正在加载视频...

视频加载失败

This is why having AI write code for you sucks. Literally just have the training wheels on forever. Great point by DHH

951,351 次观看 • 2 年前 •via X (Twitter)

42 条评论

Panta 的头像
Panta2 年前

@dhh This will be a big split among programmers. Those who are process oriented vs those who are outcome oriented.

Sebastian Raschka 的头像
Sebastian Raschka2 年前

@stephenoladele_ @dhh It’s probably going to be both: - assisted coding Mon-Fri to get things done (quicker) - and unassisted coding on the weekend because coding is fun!

Valentin Mihov 🦞 的头像
Valentin Mihov 🦞2 年前

@dhh I am coding for the outcome, not for the "text manipulation" part. AI is a great tool for me, as it really allows me to think on the bigger picture and not on the "what were the parameters of this function again?"

adam 🏕️ 的头像
adam 🏕️2 年前

for me personally, this argument is akin to saying “I don’t like self driving cars because I like driving” I think driving is perfectly fine thing to enjoy doing for me tho, as long as I get where I need to go and there is less I have to do to make that happen—I prefer that mode

neuropunk 的头像
neuropunk2 年前

@dhh Mfers act like they've never talked to another human or visited stack overflow in their lives

brian lowery 的头像
brian lowery2 年前

@0xholman @dhh But you're learning by doing that. You're not learning by having AI complete the entire problem for you without you having to think

Ravonus 的头像
Ravonus2 年前

@0xholman @dhh When I use AI and it actually answers a question I don't understand, I have dialect with it and break it down with my own understanding. It helps be grasp it much faster than I have ever before. So not even sure where this is coming from.

brian lowery 的头像
brian lowery2 年前

@0xholman @dhh That’s what you should do. But, a lot of people don’t do that.

neuropunk 的头像
neuropunk2 年前

@R4vonus @dhh mostly cope

neuropunk 的头像
neuropunk2 年前

@R4vonus @dhh people who care about programming and creating good software will make sure they understand what's going on. The people who don't, likely would have never gotten into programming the first place, or as was written, just wants the assignment done.

neuropunk 的头像
neuropunk2 年前

@R4vonus @dhh And that's at a very high level, most people I know writing production software with a framework don't know wtf the framework is actually doing

neuropunk 的头像
neuropunk2 年前

@R4vonus @dhh at the end of the day it's all for fun

Omiron - e/acc 的头像
Omiron - e/acc2 年前

Silly take. If you want to deep dive in to theory and structures and paradigms of code by hand, fine, but pretending it’s anything other than for your own edification is false AI will soon be doing everything. I memorized the Greek alphabet when I was 8. It’s never served any purpose. Knowing how to write a bash script from memory will be just as worthless

brian lowery 的头像
brian lowery2 年前

Full video here

Phil 的头像
Phil2 年前

@dhh There was nothing stopping him from taking the training wheels off and learning those things.

Carl Poppa 🛸 的头像
Carl Poppa 🛸2 年前

@dhh Isaac Asimov made exactly the same point 66 years ago in ‘The Feeling of Power’ — a dystopian, futuristic short story

Amit 的头像
Amit2 年前

I remember my old school teachers in engineering school insisted on logarithmic tables and not using calculators in class - these dudes sound like that. Unless you wanna code as a form of artistic expression, using ai to code solutions to your or someone’s real world problem quickly is great. Training wheels or not.

manuel 的头像
manuel2 年前

@dhh Programming will be less about the semantic details and more about the bigger picture. Learning syntax will be still necessary but only for reading the code you generated. Maybe it's time to move from the bicycle to the airplane.

K.O.O 的头像
K.O.O2 年前

@dhh Yup I deleted Cursor after this. I like to code too much to not be in full control

NTC 的头像
NTC2 年前

@dhh they aren't training wheels they are rocket boosters and they've barely started testing strap in

Csaba 的头像
Csaba2 年前

@dhh Many programmers like the “on-the-spectrum” part of programming like solving small riddles, writing beautiful code etc. AI will work to their detriment and benefit those who care about the outcome. Ultimately you need to be in love with bulding a house not hammering a nail.

{ nestedlogic } 的头像
{ nestedlogic }2 年前

@dhh It depends. It can be training wheels, or it can be an engine. I think you lose something regardless, but sometimes you just have to get to where you're going. You can still focus on craft and quality if you drive it, instead of letting it drive you.

macca 的头像
macca2 年前

@dhh *Sliced bread is invented* "Nah i don't like sliced bread because i enjoy the process of slicing the bread myself"

Jonas Vetterle 的头像
Jonas Vetterle2 年前

I think there levels to this and in the end every programmer has to find the balance that’s right for them. Totally agree that it can make you lazy and stop you from learning. On the other hand if you actually do know how to do something because you’ve done it a thousand times and can’t be asked to do it again, it’s nice to automate that imho That said, I find myself editing AI code suggestions more in language that I’m more familiar with than ones that I’m not. Probably because I spot more of the mistakes that AI makes, and I’m just more opinionated in languages I know well

GoldMagikarp 的头像
GoldMagikarp2 年前

This is a skill and tooling issue. As people figure out the right processes and tools, writing code by hand is like using a hammer and chisel to make a sculpture. There’s always going to be need for a few artisans to do this. But most value will be created by the people using pneumatic chisels and CNC carvers. Tools like aider give a small glimpse of where this is going. It’s going to go a lot farther. Engineers will be like sky crane operators, moving tons and tons of metal with a tiny joy stick.

Jason Leow 的头像
Jason Leow2 年前

@dhh Here's an alternative point of view. AI is just a tool. We can use it like training wheels that hinders our growth, or we can use it as a springboard to 10x our learning:

Alan Jones 的头像
Alan Jones2 年前

@dhh The age of the hippie programmers.. Your time has passed old man!

Jean-Paul 的头像
Jean-Paul2 年前

@dhh His last 7 words in the clip are most important. It doesn't suck. It sucks in certain situations. Many code as a means to an end. The sooner they can get an MVP out, the better. Otherwise, they'll spend months working on software that no one might ever use.

Data™ 的头像
Data™2 年前

@dhh

Lawrie ✨ 的头像
Lawrie ✨2 年前

@dhh I’m not negating the point, but DHH hasn’t had anything ‘great’ to say since he publicly sharted any goodwill he had right down the drain. He has his replies off for a reason.

Derek Neighbors 的头像
Derek Neighbors2 年前

@dhh Most code that gets written is “just for the outcome”. 🤷‍♂️

div 的头像
div2 年前

@dhh Kinda like “I don’t wanna use a library, i wanna implement it myself”

OMF 的头像
OMF2 年前

@dhh Fully agree here. I stopped using LLMs as a programming assistant after a few weeks because I could feel myself offloading much of the thinking responsibility onto them, and my own skills were starting to feel rusty. It might be good for short term wins, but not long-term growth.

ibrahim demir 的头像
ibrahim demir2 年前

@dhh That's why I always ask what it did and why for every answer it gives, instead of copying and pasting.

Rob Williams 的头像
Rob Williams2 年前

This is not the amazing take these two bobbleheads think it is. This is basically indistinguishable from people ranting about wanting to develop their own canister film in 2024, at least he is honest, in an obiter dicta kind of way. In Psychology this is what happens when people over identify with what they do. Then the idea of not doing it is equal to identity loss, which is terrifying to the self-obsessed numbkin in question. Furthermore, this is a guy who has been plying the same stack for a decade and a half. I will take the opinion of someone who did not trivial things in a bunch of languages over the Bernie Sanders my stack was right back then and it's still right. He's an expert on learning? for realz? no.

Arne Völker 的头像
Arne Völker2 年前

@dhh Should I be able to write the code myself, or keep on training wheels? For me, it's training wheels. They make it possible for me to code at all. But I have learned some things verbatim that I like typing out. Like: git push origin main Feels like patting myself on the back

Josep de Cid 的头像
Josep de Cid2 年前

@dhh Twitch gameplay enjoyers listening to @dhh's take about watching a game instead of playing it:

Sid Justice ⚖️ 的头像
Sid Justice ⚖️2 年前

@dhh valid point for short term side effects but by his logic we should all be writing code in notepad a tool is a tool we should make sure we are the drivers

KE programmer 的头像
KE programmer2 年前

@dhh DHH gets it

David Ackerman 的头像
David Ackerman2 年前

@dhh I disagree. I agree that knowing things is good but not everything is worth knowing. To eschew AI just to avoid falling into the non-learning trap is to throw the baby out with the bath water

Hyperdimensional Hegelian Chaos Giga Wizard™️ 的头像
Hyperdimensional Hegelian Chaos Giga Wizard™️2 年前

Yes, this is absolutely true. Here’s a deal, you’re not gonna be able to use AI to help you write code, unless you already know how to write code. You already have to understand the entire ecosystem of the platform you’re coding on for the AI to be useful, otherwise the AI will completely destroy your project. You have to be a super expert on the platform you’re coding, otherwise it’s even worse.

Hao fuwei 的头像
Hao fuwei2 年前

@dhh Who cares. AGI possibilities scale is Dyson spheres and you're talking about muh vim

相关视频