55% is a blatant lie there's studies that best AIs can get the code generation even write 16-17% of the time... how does that translate into 55% productivity boost?
one billion people using AI to code... nice technical debt.... half the human population hasnt even used the internet foir the first time... and their talking about billion people coding with coipilot, like riding a bicycle.... even countries where 30 years computers are normacy... only 5% or less are power users... even the people who think they are good computer users arent actually on average level even.... accorrding to the criteria wich have been layed out....
@@tanmaysingewar I was debating that route too. Kinda like a "Hey look guys, this whole article showcasing our new product, was actually written by our product! Isn't that great?!"
@@d2rey596 Arguably, whether it was an LLM or if it was a human writer or team of human writers, they missed the mark. (You didn't ask for this advice, but it may help someone else) One of the best pieces of advice I have been given: "Read your writing aloud, because your brain will automatically catch grammatical errors". The only caveat to that advice: pay special attention to what was actually written. Because your writing can be so "off the mark" that when read, you say a word or phrase that doesn't exist.
FSD - Full Self Developing. When you put your 11 year old nephew on the driver seat of your car and pretend supervising him is less stressful than driving yourself.
If it's not my nephew but some Indian kid I've never met, I'm not sitting in the car and my insurance will take care of any outcome: fuck yeah, let's do a turbocapitalism, baby!
This comment is transcending the traditional social communication medium allowing the viewers to move into a new paradigm of idea sharing and thought provocation beyond the current limitations of the human brain.
I think this blogpost is meant to be appealing to shareholders and business people, that do not understand the process, but would like to cut down on devs
AI is promising the same thing 5GL language promised and failed to deliver. To say I'm sceptical is an understatement. A developer's job isn't to write code, it's to tell a computer what it must do. Changing the language in which the devs describes problems from a 1-4GL to a 5GL or to an AI "natural language" description doesn't change what a dev needs to describe. The hard part of the job is still in specifying things in a way so a computer will not misunderstand.
I really like this way of putting it. Until AI knows better in terms of high level thinking, planning, architecture, future maintainability, etc. it really is just another way to produce steaming piles of code.
There must be a point where natural language description ends up being longer than the actual code, and comes with major organisational problems, such as where do I put "...should have a border radius 3, but not when hovering, or next to a widget, and they should be laid out horizontally...". Inform 7 is an example.
The fact that “mobile friendly” they show with a web version and not iOS or Android apps shows a lot about where those apps are in terms of of gh priorities
Lol what are you talking about. "Mobile friendly" has always mean't a responsively designed web application so it works well on small screens. Mobile friendly does not mean native mobile app.
24:10 I checked the repo: the diff is because the file was checked in as CRLF in the initial commit, but the Copilot Workspace version has been converted to LF. This is actually also a somewhat common mistake. A human developer would be expected to explain this, possibly also create checks to avoid this situation in the future (with an .editorconfig perhaps), while the LLM is just oblivious. This is also one thing that GitHub is terrible at showing. If there are whitespace differences maybe the UI should show me the whitespace being different, just like the last newline in the file? But no, it just looks identical.
I was pretty hyped after seeing this on fireship’s channel. But seeing you dive into the details of what it’s doing wrong in the _example_ code pretty much quashed all my excitement.
The article reads as if they gave a group of junior engineers "focus medication," and told them to write an article, where the sky is the limit. That would explain why some paragraphs are made up of only one sentence 😅
I worked as a translator in China for about 8 years. I'm guessing this article was probably written in Chinese originally. It's not just VC speak, it feels like Chinese VC speak to me. A lot of those sentences that have a lot of words that feel awkward in English actually feel very fluid in written Chinese. I don't think it came from AI beyond partial translation assistance probably.
People's reaction to these technologies crack me up. Instead of being blown away by the rapid pace that things are developing, they look at the current state and capability and do the developer equivalent of "My dog is so stupid! I can beat him at chess 3 out of 4 games, what an idiot dog!"
Look at where things are today, look at where they were a year ago, look at where they were a year before that. Draw a line. Where do you think it will be in a year?
Oh boy! Programming in natural language sounds like a great idea! Because that's exactly what we need. More ambiguity! More fuzzy shit! Consistency and well defined language rules are soooooo 1970! Precise control over program behaviour? Who needs that? I'm 1000% sure nothing could possibly go wrong.
It's interesting to me that most people think that getting started isn't the hardest part. When Theo said staying with it is the hardest part, it clicked for me, but before that, I totally thought getting started was the hardest part. When you don't even know the right questions to be asking, looking back at it feels daunting to me.
Right off the bat, it's way too expensive. Clearly it needs a lot of work, just their sample shows that much. They should open it up for beta testing, let people use it for free until they work out the bugs. Even after that, charging by the hour is a deal breaker. Thank you for the review, Theo. It's far more interesting than enticing.
These posts and tools are made for non-technical managers. GitHub/MS wants to sell them licenses, on the premise that they will revolutionize their company/department.
At this point, AI products are so obviously forced and just capturing the remaining VC money before reaching the tough of despair in the Gartner hype cycle
Many years ago, I read an article about how the word "so" entered everyday English because of the show Friends. I wonder if the same will now happen with the word "delve"
the comparison of supermaven and copilot was not fair at all.... not only did you not test the two on the same completion task, you made a point that copilot did not use tailwind in the solution, but supermaven also did the same right before. If you're going to showcase something you have vested interest in. do it fairly.
Honestly think the language difference is more technical vs non-technical people, were those papers in particular areas? Because when I moved from international relations to applied informatics I was actively told off for this very thing. I wouldn't flinch at 'delve' tbh
I partially agree. The business speak makes me highly skeptical and the output is an embarrassing pile of garbage. But there are enough non-technical "decision makers", who will eat that s**t right up. Incidentally, these are the types of people I'm not going to feel sorry for. Also, it's hard to say if a directional change is the right call. We might be one or two LLM generations plus a few tweaks away from this being amazing, but that's hard to say now. Anyway, the potential benefits are insane. Launching early, however embarrassing the product may be right now, might even be correct. Iterating with customer feedback and collecting training data is just that valuable.
Wtf we live in a world where we are short on professional developers? Literally when every company is laying off half of their professional developers 😂
I read that essay you mentioned was good, 'Mind the Gap', as well as a few others. I know it was 2004 but wow, that was an extremely boneheaded self serving piece that is just working backwards from a desired conclusion with audaciously incorrect claims throughout. Very weak stuff typical of the era.
Why do companies keep saying they need to lower barrier of entry for programmers but on the other hand they fire 60% of their junior devs? If you don't want junior devs, why do you want more junior devs to commit code?
I spent some time trying to write a (much less ambitious) version of this and getting a bot to edit files and do multiple steps autonomously without loosing the plot... project managers just don't realize how hard that is. I wish project managers/marketers could let things be cool for what they can do without over promising.
I was using cody AI , aws toolkit and none of them was faster and better than supermaven thank you for sharing this cool copilot I really appreciate that
Whenever i need to write a function that validates if a url is a valid github url, or check if we're using https, now I can. I can't wait to get my hands on this tool.
15:38 It's stunning what a stupid idea this is and how no one at GitHub seemed to notice or try to stop or change this messaging. We don't need one billion of anything. If anyone thinks software actually matters that much, they need to take a break from computers.
I agree that such tools should run in the IDE you're already using. I agree the demo sucks; however, a configurable multi-agent to help with larger scope code changes that actually works is something I'm looking forwards to. Human in the driving seat is a must.
the model github copilot uses is super not good enough to support this feature, too much hallucination and misunderstanding, i assume this is planned for gpt-5
"As programming in natural language...." 1. Most people are barely able to convey their idea in natural language to begin with, AI can not create anything useful from that as it currently stands without tons of extra context, which brings me to 2. Programming languages are designed to enable expressing ideas in a short and concise formatting, saving the amount of keystrokes necessary to develop a system. Natural language sucks at being concise (try writing an if/else flow in natural language and your preferred programming language, you'll see what I mean).
"The devs that are boosted that hard - they needed the help, let's be real" - Yeah, I'm sure Andrej Karpathy is a huge slacker and desperately needed that help
Somebody in Github got the memo that Devin was on the way, so the quickly set up a team of 50 engineers to make their own version of it. This is their way to justify their employment.
as a guy trying to learn how to code, and i have pretty bad dyslexia id currently say trying to find useful resources for me to learn from is very difficult, i have a diploma in chemistry biology physics and maths being dyslexic isnt a significant problem because i can just use my imagination to understand it and see diagrams and maths which i dont struggle with, but coding is a tough one thats what im excited about the new gpt 4o desktop app whenever they decide to release the windows version, i think using the screen ''streaming'' to chat gpt with the voice is gonna be great for learning hopefully the same way it'll help blind people, just describing what im doing as i go i think will inscribe the language into my head a bit better lol
Ugh, so I just watched the demo video as a bit of a hate watch, and came across something that should be cleared up. The changes of the readme and the contributing were manually added to the to do lists. Copilot did not decide this on it's own. The fact that this wasn't clear in the blog post just shows how badly written the post was. This is githubs demo btw: ua-cam.com/video/L5Xny6yehUg/v-deo.html
It's bullshit like this why I still don't use GH, but especially the thought that it might change your licensing and absorb your codebase without your permission. That was a smooth move editing the article to promote your channel. And I feel called out for both using the word delve and starting sentences with `and`. I delve into different things all the time, and I've never had anyone complain about it, at least not to my face.
Also listening to Theo rant about the quality of the products... Having a couple friends working at M$ that are Principle Devs, hearing them talk about new hires and the quality of people getting hired. One would say "Well it isn't like it was in the old days" and will continue on about why DEI is sh1t... Think, some new hires he encounters he will agree they are pretty smart but is surprised some of the more lower level basic things they have no clue about. Example might be memory management and handling stuff like that and performance, why or when to use a link list, or avl tree, quickorts are good except when...?
Don't get the part of github not having single dev environment solution. The approaches are different, it doesn't make sense to merge everything. Especially if some ideas don't stick, it would be easier to abandon them.
I mean, yeah, looking back makes me thing people simply does only want to dish out content for the heck of it. First they say they hate it, months later, you guys become patrons of it. I mean, I'm one of them, I hated copilot. But once someone said, that I should try the free trial, I myself have fallen in love with it. So yes, we can be really stupid sometimes, lol.
Ah yes, let's throw real world problems at the thing that was only tested against really simple interview questions like "make a function that can tell me if an array contains a negative number". Surely nothing will go wrong. I genuinely hate that everyone goes so crazy over LLMs getting 80%+ accuracy on publicly available coding question sets before they look at what questions are being asked. The datasets are literally open source. Shit, most of them are on huggingface so you don't even need to download a csv.
I'm still surprised by how insanely bad the github mobile app is on android. maybe microsoft needs to push the same requirement as facebook, and make their app devs use android for a while.
I'd rather have a tool to say you forgot to delete the variable on line 50 dummy, now you have a memory leak than an AI that occasionally gives me good code (sometimes). The number of times I've had to backspace out X == null to X is null because the compiler checks to see if the object can be null, is frustrating.
hot take. creativity is a significant part of being a developer, and if you're struggling to know how to start a project, then maybe you need to practice your creative thinking
The "reach" language drives me crazy, but it's hardly unique to copilot. It's been frustrating me for years as I try to look for and learn about tools to help be build things, to judge what will be worth investing my time into. It's ironic how much it all sounds the same, when marketing people write enthusiastic and yet generic buzzword fluff about why their product is great instead of telling me *what it does*. It could be AI, but it doesn't have to be.
I don’t why you hating on that. This is a nice product, pretty much what cursor is doing, but they try to move it to the browser. I agree with the not good separation of products, this should be part of the vscode copilot
Another thing that is TERIBLE about llms here is the fact they scale so poorly on large repos... Like this grows quadraticaly in repo size... like a small repo takes 20 minutes a repo 4 times the size should take 4*4*20=4hours+ ... This is ridiculous
Copilot sucks so bad, it's a total confusing mess ever since they split it into workspace/terminal/vscode areas and then added slash commands it gets autocompletions wrong mixing up camel case and snake case and it still hallucinates simple things like telling you to use npx with some template that doesn't even exist..
I forgot to mention I AM ALSO INVESTED IN MICROSOFT. Even more money than Supermaven, lol.
And they ain't still half as fast as SuperMaven
55% is a blatant lie
there's studies that best AIs can get the code generation even write 16-17% of the time... how does that translate into 55% productivity boost?
one billion people using AI to code... nice technical debt....
half the human population hasnt even used the internet foir the first time...
and their talking about billion people coding with coipilot, like riding a bicycle.... even countries where 30 years computers are normacy... only 5% or less are power users...
even the people who think they are good computer users arent actually on average level even.... accorrding to the criteria wich have been layed out....
I honestly think the entire article was written by an LLM, and then just edited for "correctness" without editing for tone
I think entire project is built by copilot code, and they even forgot to validate it.
@@tanmaysingewar I was debating that route too. Kinda like a "Hey look guys, this whole article showcasing our new product, was actually written by our product! Isn't that great?!"
there is NO SHOT that article was written by a human
@@d2rey596 Arguably, whether it was an LLM or if it was a human writer or team of human writers, they missed the mark.
(You didn't ask for this advice, but it may help someone else)
One of the best pieces of advice I have been given: "Read your writing aloud, because your brain will automatically catch grammatical errors".
The only caveat to that advice: pay special attention to what was actually written. Because your writing can be so "off the mark" that when read, you say a word or phrase that doesn't exist.
@@tanmaysingewar I think copilot is releasing its own products at this point.
FSD - Full Self Developing. When you put your 11 year old nephew on the driver seat of your car and pretend supervising him is less stressful than driving yourself.
If it's not my nephew but some Indian kid I've never met, I'm not sitting in the car and my insurance will take care of any outcome: fuck yeah, let's do a turbocapitalism, baby!
Don't worry its FSD (Supervised)
@@LordMegatherium car/acc accelerate!!
This comment is transcending the traditional social communication medium allowing the viewers to move into a new paradigm of idea sharing and thought provocation beyond the current limitations of the human brain.
Enthralling concept. Thou missed one exception the cohesion of thoughts synergised into one concept
So.... Meta ™️
I have translate to English link next to your comment.
As soon as corporate speak is so prevalent in the announcement you know the tool is built by non-technical people.
It’s obviously built by very technical individuals… but that’s not to say they were the decision makers
To be concise, verbose.
@@Kane0123 Originally written by a technical person only for a corpo-person to rewrite it in corpo-speak
@@chinesesparrows Those words are contradictory, if you didn't know.
I think this blogpost is meant to be appealing to shareholders and business people, that do not understand the process, but would like to cut down on devs
That's the sign of a gassed up product that is not designed to do the actual job and will fail.
yeah, looks like they'd prefer to learn the hard way.
AI is promising the same thing 5GL language promised and failed to deliver. To say I'm sceptical is an understatement.
A developer's job isn't to write code, it's to tell a computer what it must do. Changing the language in which the devs describes problems from a 1-4GL to a 5GL or to an AI "natural language" description doesn't change what a dev needs to describe. The hard part of the job is still in specifying things in a way so a computer will not misunderstand.
I really like this way of putting it. Until AI knows better in terms of high level thinking, planning, architecture, future maintainability, etc. it really is just another way to produce steaming piles of code.
There must be a point where natural language description ends up being longer than the actual code, and comes with major organisational problems, such as where do I put "...should have a border radius 3, but not when hovering, or next to a widget, and they should be laid out horizontally...". Inform 7 is an example.
@@nomad3245mmm dat scat code fetish
@@nomad3245 u mean piles of unused code?
Oh, another AI tool that can generate a counter and TODO app, thanks
This looks like 100% more effort than just writing the code
That subscribe insert caught me off-guard
Felt like like dropping soap in a public shower
@@chinesesparrows Ew, bad joke.
@@Lexi_Sharp Ew, snowflake humor.
it does improve life for junior engineers because when people try this they'll want to go back in hiring juniors right away
The hardest thing is when you need to start marketing smth and you have no idea how to start...
Its like the developers were coding Workspaces, and while giving it a test run, it became quasi-sentient, and wrote its own promotional copy.
The fact that “mobile friendly” they show with a web version and not iOS or Android apps shows a lot about where those apps are in terms of of gh priorities
Lol what are you talking about. "Mobile friendly" has always mean't a responsively designed web application so it works well on small screens. Mobile friendly does not mean native mobile app.
24:10 I checked the repo: the diff is because the file was checked in as CRLF in the initial commit, but the Copilot Workspace version has been converted to LF. This is actually also a somewhat common mistake.
A human developer would be expected to explain this, possibly also create checks to avoid this situation in the future (with an .editorconfig perhaps), while the LLM is just oblivious.
This is also one thing that GitHub is terrible at showing. If there are whitespace differences maybe the UI should show me the whitespace being different, just like the last newline in the file? But no, it just looks identical.
Never heard of this issue until this happened me today - didn’t know such a thing existed
This looks like Devin's older brother who is allowed to go to bars.
I was pretty hyped after seeing this on fireship’s channel. But seeing you dive into the details of what it’s doing wrong in the _example_ code pretty much quashed all my excitement.
The article reads as if they gave a group of junior engineers "focus medication," and told them to write an article, where the sky is the limit. That would explain why some paragraphs are made up of only one sentence 😅
That blog post was probably written by copilot. Sure as hell *looks* like it was written, or highly edited, or enhanced, by AI.
copilot should be like an emmet thats it, this will make new dev lazy, or conspiracy theory that they secretly make all new dev dependent on it
I worked as a translator in China for about 8 years. I'm guessing this article was probably written in Chinese originally. It's not just VC speak, it feels like Chinese VC speak to me. A lot of those sentences that have a lot of words that feel awkward in English actually feel very fluid in written Chinese. I don't think it came from AI beyond partial translation assistance probably.
People's reaction to these technologies crack me up. Instead of being blown away by the rapid pace that things are developing, they look at the current state and capability and do the developer equivalent of "My dog is so stupid! I can beat him at chess 3 out of 4 games, what an idiot dog!"
Look at where things are today, look at where they were a year ago, look at where they were a year before that. Draw a line. Where do you think it will be in a year?
Is the appartment unit of time linear?
Saying "cringe" all the time is bad practice.
Oh boy! Programming in natural language sounds like a great idea!
Because that's exactly what we need. More ambiguity! More fuzzy shit! Consistency and well defined language rules are soooooo 1970! Precise control over program behaviour? Who needs that?
I'm 1000% sure nothing could possibly go wrong.
It's interesting to me that most people think that getting started isn't the hardest part. When Theo said staying with it is the hardest part, it clicked for me, but before that, I totally thought getting started was the hardest part. When you don't even know the right questions to be asking, looking back at it feels daunting to me.
Hot take from people who love to extrapolate into the future ... "But Gpt-5 will solve all these problems" or "For now", my favorite one liner
This is targeted at C-suite folks who failed to get their devs to be "more productive".
Right off the bat, it's way too expensive. Clearly it needs a lot of work, just their sample shows that much. They should open it up for beta testing, let people use it for free until they work out the bugs. Even after that, charging by the hour is a deal breaker. Thank you for the review, Theo. It's far more interesting than enticing.
They advertised AI with AI using the grammar in their blog
5:32 THAT did changed my life 😳
These posts and tools are made for non-technical managers. GitHub/MS wants to sell them licenses, on the premise that they will revolutionize their company/department.
At this point, AI products are so obviously forced and just capturing the remaining VC money before reaching the tough of despair in the Gartner hype cycle
Many years ago, I read an article about how the word "so" entered everyday English because of the show Friends. I wonder if the same will now happen with the word "delve"
the comparison of supermaven and copilot was not fair at all.... not only did you not test the two on the same completion task, you made a point that copilot did not use tailwind in the solution, but supermaven also did the same right before.
If you're going to showcase something you have vested interest in. do it fairly.
"delve" is fancy language now? If we set the bar any lower, life will be one long limbo contest.
I just wanted to try it out but the very first step (Issue Specification) keeps blowing up in an error. Zero use
Honestly think the language difference is more technical vs non-technical people, were those papers in particular areas?
Because when I moved from international relations to applied informatics I was actively told off for this very thing.
I wouldn't flinch at 'delve' tbh
I partially agree. The business speak makes me highly skeptical and the output is an embarrassing pile of garbage.
But there are enough non-technical "decision makers", who will eat that s**t right up. Incidentally, these are the types of people I'm not going to feel sorry for.
Also, it's hard to say if a directional change is the right call. We might be one or two LLM generations plus a few tweaks away from this being amazing, but that's hard to say now. Anyway, the potential benefits are insane.
Launching early, however embarrassing the product may be right now, might even be correct. Iterating with customer feedback and collecting training data is just that valuable.
nothing wrong with starting sentence with And
i would like to try the experiment where they made a shared programming environment (was it github realtime iirc?)
13:12 "I'm sorry to the author"
Don't be, the author doesn't exist anyway.
Wtf we live in a world where we are short on professional developers? Literally when every company is laying off half of their professional developers 😂
This seems cool for learning a new stack or something like that. That's a win, right?
I read that essay you mentioned was good, 'Mind the Gap', as well as a few others. I know it was 2004 but wow, that was an extremely boneheaded self serving piece that is just working backwards from a desired conclusion with audaciously incorrect claims throughout. Very weak stuff typical of the era.
Why do companies keep saying they need to lower barrier of entry for programmers but on the other hand they fire 60% of their junior devs? If you don't want junior devs, why do you want more junior devs to commit code?
I spent some time trying to write a (much less ambitious) version of this and getting a bot to edit files and do multiple steps autonomously without loosing the plot... project managers just don't realize how hard that is. I wish project managers/marketers could let things be cool for what they can do without over promising.
I was using cody AI , aws toolkit and none of them was faster and better than supermaven thank you for sharing this cool copilot I really appreciate that
The "." Key I had no idea thank you!
Whenever i need to write a function that validates if a url is a valid github url, or check if we're using https, now I can. I can't wait to get my hands on this tool.
Notice how the generated code at 23:28 is also just wrong. Yay autocomplete!
@Theo - Of all the UA-camr's I watch, yours is the one where I really, really hope to get your personal reply to my comments!
15:38 It's stunning what a stupid idea this is and how no one at GitHub seemed to notice or try to stop or change this messaging. We don't need one billion of anything. If anyone thinks software actually matters that much, they need to take a break from computers.
I agree that such tools should run in the IDE you're already using. I agree the demo sucks; however, a configurable multi-agent to help with larger scope code changes that actually works is something I'm looking forwards to. Human in the driving seat is a must.
It missed compiling the regex. So the code is kinda almost broken from being slow IMO.
It’s honestly sad to see so much money and engineering time going towards pointless projects, rather than developing AIs that actually provide value.
the model github copilot uses is super not good enough to support this feature, too much hallucination and misunderstanding, i assume this is planned for gpt-5
"As programming in natural language...."
1. Most people are barely able to convey their idea in natural language to begin with, AI can not create anything useful from that as it currently stands without tons of extra context, which brings me to
2. Programming languages are designed to enable expressing ideas in a short and concise formatting, saving the amount of keystrokes necessary to develop a system. Natural language sucks at being concise (try writing an if/else flow in natural language and your preferred programming language, you'll see what I mean).
Agreed... I'm sick of setting up these toys only to find they're more trouble than they're worth. Great video, again 👍
15:11
This was sneaky 😂😂😂
that last part of GitHub's announcements felt like McKinsey consultants explaining Github lmfao
"The devs that are boosted that hard - they needed the help, let's be real" - Yeah, I'm sure Andrej Karpathy is a huge slacker and desperately needed that help
Somebody in Github got the memo that Devin was on the way, so the quickly set up a team of 50 engineers to make their own version of it. This is their way to justify their employment.
For developers with ADHD, getting started is often an issue - not knowing what the first step should be.
Imagine the codebase for this project...
as a guy trying to learn how to code, and i have pretty bad dyslexia id currently say trying to find useful resources for me to learn from is very difficult, i have a diploma in chemistry biology physics and maths being dyslexic isnt a significant problem because i can just use my imagination to understand it and see diagrams and maths which i dont struggle with, but coding is a tough one thats what im excited about the new gpt 4o desktop app whenever they decide to release the windows version, i think using the screen ''streaming'' to chat gpt with the voice is gonna be great for learning hopefully the same way it'll help blind people, just describing what im doing as i go i think will inscribe the language into my head a bit better lol
Ugh, so I just watched the demo video as a bit of a hate watch, and came across something that should be cleared up. The changes of the readme and the contributing were manually added to the to do lists. Copilot did not decide this on it's own. The fact that this wasn't clear in the blog post just shows how badly written the post was.
This is githubs demo btw: ua-cam.com/video/L5Xny6yehUg/v-deo.html
It's bullshit like this why I still don't use GH, but especially the thought that it might change your licensing and absorb your codebase without your permission. That was a smooth move editing the article to promote your channel. And I feel called out for both using the word delve and starting sentences with `and`. I delve into different things all the time, and I've never had anyone complain about it, at least not to my face.
Ffs it's self driving cars all over again
With AI, we're going to need the ESC key closer, or there will be a lot of repetitive strain injuries....
Maybe good to see idiomatic solutions if you're new to a language/framework to get a feel for it.. But yeah very limited use case that I can see.
Can’t wait for the day when devs are redundant and all of GitHub is running on copilot entirely
Lol your demo of copilot vs maven speed showed the same ostensible latency and speed. Investor colored glasses.
What Github Codespaces charges per hour, Unity Build Automation charges PER MINUTE.
The hard part is when you code yourself into a catch 22
That "treadmill" is the Microsoft's way. Always has been, ever since Win 3.x
Also listening to Theo rant about the quality of the products... Having a couple friends working at M$ that are Principle Devs, hearing them talk about new hires and the quality of people getting hired. One would say "Well it isn't like it was in the old days" and will continue on about why DEI is sh1t... Think, some new hires he encounters he will agree they are pretty smart but is surprised some of the more lower level basic things they have no clue about. Example might be memory management and handling stuff like that and performance, why or when to use a link list, or avl tree, quickorts are good except when...?
Measuring time by number of apartments lived in.
Don't get the part of github not having single dev environment solution. The approaches are different, it doesn't make sense to merge everything. Especially if some ideas don't stick, it would be easier to abandon them.
I mean, yeah, looking back makes me thing people simply does only want to dish out content for the heck of it. First they say they hate it, months later, you guys become patrons of it. I mean, I'm one of them, I hated copilot. But once someone said, that I should try the free trial, I myself have fallen in love with it. So yes, we can be really stupid sometimes, lol.
What's with the complaints about long sentences? Can you only read twitter posts nowadays? I agree the corpo speech is painful to read though.
I'm a typescript dev and I have a lot to say about github copilot
Ah yes, let's throw real world problems at the thing that was only tested against really simple interview questions like "make a function that can tell me if an array contains a negative number". Surely nothing will go wrong.
I genuinely hate that everyone goes so crazy over LLMs getting 80%+ accuracy on publicly available coding question sets before they look at what questions are being asked. The datasets are literally open source. Shit, most of them are on huggingface so you don't even need to download a csv.
But what if you are exactly having 3 years of experience. It should have been >= 3 years. Or =< 3 years 😅
I know.. Bounty check is still hard for Theo. 😊
If it happens to be exactly 3 years, just wait 1 nanosecond and you’re over 3 years :)
I'm still surprised by how insanely bad the github mobile app is on android.
maybe microsoft needs to push the same requirement as facebook, and make their app devs use android for a while.
this is the worst most embarrassing.... version of this tool you will ever use
I'd rather have a tool to say you forgot to delete the variable on line 50 dummy, now you have a memory leak than an AI that occasionally gives me good code (sometimes). The number of times I've had to backspace out X == null to X is null because the compiler checks to see if the object can be null, is frustrating.
maybe all of this isn't aimed at developers but business people who might pay money for it, even if devs don't like it
hot take. creativity is a significant part of being a developer, and if you're struggling to know how to start a project, then maybe you need to practice your creative thinking
The "reach" language drives me crazy, but it's hardly unique to copilot. It's been frustrating me for years as I try to look for and learn about tools to help be build things, to judge what will be worth investing my time into. It's ironic how much it all sounds the same, when marketing people write enthusiastic and yet generic buzzword fluff about why their product is great instead of telling me *what it does*. It could be AI, but it doesn't have to be.
It's 100% is what happens when marketing does the screenshots :P.
I don’t why you hating on that. This is a nice product, pretty much what cursor is doing, but they try to move it to the browser.
I agree with the not good separation of products, this should be part of the vscode copilot
🤣 "They needed the help. Let's get real" - Loved that one!
Another thing that is TERIBLE about llms here is the fact they scale so poorly on large repos...
Like this grows quadraticaly in repo size... like a small repo takes 20 minutes a repo 4 times the size should take 4*4*20=4hours+ ...
This is ridiculous
Copilot sucks so bad, it's a total confusing mess ever since they split it into workspace/terminal/vscode areas and then added slash commands it gets autocompletions wrong mixing up camel case and snake case and it still hallucinates simple things like telling you to use npx with some template that doesn't even exist..
18:50 "increase labour demand"? 🤨
There are lots of free, offline LLMs out there that are more than competent.
I only use chat, to generate unit tests
Thank you! That's exactly been my thoughts about AI tools as well!
Ooh fuck yea this supermaven is real quick, just got it after watching this.