
next few years but really none of this
was impressive until I saw this where it
asked Claude to perform in an analysis
of the world's economy looking at more
than just one country in this case
although they didn't disclose it it
looks like they were using some sort of
large language model framework in order
to implement agents which all of these
agents were working in parallel
collecting all the data they needed for
these top countries and processing it
pretty dang impressive for the final
results it provided these pie charts
comparing the two values side note I was
a little disappointed with this because
pie charts are actually really bad at
comparing values but nonetheless it not
only provided an analysis it also
provided a final summary detailing how
the major countries planed a fair over
the next few years now I thought this
was more impressive because it
demonstrated how you can actually use
coding such as python to perform an
analysis with a large language model and
frankly this is where I see analytics
going into the future personally I'm
going to be exploring more on this
channel how to use things like python in
conjunction with libraries that build
out agents for large language models to
solve more complex problems all right as
always you got value out this video
smash that like button and if you'd like
to learn more about how to start coding
in data analytics I just made this
tutorial right here on how to learn SQL
all right with that see you in the next
one
浏览更多相关视频
Künstliche Intelligenz in der Schule | BR24
GitHub Copilot tips and tricks
I used the first AI Software Engineer for a week. This is happening.
How to Describe a Person in English - Spoken English Lesson
Bash Scripting on Linux (The Complete Guide) Class 01 - Course Introduction
The REAL Cost to Build My 450SF Tiny Home in 2024
10 Cara Untuk MEMENANGKAN Debat Apapun
5.0 / 5 (0 votes)