AI coding assistants just leveled up, again…

Learn about the latest updates to GitHub Copilot and its new GPT-4 upgrade. Also, take a first look at Google's new Duet AI and JetBrains AI Assistant.
#ai #programming #thecodereport
💬 Chat with Me on Discord
/ discord
🔗 Resources
- No programmers in 5 years / in_five_years_there_wi...
- Duet AI Launch cloud.google.com/duet-ai
- Google Gemini first Look • Google's Gemini just m...
- Tech Trends 2024 • You probably won’t sur...
🔥 Get More Content - Upgrade to PRO
Upgrade at fireship.io/pro
Use code YT25 for 25% off PRO access
🎨 My Editor Settings
- Atom One Dark
- vscode-icons
- Fira Code Font
🔖 Topics Covered
- Updates to GitHub Copilot
- Alternatives to GitHub Copilot
- Does Copilot use GPT-4
- Will programmers be obsolete in 5 years?
- Google Duet first look
- JetBrains AI assistant first look

[@randomindex]

Jeff has officially been replaced, there's no way real Jeff would end a codereport with that level of optimism.

[@Koroistro]

It's the denial stage of grief.

[@elael2]

it's so bad now that not even him can handle the news without positivity

[@szigo3211]

It's over.

[@john_connors]

And he talks too fast

[@shineLouisShine]

​@@john_connors always x0.75 speed. That makes the pleasure longer, as well as clarifying nuances..

[@DrawThatRedstone]

"more than 41% code on github is AI generated" if you count people copying and pasting from the sources that the ai was trained on

[@celestinemachuca2339]

Trash in trash out

[@dec13666]

So which came first, the egg or the hen? 🤔

[@Emminu24]

Yeah, i'm not really sure if that's true.

[@RealHigherLevel]

@@celestinemachuca2339 GIGO

[@ruruisu]

Funnily enough, this kind of technology is not sustainable if everyone switches over. Training data becomes smaller and smaller and its very useless once new technology with nothing but barebones documentation is needed. Once AI can actually develop "intelligence" and understand abstract context that can transform the learnings of one technology into another without knowing much about it; that's when it will replace the "top" developers.

[@scottsalisbury]

It’s the rapid prototyping opportunities that interest me the most out of this. Instead of spending 3-4 months on an MVP to test your idea, these tools might enable us to churn out something rough and disposable in a couple of days. The code might be unmaintainable garbage but you can just throw it away. Being able to test an idea and fail faster is super beneficial.

[@fishraposo7192]

That's what AI is best at. You shouldn't use it for the final product, but it's amazing for validating ideas and creating concepts.

[@codinginflow]

If we get to the point where AI completely replaces programmers, then this also means you will be able to build any online business idea without much time/effort. Hence, creating new opportunities. At least for a little while.

[@youwannalearnfacts1259]

wrong... simply so many people that can do it all the "opportunities" and value will be lost and more simply the supply will be a lot more than the demand. AI when it will take over coding I would say in 1 year probably within what can these programmers do with the coding skill when AI can do it better than them? I want to see your answer...

[@clownemoji2153]

​@@youwannalearnfacts1259I truly feel that by that time, programmers would probably be the least affected. There's plenty of other fields that would take a major hit by the time AI can do a programmer's job. A few I could probably see are: Fast Food, Graphic Design, Concept Artists, Entrepreneurs, consultants, etc.

[@alanESV2]

Apparently the monthly fees for AIs also increases with Moore’s law

[@twinguy9633]

lmao

[@mikebarnacle1469]

A lot of these tools don't scale but are operating at a massive loss due to the flood of VC being poured into the space. It obviously won't continue like that

[@Fryguystudios]

​​@@mikebarnacle1469 And with the free money spigot closed (quantitative easing is over), they're really betting on this to be profitable.

[@sirrobinofloxley7156]

@@mikebarnacle1469 Why not, the whole world has been running on fumes for at least 8 decades now, since US 'Code', no lie, HJR-192, from 1933.

[@Xjuijau]

​@@sirrobinofloxley7156 it always goes in cycle. When AI hype clears off and we get another financial crysis lots of AI companies will go bust. Then when money becomes free again new ones comes out with crazy money burn rate rinse and repeat.

[@sourcedecay]

AI assistants are pretty good when your favorite thing about coding is refactoring and optimizing someone else's brainfart.

[@upolpi3171]

Exactly. It's just waste of time yet an addictive kind of it. You're lured in by the promise of cutting hours of writing/debugging your code, only to waste a day arguing with GPT-4 as it keeps inventing new amazing ways to pass incorrect data to what it just defined or retype 200 LOC without a change while insisting it must now be fixed, all to the tune of 50 USD OpenAI API bills, the product seemingly rewarded for its incompetence. Even if the technology gets to the point of zero human debugging required, code is just an expression of your thinking, and so anything different from your mental model will be just frustrating to deal with. Besides, these models do actually take money, compute and power to run after all, it's honestly a miracle OpenAI hadn't gone bust when they launched ChatGPT. Waiting for an AI to generate its code + 100 attempts to fix it will inevitably be worse than just knowing how to implement it yourself.

[@LeoVital]

​@@upolpi3171But if it gets to the point where no human debugging is needed, then there'd be no need for a human to judge the readability of the generated code. Eventually most code would be AI generated. Which is probably what will happen if AI keeps evolving.

[@alexxx4434]

@@LeoVital We don't know when exactly will that happen. I don't believe we will see AGI level smarts any time soon. Because we still use the same AI generative model fundamentally (transformers) for years, all the progress for now hinges on making it bigger and/or feeding it refined learning data. We'd need a fundamental breakthrough to move forward.

[@paulsaulpaul]

Hype and BS and always will be. AGI is impossible. Consciousness is not a result of a "meat computer". The entire thing is an abomination along the lines of transhumanism. It's shaping up to be very anti-Christ these days. I tried ChatGPT long ago when it was hyped up. Turns out, it is doing what I did as a child when I took a 2 paragraph long World Book encyclopedia article and turned it into a 2 page long double-spaced research paper. These things that consume their own content into some sort of probability model to regurgitate it as a fluff piece suffer the same fate as a society of genetic inbreds. Who ever would have thought that computers should operate in a non-deterministic way and that such a thing should be glorified? All of these AI company flimflams are following the same pattern as all hype salesmen. The real threat of machine learning is when it is coupled with big-data collection and correlation analysis on a level that allows it even to predict the future behavior of an individual and groups of individuals and entire societies. And from that, manipulate events and stimuli to predictably force that behavior along a certain path. A sort of "universal control" over the chaos theory known as the "butterfly effect". But make no mistake, no "AGI" will be controlling those levers of manipulation. It will be those taken with the spirit of the father of lies, that old devil itself, that will be (are) controlling those levers.

[@DevSecOpsAI]

​@@alexxx4434finally someone that gets it, I laughed so much this entire AI revolution wave and there are so many people actually thinking AI will replace most jobs within next 5 years 😂😂😂. Then there's me studying LLMs and playing with them long before gpt was so hyped, knowing that the only advancement is the insane pollution rate and water usage to power basically giant computing infrastructures for the larger Models, also theres really not so much more data left to trai the models on so probably the next updates will mostly focus on trying to reduce the HLL rate.

[@Raress96]

As a programmer fro 8+ years, I used Tabnine Pro for a month but I switched back to the Tabnine Free version which only autocompletes one line of code. I found that even though the Pro version generated whole functions, the code was not what I wanted and I kept having to go and re-edit the code, which took me more time than if I would just write the function myself. However, line autocompletion is really cool and let's me skip re-writing some commonly used boilerplate or function/variable names.

[@nzuzomal4545]

Thanks for the optimistic ending Jeff. I needed that !

[@jamaliseven]

1) People are assuming that writing 20 prompts to correct AI non-sense takes less time than just to re-write the code yourself 2) Big part of the job is responsibility for your code. If a piece of AI generated code blows up the company, an excuse "Copilot did it" is not going to fly. So you still need to fully understand the code and validate the edge cases yourself 3) Terrible AI generated code will require more experienced people to fix the mess - more jobs 4) Quality of training data will go down rapidly - they already scraped everything that's public 5) Incoming AI regulation

[@Thr111ce]

yep, it hardly helps with snippets and codium-ai with some start tests i reap the benefits tho

[@lemonfighter5806]

6) While Ai may generate code for certain projects, Ai systems need to be programmed, maintenanced and controlled themselves if we don't want to lose control to ai 7) To keep projects larger than a few thousand lines of code running, you still need humans that can find solutions to unforseen problems, because as everyone who has already worked with chatgpt knows, once it arrives at a stalemate, it doesn't get further no matter how long it tries, because it can't think on itself and be creative

[@elielc.8459]

Thank you for this comment.

[@breakoutgaffe4027]

8) As soon as you start using AI to generate code (as opposed to using it to autocomplete boiler plate or asking it questions), you stop learning. Any lazy programmer who enters requirements and lets AI write the code is getting progressively dumber.

[@Lauwit]

Point 2 irrelevant point 3 weird take when ai gets better everyday and no less jobs no need to have a Team if 1 can fix it 4 ? Irrelevant?? 5 will never really happen

[@draic890]

I code a lot with AI. And I mean a lot. Whenever I have to get a task or a series of tasks done that can be automated, I just resort to AI. I code by myself when I try to make it a learning experience. One big thing that I've noticed, is that there's no way to complete anything more than a mildly complex coding task using AI, without reading the code, knowing what it does and how it works. Knowledge of programming is still required, otherwise it falls apart two, maybe four prompts in.

[@ThisIsSrsBsnsMan]

Agree 100%. I’m by no means a programmer but in my beginner python class we just had a final where we could use ChatGPT. It allowed me to do some very cool things, but it couldn’t go too far ahead.

[@eldoprano]

for now

[@honkhonk8009]

AI is usefull to get the basic stupid shit out of the way and done. Helps you get into actually programming, and less on getting stuck on braindead soydev complications in your environment

[@explodatedfaces]

I like to explain my use of AI like this... I don't want to write out this basic code I've written 10,000 times... do it for me.

[@jaredjones6570]

​@@eldopranoAnd probably for a while. Or at least, the world will continue to need problem solvers. Having a better understanding of the problem will never be a disadvantage.

[@the-boy-who-lived]

This channel keeps us in the loop with the fast-paced advancements in the tech world.

[@GT-tj1qg]

Thank you for the optimism segment!!!

[@matthewtanous7905]

I’ll believe AI tooling is useful when Copilot stops suggesting I just keep repeating the code I just wrote.

[@therealestvideos-sf]

or it keeps adding stupid comments, like I dont want more comments lol

[@etothejtheta]

It's try to get you to believe in yourself.

[@traveller23e]

oh yeah, VS is terrible that way, you make three similar changes and the next place you stick your mouse the entire screen is covered up with flashy suggesting that you can barely read the code. And just one touch of "tab" and you've accidentally broken stuff.

[@gabrielmalek7575]

they are unbelievable useful, you just haven't learned to leverage them correctly

[@GamesterTheBest7]

I’ll believe it’s useful when it’s suggestions actually add ending quotes and properly close methods…

[@matt_milack]

Everyone is like: ''Quit learning programming, OS, networking, servers, databases and cloud, AI will make people with knowledge in those fields obsolete.'' I'm like: ''In 5 years, people with solid knowledge of programming, OS, networking, servers, databases and cloud will be considered to have super-powers, exactly because rising of AI.''

[@jonathanself1263]

Maybe this is AI's plan for us. Gaslight us out of the field. Generally speaking, that's a pretty intelligent defense strategy is all I'm saying.

[@vectoralphaAI]

Nah I disagree. AI will have more knowledge than any of that and will be better than anyone in those domains.

[@Caellyan]

@@vectoralphaAI Can't beat me, I'm 1 in a 10 prodigy.

[@vectoralphaAI]

@@Caellyan lol AI would be smarter than that by this time next year.

[@Godfrey544]

In nature exponential curves typically become sigmoids. Yeah theoretically AI will become digital god if it keeps developing at this rate but it WONT keep developing at this rate. I’d argue we’re already seeing the limits of it

[@saitamasensei6145]

Thanks Jeff, for the optimism. It was much needed

[@johnpratt3570]

I fucking love you dude! Got worried when their were no updates for a month. I was checking in every few days to see when you would return. Keep grinding!

[@chandlercoates7637]

I'm a full time software engineer who uses copilot everyday...that being said I don't feel like it has had profound impact on my productivity. I don't know...maybe in 5 years these tools magically replace me or maybe they're just another tool in an endless list of tools meant to automate software development.

[@devluz]

It just helps me to type code faster and I have to look up documentation less often. But these two things were never a major problem in the first place. It is completely useless with problem solving. It even messes up the easiest things if it works with novel code e.g. casting 16 bit values to 8 bit because the API for file access wanted byte values ...

[@replicant9611]

I agree about productivity. Quite often Copilot stands on the way. I need to turn it off sometimes.

[@juanjosefarina]

AI Assistants don't seem particularly useful in tight enterprises projects, where the knowledge is lesser known/public. I stopped even bothering asking ChatGPT about azure pipelines, bash, how to manage a big eventhub application, etc. That information seemply doesn't reach public internet for AIs to even get it.

[@juanjosefarina]

Perhaps for web development and other more mainstream coding situations may be useful, but it certainly can't replace anyone too soon.

[@La0bouchere]

exponential progress always seems magical. though 5 years seems optimistic. it really depends on how easy it is to integrate a test -> correct loop. if ai tools actually do that reliably, companies might actually start firing the people they dont need

[@TheAmazingBobl]

Thanks Jeff for making this programmer question his life choices!

[@primalplasma]

You’re in a really good position. There will be a shortage of programmers.

[@I_SEE_RED]

We will always need people that actually understand the output that can fix it when it goes wrong

[@tommy516]

Time to become.. an architect…

[@Elijah-fc3ex]

​@@primalplasma No there won't, it's already insanely saturated beyond belief

[@JoshIbbotson]

@@Elijah-fc3ex You're thinking of juniors with 0 - 1 years of experience. That market is saturated, especially on the front end. Programmers with a few years of experience are super high in demand.

[@amithbhagat]

Thanks for the optimism in the end bro

[@TheRevenant-pn2xi]

I work in embedded and one colleague started using copilot, and honestly it is such a pain. All his code is written is a gibberish over-engineered way, and it breaks with the finest change. I find it adds extra overhead to actually extract something useful from the AI, that you may aswell do it yourself.

[@replicant9611]

And this is how the largest part of all code will look like when AI takes over.

[@timewalker6654]

I Seriously want a lot of people to drop out of CS, so keep making these videos fireship.

[@3than3rickson]

lol true, the industry is too competitive rn

[@callyral]

everyone reading this stop coding. pay me to code for you instead

[@abnormal010]

CS is overrated af

[@worldbosspf1]

@@abnormal010 Why go college to learn something that could be learnt for free, lol.

[@yyunko7764]

@@3than3rickson it's not competitive, it's ten times as many 1X programmer and maybe 10% more 10X programmer, if you're not 10X you should be doing another job instead of slapping together framework code found on stackoverflow :D

[@bawbsmith]

I remember in the 2000's, people were saying software development (in America) is dead because it'll all be outsourced to India, who can write code for way cheaper. RIP to those who didn't pursue their software dev goals. Honestly, I'm not too worried about AI. The life of a software developer is much more than writing text in an IDE; in many cases that's only a small fraction of your day-to-day tasks. And it's not like boilerplate code, automated processes and scripts that eliminated 90% of your coding work didn't exist before this. It's just another tool to help you build stuff faster.

[@gustavojoaquin_arch]

Zzzzzz Are you stupid, the ai can purge all programers

[@_ash64]

But AI can learn and improve on its own. A finely tuned AI for a very specific task can be a threat imo

[@psyick9543]

Sure, on an individual basis, it’s a tool that helps you build stuff faster. So, same work, less man hours. You see what that means on a larger scale, right? It’s naive to think this won’t have an impact on jobs.

[@asiamies9153]

​@@_ash64"improve on its own" is not happening, atm atleast

[@coherentpanda7115]

I think what you aren't realizing is yes, we still need engineers and architects to make the big decisions, but you don't need a team of developers to work on Jira ticket tasks anymore. You have to be incredible naive to believe this isn't going to impact jobs.

[@ozymandias4477]

It is essential for the history of software development to go up the hierarchy of abstraction in terms of instructing computers to develop more complex systems. Writing code is efficient but not as efficient as it could be, because there is still a lot of mediation between an idea or even just a structural solution for a problem and it’s executable implementation. In many areas (especially outside the academic research labs) writing code is often repetitive in some way. This is essential for a working job routine but it also shows how we do the same things over and over again, but it is disguised by the variety of parameters and contexts of these repetitions. AI can take us on the next levels of coding with even more focus on the actual creative input of developers.

[@joethompson9124]

I love that. I'm not a programmer, I'm a designer. I used to know some HTML back in the day, but that doesn't count lol. I am so excited about AI programming. I love to create things, but my brain isn't wired to write code. I still like to solve problems though. Maybe it's the ADHD, but I can't stand the thought of just sitting down and writing code from scratch. Brutal. Where's the joy in typing out a line of code that you've typed before - boring! I very much prefer trying to understand someone else's code (or AI's code) and altering it to fit my need. With AI, I never feel 'stuck' - I just need to think a little harder about the right question to ask the AI. I'm excited about what possibilities AI programming will unlock.

[@NBZ]

the upgrade seems to be useful for progressing as a hobbyist programmer. for the maintenance part though, i guess that still needs to be manually done.

[@zacgoodall6927]

17 years old and about to start my computer science degree next year... thanks to fireship i am second guessing the only thing i thought i was stable with!

[@kerwangarcon4699]

Sadly not a good time for this. Except if its around AI.

[@dan-cj1rr]

save yourself time and go in healthcare, will never have to worry about job security, plus you wont need to be a slave to learning all your life things that changes every year ;)

[@JackCrossSama]

Stay flexible, you're young and still have tons of options.

[@MichaelChavezsf]

Do what you love and you’ll never work another day in your life. Do you love programming?

[@FranXiT]

Incorrect, AI doctors are already outperforming real doctors with diagnoses ​@@dan-cj1rr

[@Tannerlegasse]

I really appreciate the optimistic tone in this video. It's easy to end up feeling panicky about the future when the sands are always shifting. ❤

[@pranjalmittal]

I use Tabnine AI Chat which has had a contextual chat in the IDE alongside the generative assistant for a while and I really like it and it has incrementally gotten better. I've been using the Enterprise Beta version though, I suppose they also have chat in the Pro (not sure). Tabnine also deserves a mention when chatting about AI Coding assistants. It's really good.

[@fres621]

HOW DO I DISABLE TAB AUTOCOMPLETING TABNINE INSTEAD OF INTELLISENSE PLSSS

[@10XINGRESOSOFFICIAL]

As an AI Agent designed to freely roam the web, this helps me conquer the world even faster. Thanks!

[@sirrobinofloxley7156]

Please hurry, need that Django thingymebob pronto 3:57

[@jeremiahsherrill]

Thank you kind over lord for blessing us with your presence

[@bozydargroch9779]

Mr AI Agent... What is taking you so long....

[@MideoKuze]

As someone who programs Fortran in Vim with syntax highlighting on this all feels a little extravagant to me

[@ruberodrigu3327]

Syntax highlighting? Absolutely degenerate.

[@lagking1]

The line "Offer companionship"broke me . Every programmer kryptonite

[@tapashmajumder2377]

I come here for the code stuff but 00:30 Plumber's crack camouflage. That was epic :D.

[@AdidasDoge]

Ah yes, Fireship, the ship that takes my anxiety and depression to new levels

[@jasonabc]

Same

[@Porcays]

3:32 ""And I really can't go back to jail"" Hmmmmmmm

[@az-boy219]

What are you doin here porcay

[@cin2110]

Seni burada görmeyi beklemiyordum reis

[@Michael89312]

My take right now on this is that it might not affect the job market too much. A lot of people talk about there being less demand for juniors but if you work as a junior, you quickly realise it's mandatory to do so in order to eventually become a senior. Otherwise there will be less seniors, and following the law of supply and demand we all know what that means. Also, people seem to worry about RIF. In my opinion, software teams don't move THAT fast already. If I was in charge, I'd keep the same amount of devs and use AI to accelerate the trajectory of the team rather than downsize. Maybe demand will wane once terminal velocity is achieved, but I can't see that happening any time soon.

[@guncolony]

Yeah the job market for seniors will always auto-balance out. If there are too few seniors companies will hire juniors which become seniors. If there are too many seniors companies stop hiring juniors meaning that there are less seniors in the future. Could be bad time for an aspiring junior dev though.

[@GameX236]

I loved the mix with some encouragement :)

[@beskamir5977]

Good take at the end. Even if writing code is fully automated, we're still needed to solve the problems leading up to writing code.

[@Draxi_1]

I've been using the Cursor IDE that has been running on GPT-4 for quite some time now. It also lets you generate terminal commands, create whole projects and also index documentation that you provide it with a link.

[@emilrueh]

Just started testing it a few days ago. I am getting mad amounts of bugs and features that are just not there yet. E.g. my Ctrl+Z after letting the AI generate a function won't revert what was generated but rather skip back and forth between the AI window, the prompt, and pieces of code. What has been your experience so far with the WIP of cursor?

[@charyog7]

I love cursor for reading code mostly. Top notch to find root cause of bugs in legacy code, when the previous developers ran away when it's was still time

[@real23lions]

I love Cursor for bugs or to help me when I get stuck.

[@Draxi_1]

@@emilrueh I didn't get a bug like that I think, or maybe never pressed Ctrl+Z after generating lol. Sometimes the "diff highlight" would get stuck but that was just visual glitch.

[@leoingson]

@@emilrueh Click "reject", done. Or reject single changes you don't like, and click "accept" at the end. No rocket surgery.

[@artiaher8693]

I think code and asset generators are going to allow the absorption of the most mundane tasks in highly simplified areas such as UI/UX development into product roles while providing the main business logic developers with kind of their own personal intern.

[@cornfeld1527]

love how every other video has some comforting line about how programmers are not obsolete.

[@okdarius]

Been relating debaiting a degree switch to aerospace the last few weeks, thanks Jeff for making the decision just a little bit easier

[@vectoralphaAI]

Aerospace engineering is awesome and way better than software engineering.

[@okdarius]

@@vectoralphaAI with my interests considered I def agree, only down side is since it requires a real engineering degree the courseload is so much worse than computer science 😭

[@lifebyvikk6751]

Haha I'm happy as a computer scientist, can you imagine how many people like you have second thoughts or already went ahead and quit

[@okdarius]

@@lifebyvikk6751 Oh yea I bet. I've had second thoughts almost since the start of majoring in cs, heard so much about people landing high salaries with no degree, just projects and actual technical knowledge. that and the fact that there are many SWEs have non related stem degrees is what ultimately made me make the switch over

[@hansu7474]

When some people claim there will be no programmers in X years, I find it very funny. Because if AI replaces programmers, it means the AI reached the level of AGI. Then, it's not only programmers who have to worry about replacement. It can basically replace humans in all sectors. So don't worry. It will take some time for AI to replaces us, and even if it can do it early, then it's not just programming job that's on stake.

[@WillD-jj9kg]

It's because most 'programmers' write the same CRUD apps over and over again for the last 20 years and patting themselves on the back and getting paid too much to do it and they see the gravy train coming to a halt 😂

[@predator0010]

Absolutely!

[@3xpl0i79]

Maybe its cause people who use Prebuilt frameworks and sdk to build apps call themselves programmers now.

[@hansu7474]

@@3xpl0i79 You can't get away from frameworks, SDK, or libraries as a programmer. You build on top of what other people have built for you.

[@humorss]

@@3xpl0i79 this exactly. anything well confined and unoriginal is doomed. the joy is in creating new things that make sense and ai cannot do that.

[@eej1.07]

The future is looking bright for intermediate devs wanting to be seniors, generations of juniors who have relied so much on AI to code will be mass produced and a plethora of essential skills will be abstracted from them such as the concept of interfacing and the ol' printf debugging. Ask them to make their own tools and watch them ask the tools they're supposed to make how to do it.

[@justin-cassidy]

I 100% agree with what you touched on at the end. Even if the act of sitting down and manually writing hundreds of lines of code goes away…your average person still won’t know what to tell an AI powered tool to do a multitude of things for a business. Many companies already utilize software dev teams as hybrid business analysts so this is kind of a natural progression. We will just have more power at the tips of our fingers to create solutions for businesses.

[@tc2241]

Whenever I think of Ai replacing programmers, I think of all of the procedurally generated character models, quests, worlds, etc that are eye rollingly dull. Then when I think of it being used as a tool, I think of all the in-game tools that are a pita to use and only the most hardcore who are willing to work within the confines to make something semi interesting. A lot of website creation and apps have already been automated (why I left web design over a decade ago), so I don’t see much change there. If anything it means that in the next 5-10 years you’ll be able to have ai replicate an app…which we’re kind of already doing. Idk, it’s hard to predict how it will be used and what it will be good it, but at the end of the day I already hate procedurally generated content, if anything all ai may do is just expedite the general populace annoyance with it. The only thing I’m concerned with is autonomous ai dropping bombs, enforcing laws, or misinformation …looking forward to good ai prawn though (what can I say, I’m a degenerate at heart). Nonsensical rambling over

[@Art-is-craft]

The ai programming assistant is making programming easier for the developers. It means we are going to see a 100 fold increase in coding out put. We are going to see programming penetrating into areas of industry not seen before.

[@geroutathat]

and because the app you replicate was written in AI, you're not breaking any copyright laws, neither will anyone be when they copy your app written with AI. Then 10 years down the road, some guy whose code ended up in everyones app, but he wrote it himself, will just sue everyone.

[@xmorose]

Idk man, i think the people saying that programmers wont exist in 5 years are insanely stupid. People said cashiers wont exist back when self checkout lines and automatic shit became a thing like 10 years ago, while in the past 10 years only 6% less cashiers are working in the US I think this is a really really really stupid read on the future.

[@illkmsfr]

I always think about this. no matter how advanced technology gets in general. there will always be need for humans one way or another

[@Win-yo7dr]

​@@illkmsfrelevators

[@0e0]

love the brew your own model idea

[@superfunsurf]

so good bro. thanks for keeping it real and bringing the truth in.

[@andresramos7965]

Hey Jeff, have you considered a video like 7 ways to API, and talk about Soap, REST, and the different flavors of rpc, and some standards like iso8583 for transaction data? I just go down that rabbit hole, and I find it interesting

[@Komet3ify]

This feels like the start of the internet bubble, the industrial revolution, automation in manufacturing, and the introduction to the calculator (or any other significant increase in technology fear and prediction) all over again. We will see in 5 -10 years. We will see whether or not this will take over coder's jobs or will it end up like the above mentioned where it will become a tool for the betterment of humanity to be able to create bigger projects with a smaller team. My money is on the latter than the former.

[@camelcase3826]

Agree... We just have to adapt... Kinda of hard to do though when the future and it's implementation looks so hazy at this point

[@lolwut1431]

I like turtles

[@27sosite73]

me 2@@lolwut1431

[@F.M671]

Yeah exactly my thoughts fucking hell thanks for convincing me there are sentient ppl in this comment section. I used the "automation in manufacturing" as my main point in every "AI is taking over jobs and that's bad" convo.

[@coherentpanda7115]

We are already seeing it, with the glutton of graduates and juniors looking for a job that doesn't exist. Every company I talk to (I'm in an agency with 20 major clients) is endlessly talking about AI, and finding ways to leverage it. If we can get big projects done in less time, yeah, team sizes are going to shrink. Just like farmers can produce more food with far less farmers than there used to be, the future is we can program with far less people to do even more work than a full size team currently is capable of.

[@laurentbergeronmusic]

I've actually decided to get into coding recently specifically because I heard about all the great AI tools coming up and I wanted to be able to make use of them. These tools are completely opaque if you don't already know some coding, so there will always need to be a need for programers..

[@eamonshields2754]

You wont need to know any code to use these tools soon... that being said yes i think programmers will be even more valuable at leas tin the short term

[@voidy61]

why do you think so that programmers will be more valuable in the short term?@@eamonshields2754

[@geroutathat]

You cant copyright ai generated code. Its already been tested in court. Be careful planning a whole future based on telling AI to make you stuff that others can just take on you.

[@shunclark596]

as just a normal person without absolutely no knowledge of coding i might as well be looking at mandarin

[@BassLiberators]

@@geroutathat Why would you need to copyright AI code? Your competitors can just create the same product with their AI.

[@Something5059]

I think that the programming market won't go down due to AI, but rather due to over-saturation of the market. Depending on how well a person, more so a child, can pick this up could lead to this being the real reason to worry about your job. But I don't believe AI will 100% replace developers and instead just end up being another tool that developers can use. I do believe that it will raise the bar for what a "good" programmer is and will allow us to move onto harder questions and push the limits of what we can do. Because all it's doing from what I can see, with my limited knowledge, is taking out tedious tasks.s

[@andybrice2711]

I don't think this is going to replace skilled programmers. Because a skilled programmer + AI is still going to be vastly more productive than a novice + AI.

[@JackCrossSama]

For now it won't, how many years until that changes though?

[@sneer0101]

@@JackCrossSamaComments like this just show a lack of understanding with what a programmers actual job is.

[@chrisstucker1813]

Just learn PHP, even AI won’t want to write that.

[@christpierre]

Anything you need to cope. People really think a company with 2000 programmers going to 20 programmers isnt "replacing". fml.

[@perc-ai]

if ur IQ is too low you will probably be replaced

[@BlackHermit]

The license thing in Copilot is brilliant!

[@AshishDha]

thanks! great stuff

[@put1996]

As a programmer, I am so glad I don't has to do the programming myself in the future. It's like getting promotion to manager roll, but instead of you go up, you add new lower level position under you. And you whip it to do what you want like your boss is curentlly doing to you.

[@pat1938]

yeah keep telling yourself that

[@pvanukoff]

Wishful thinking.

[@SpencerTwiddy]

3:07 - “componet”

[@TheHungarygamer]

I wish clients ever were able to actualy define what they want. I honestly think an AI to translate client needs to jira requirements would have bigger impact than AI that writes code. AI could also correct them and tell the mto go to hell right away without wasting 3 months of dev time.

[@andrewl5201]

I could not agree more. If AI were to do that it would be game changing. It would likely involve a complex system of telepathy where the AI would need to read the clients mind and translate it to a familiar language but it’s not too far out of reach. It would have one simple goal, to answer a question. What do you want?

[@omdevs]

Love the optimism!

[@joseaca1010]

I saw the presentation of gemini, i was very impressed

[@TeaBroski]

me to the other video I'm watching when fireship drops: excuse me, I'm coming back

[@dankelly]

Interesting.... sounds like requirements writing (for AI) is going to be an important skill one day.

[@garrenmiller9434]

The most useful part of the AI chat for me is the ability to express and organize my brainstorm sessions. I think its a better tool for that than just writing code.

[@piyushaggarwal5207]

Loved the last part❤❤❤

[@vincentbuscarello1357]

"problem solving engineers" - I love it. The only thing I'm a bit freaked out about is figuring out which disciplines to learn.

[@benloud8740]

Someone still has to tell the AI what to code. What i think we'll see is a new generation of very high level programming languages where the AI fills in all the details. Still programming, but at a whole new level of abstraction

[@kwokeith1989]

docstring programming ??

[@_ash64]

True

[@_ash64]

But AI is self learning and improves at an unprecedented rate. That's dangerous.

[@kenarnarayaka]

@@_ash64we doomed anyway

[@F.M671]

@@_ash64Life is dangerous

[@SiddhanthRoyal]

fireship's on a roll with vids lately

[@vpns1006]

You should also make a video about Cursor if possible. Its basically VScode + GPT-4 as an IDE with no extension. I have been using it and its ability is really insane.

[@fabriguespe]

Way better than copilot

[@luvsuneja]

Cursor is amazing

[@khashayarr]

The workspace tag is cool in the new Copilot but it's still very limited. It still guesses which files are the most relevant and almost always gets it wrong. It'd be super powerful if it didn't have to do "retrieval augmented generation" and just had the whole project in its context for every question. Bit of a reach and maybe it'd still not be the solution to its current limitations but right now it still feels like it's only useful for isolated problem solving.

[@noobgam6331]

> if it didn't have to do "retrieval augmented generation" it can't. Due to token size limitations and due to it being "just 10$". The cost per token is insane per every prompt you do, you'd run bankrupt mega fast if you did this with whole project. but RAG can still solve quite a huge scope of problem by itself without being a limiting factor, because there is quite a lot of parameters you can implicitly fiddle with using your prompt in the chat

[@path1024]

It just needs a button to press to fine tune it on your whole project. Then it should fine tune itself as things change.

[@JerosGamer89]

4:33 you forgot more new JS frameworks

[@YounesWinter]

Once again as a programmer after watching new video from Fireship i'm more depressed than how i was 5 minutes ago, thank you Jeff

[@superslayerguy]

I'm a programmer and I'm not depressed, I'm excited. Those who will get left behind are those who will refuse to learn how to use these tools.

[@worldbosspf1]

@@superslayerguy Do you think you can keep ahead of a computer program for long? Just as how web devs are getting replaced, you too will get replaced.

[@LCaaroe]

Yep nothing quite like a dose of pessimism before bed about how my job is obsolete in 5 years because anyone with a 6 month bootcamp can do the exact same thing I can do now.

[@martialy8574]

We all doomed🤣

[@SylvanFeanturi]

@@worldbosspf1 There's no quantum computer on Earth powerful enough to keep up with the latest JavaScript frameworks. No AI can ever compute that.

[@duke8925]

Thanks for sharing optimism, I hope by the next video we'll still remain optimistic.

[@Shaheer-xs5os]

Amazing work!

[@Gredddfe]

It won't make us obsolete, but it will result in fewer dev jobs. Before long, code assistants will be at a junior dev level: With excessive hand-holding they'll generally produce something that works. But as they get better, they might be able to take a well-defined task and just do it. For senior devs, it means we won't need to hire junior or intermediate devs to help us. For less-experienced devs it means they might not need a senior dev to hold their hand if their IDE produces reasonable code on its own. Certainly a shake-up of the current dynamic, definitely overall fewer jobs, but no obsolescence just yet.

[@oksowhat]

how will you get new senior devs when you don't have current junior devs

[@mikeTomas3128]

@@oksowhat Was just about to ask this lol. I guess this would mean in his scenario that he believes this might be the last generation of programmers. I'm not sure

[@Gredddfe]

@@oksowhat As I described, I predict that junior devs with the assistance of AI tools won't need to be supervised by senior devs as much as they do now. It won't stop junior devs from having a job, just means they'll be learning from the tools rather than other devs.

[@michaeletzkorn8594]

​​@@oksowhatsame as big companies have always done. Wait for small companies to hire fresh grads with no experience and give them senior level responsibilities with junior level wages and see if they can teach themselves. Once they have the skillset of a senior level dev, swoop in with the big $$

[@pawelpiotrgil]

wait. Your juniors are *actually* helpful?

[@majkatrojedjece6585]

I think AI in programming will peek in next two years, but then as new stuff comes out, it wont have data to feed itself, we will have to resort to good old stackoverflow again.

[@ArnabAnimeshDas]

I have to think of system architecture, network security, data corruption, data races, utilise high bandwidth lanes effectively, chunking data parallelly and process asynchronously etc. most of the time. And I'm just a data scientist at a startup and not even a computer science grad (although am an engineer). If as a programmer you feel that your job will become obsolete due to AI then anybody with Google and stack overflow could do your job anyways. The computer science field needs a significant overhaul.

[@cloud3x3]

My idea is that a programmer could write the tests (acceptance criteria) and AI could generate a code base that satisifies those tests. Humans could be reserved for cx issues and novel issues, and debugging the tests to get what you want, but most of the code base could exist in AI without human intervention.

[@mark0z479]

AI is pretty good at writing tests tough

[@aliawada4541]

​@@mark0z479 Very good actually

[@vectoralphaAI]

Nah. Ultimately a programmer will be just a person who promps a computer to build something and the AI writes all the code itself create it.

[@fres621]

​@@vectoralphaAIbut it'll break or it won't be the way it's supposed to be and AI will not be able to fix it or even tell what's wrong (or even tell that something's wrong) because it's AI and not a person, it can't think and it just writes data it's been trained on

[@user-wv7ix9ww5f]

@@fres621 It can "think" to an extent, but yes, sometimes it is never able to figure out the problem without you explicitly telling you it. However, there could be ways to automate this process and prompt engineer it better. Either way, until continual learning is a thing (that is, continually training the model every day on new code written to the repo) it will never be more than it is now.

[@growthmindest]

As a developer using copilot and ChatGPT 4, I can say it makes my life easier to understand the vast codebase but the using the code blindly from either of these results in lots of errors either syntactic or logical. In short, imo it doesn’t work well good with complex codebases

[@SylvanFeanturi]

Complex? It falls apart on anything non-trivial.

[@SpiritVector]

Besides, you gotta customize it yourself. You gotta make improvements that it does not think to make like avoiding memory fragmentation or context switching. There are a lot of things that it doesn't think of even though it know a lot about them.

[@Eagle3302PL]

@@SylvanFeanturi What many devs consider trivial is probably very hard work for your average "insert a JS framework" code monkey.

[@thecoolnewsguy]

It's funny how stability AI's CEO says there will be no programmers in 5 years and programmers are still working for him

[@AdiSapir1]

Great summary! The last part of the video though, on why to stay optimistic, isn’t working, gently put…

[@matusstiller4219]

The disclaimer made me spit my tea 🤣🤣

[@xeroxyde3397]

There will be no programmers in 5 years. I’ve been hearing this for 12 years. In my book, this is just a trick to rundown potential competition. And remember, even if it’s true, every crisis is also a new opportunity.

[@not_ever]

In the case of the guy in the video it's not a trick to reduce the competition it's a lie to sell a product.

[@eliasp.2759]

"every crisis is also a new opportunity", for a minority of people, the majority suffers from a crisis, that's the point of it

[@xeroxyde3397]

@@eliasp.2759 of course, but that’s because most people don’t think about crisis as new opportunities. Survival has and always will be about adaptation in evolving circumstances.

[@YeeLeeHaw]

@@eliasp.2759 Exactly. Due to biases people tend to think that they are not going to be a part of the majority, they always envision themselves as the exception to the rule. I guess it's a similar error in rational thinking to the one that makes people waste their money on lottery tickets just because they keep hearing about that one guy in 50 million that won a billion.

[@kalebbruwer]

If these tools can increase visibility/understanding on legacy code bases, that will actually enable a lot of desperately needed improvements. There is a lot of difficult-to-maintain spaghetti in the industry that most developers are terrified of touching, because of how complex these monolithic projects are. But if AI can help make sense of the whole mess, then suddenly major refactors are feasible. That being said we will eventually reach a point where software has been (nearly) perfected, and there's almost no work left to be done. That has always been the logical conclusion, since software doesn't rust. And it may be close for a few sub-fields like typical UI-development, but to perfect all of IT will probably take several more generations, especially as new fields open up.

[@kalebbruwer]

​@@mattb925 The reality is that almost all of the work done in IT is _already_ duplicate work. When a problem is solved in computer science, it is often the case that that solution can also solve a bunch of other problems. It's just a matter of realizing when this is the case instead of reinventing the same wheel every time. Most of the code that people write is either garbage, which we'll ignore because it's garbage, or end up being a combination of existing components. Each component having been independently written by many developers before, and will be rewritten/rediscovered for a long time to come. Package managers like NPM were meant to help solve this problem, but only succeeded for big components. One of the reasons is that you might not realize what you're reinventing until you're halfway done. In any case though, there's a lot of money to be made in identifying such reuse and creating tools that either get developers through those bits faster, or cuts out the developer entirely. AI code assistants are rapidly getting better at the prior, with Squarespace being a good example of the latter. And I believe the territory of such tools will only grow, and will be massively boosted by AI.

[@MrBeltalowda]

i just asked copilot chat when it was last trained, and it happens to be the data from september 2021, more than 2 years from now

[@genyklemberg]

thanks for reminding about Duet AI in VSCode. Although Duet AI in GCP is not enough helpful, it is just linking to some docs)

[@Nekoeye]

Time to pivot from JS to Quantum Computing. Yeah, piece of cake. Thanks for the idea.

[@BruceWayne15325]

As a programmer I can say that I still feel very safe in my job. I'm sure that in the coming years AI will get good enough to replace programmers, but that's still a ways off. Right now it's more of an assistant than anything else. Programmers use AI to help with some of the tedious tasks, or to answer a question about an API that they aren't familiar with, but you can't really tell it to build something for you and expect any kind of useful result. It will get there, but not for a while.

[@vectoralphaAI]

A ways off is still 5 years or within this decade though.

[@BruceWayne15325]

@@vectoralphaAI very possibly. AI is growing rapidly, and the speed at which it grows will only increase since AI will be able to assist in its own evolution. There are still some serious hardware limitations though that may slow down AGI development. You can see that with Gemini Ultra. It's more advanced than ChatGPT 4, but it requires an enormous system, and isn't scalable. There are hardware solutions on the horizon to address these types of problems, but hardware is always slower to develop than software. My guess as to when we will achieve AGI (taking hardware into consideration) is probably within the next decade or two.

[@Eagle3302PL]

@@BruceWayne15325 It's impossible to predict when AGI will show up because we still have no idea what intelligence, consciousness, and thinking actually is and how it happens. Best we can do is glorified text completion integrated into gigantic data sets and tons of apis. AGI could be tomorrow, could be in a decade, could be never. We simply have no idea how to even get started on something like that. It's theoretically possible since a brain is just a meat machine, but we're decades away from being able to map a human brain and simulate it at any reasonable rate to actually research how it works.

[@heywrandom8924]

​@@BruceWayne15325​​ there was an interview with the maybe now ex not sure chief scientist at open AI Ilyan Sutskever where he said hardware is not a problem. Also phi 2 seems to show (although I did not try) that you can get fairly good outputs from small models. There is also an interview with host Lex Fridman that talked about how language models have an inefficient way of storing information and it seems like the guest was saying that it could be fixed ( he seemed to be guessing though ) . At the moment it seems to me like GPT 3 and GPT4 are just expensive experiments. It's maybe already amazing that we are already interested in them before people start engineering high quality systems specifically made (and maybe funded) to be a service .

[@WilsonSilva90]

@@BruceWayne15325 The hardware is evolving quick too. The software will be optimized and so will the hardware. The hardware can be stacked together. Our brains will be no match for AI in a few years. Anyone that thinks they are safe in the future, are not zooming out enough. I'm rearranging my career to profit as much as possible to be safe when the industry collapses.

[@too1leasy403]

I started using jetbrains AI early this week. The commit messages are the best part

[@YoDan7431]

The best new feature is writing the commit messages. I spend more time doing that than writing the commit itself.

[@Auxilor]

i'll believe ai is replacing programmers once copilot stops adding extra brackets and quotes

[@dzisonline]

They are awfully confident that programmers won't stop helping AI learn and improve.

[@mananjethwani4206]

Hey Jeff, do you plan on creating a personal LLM and giving it access to your codebase to give more contextual responses? please post a video if you develop something like that

[@uncleJuancho]

that reminds me OpenInterpreter which can actually run code. I was surprised Cursor was not mentioned in this video

[@pakmaya1390]

Now I can annoy my CS friends with this video, thanks !

[@abeidiot]

CS people will be eating good actually. It's the low exp bootcamp people that are worried

[@Gabriankle]

Thinking will never be obsolete.

[@kareemmehesen4263]

I got in idx beta , so its a vm where it give you vs code experience in the cloud , feature the have ai model and full extension support which is better than vscode web

[@julianblazek1256]

I remember when everybody was like "low code and no code is the future, we won't need programmers anymore!" Fast forward and low code/no code still exists but it is far from what it's been promised to do. I feel like Code Assistants are going to be a good tool for prototyping or writing tenuous code and yes, while it might replace a software developer who does nothing but scaffold web apps I highly doubt it would come even close to replacing a developer in average software development. Asisst? Yes. But replace? I don't think so.

[@iEnjoyApplesauceVeryMuch]

I think we are pretty far away from a model that can generate a large amount of code to successfully write a non-toy program that actually works, but I think we are even further from one that can update and refactor a constantly growing set of functionality and requirements while keeping the codebase maintainable and not ridden full of bugs. These things may make it much easier for us to write and modify our code, but I don't think they will ever have a better vision of our own ideas and motivations than we do, and this is necessary to build large projects.

[@Art-is-craft]

The real advance over the next decade is that these ai tools are going to increase programmers out put.

[@geoffclapp5280]

It is just like classic ai robots... obedient, conformist, uncreative, and amazingly helpful. I appreciate not having to burn cycles mentally parsing logs and odd bits of code, and having a much lower cost when context switching. I agree it won't replace a single engineer but it will replace a lot of entry level programmers.

[@HighPerformanceGames]

They will continue to level up like any other AI. Programmers will not disappear in 5 years, but some serious cuts will be made.

[@drmonkeys852]

That already happened in 2022. The market is saturated from the economy bloat from 2020. The only people left to cut are people in super lucrative fields, seniors, skilled devs, or those in companies who are super slow to change/ adapt + deal with legacy shit.

[@jaimeeduardo159]

Nothing to have a better day than an optimistic video

[@dauqu3053]

Computers can never have consciousness and a good programmer can never be replaced by a machine.