> Even as someone who uses a lot of AI, if you can't be bothered to at least give it a prompt like "Go through the documentation and comments in detail and remove any obvious AI shibboleths like emdashes, it's not x it's y, rule-of-three, 'delve', excessive grandiosity and flourishes, boldness, bullet points, etc", you should receive a brisk kick in the rear.
...is it that easy? I mean, wouldn't the AI providers just include this in their system prompts and/or harness?
I get good results doing that, but it's all about being the human who has the taste and making a tasteful result. change the prompt, try again, change the temperature, whatever it takes.
I agree with the premise, but it's quite painful in practice constantly probing and prodding for justification and explanation -- especially because _even with_ the justification, explanation, etc, one's mental map / the "topology" of the thing being built is only very loosely being populated as a result of the conversation. I say this having continuously tried to find a way to keep my learning rate comparable to if I was writing the code myself, and having somewhat failed.
I'm starting to wonder if the thing to address is the anxiety itself rather than the "fuzziness about the code" that creates the anxiety - and more explicitly model myself as an engineering and/or product manager counterpart to these things. I wonder how non-IC EMs/PMs do it - it seems maybe fundamentally anxiety-inducing? – but they _do_ do this already (tolerate the fact that the underlying technical system is not fully within their grasp).
Unlike with an ai, a pm knows what it means to understand something. Just in general, they have had the part of the human condition which is gaining an understanding of something. Not memorizing a fact or learning how top perform some recipe steps, but understanding several dimensions of something.
They don't code, but there are other things they have learned and understand, and witnessed other people showing how they do or don't understand that thing. It could be about fishing or calligraphy or being careful what you say to kids or anything not just work stuff. From that they know how to recognize it in others on other topics. Modulo the effect of the exceptionally good and bad on both sides, a good bullshitter can fool a simple pm, but the exceptions to a generality are irrelevant.
And they know that some other people do understand the code. They haven't performed all the steps to reproduce someone else's reasoning and correct final result, but they don't have to do that every time to know that it's possible to do and would arrive at the same or equivalent result. It's not faith, or not blind faith like religion. It's just letting someone else do a job that you know if you wanted to you could do exactly the same and it would work exactly the same. You don't have to carry a load of grain from one pile to another to know for a fact that it's possible to do and roughly what it takes to do it.
They also know how to detect cues about consensus, or lack of it. When most people who understand a topic, you can tell, without relying on any simple rules about what words are spoken, you can tell when most of the people who have put in the time to understand something agree or disagree on some premis. And the people you are guaging that consensus from are again other humans who you share the human condition with. You have a power to understand and interpret them that you don't for anything else.
They may also understand some level of code. It will have been explained to them at some point to some degree. They will have some sort of simple example in their history where they were walked through something in some class or a meeting.
It makes me incredibly sad to see Osmani letting AI write his stuff for him.
I went to go find some of the stuff that he wrote pre-AI and found myself on his bio. Not only is it generated, it's incredibly clumsy and boastful.
In sum, Addy Osmani’s career is a testament to the impact one engineer can have by combining technical excellence with education and community leadership.
Osmani’s journey reflects the evolution of the web itself - ever faster, smarter, and more empowering for those who use it.
Few individuals have done as much to push the web forward while uplifting its developers, and that legacy will be felt for a long time to come.
Who would put these embarrassing brags on their own website? Did he even read this?
I wanted to avoid exactly this for me, so I made https://www.writelucid.cc/. It's free (bring your own API key), hopefully someone will find it useful.
It's a nice-looking app, but "solve good writing with AI" is exactly the thing I'm not interested in.
The way to avoid creating slop is to read and write things yourself, regularly. Be more selective about what you're trying to solve with AI.
You cannot escape the fact that AI has no taste, period. The more creative the medium, the more apparent this is. There's nowhere to hide from even a moderately literate reader.
I dunno. People worry we give up vital skills doing this. I question if they'll be vital in the future? If the LLM can genuinely solve the bug today, why wouldn't it be able to tomorrow?
One thing that seems fairly certain is that llms aren't going to get any worse. They'll probably keep getting better, but there is 0% chance they'll get worse.
If you can get away with not fixing the bug yourself today, the very idea of doing it yourself will be laughable tomorrow.
I try to learn the skills that the LLMs struggle with. Some of those skills will be made irrelevant too, probably when Mythos gets released to the public. But also some of them won't. Probably. The skills that Claude has a handle on today? Waste of space in my brain!
thinking is vital. just like homework at school to improve your thinking. final output can be totally ignored .
one could argue that thinking itself would be not needed in future. But we are not there. I would love that future though. thinking is suffering. if ai can think for us then it would eliminate suffering.
But you can't successfully outsource all of "thinking" to a LLM today(or, likely, ever), so this is mostly a strawman.
Thinking isn't Boolean either. A LLM let's you not think about, say, rust lifetimes, as an example. You'll still need to think about other things.
I'm not abandoning all thinking, I just change what things I think about. Some things I delegate thinking about(like rust lifetime syntax, to stick with my example), which frees up time to think about other, more interesting things, like application design, or architecture!
I believe almost nobody thinks original thoughts. I never have. At best I applied an idea from one area to another, which is something AI can do.
Moreover, most novel advancements seem like they come when society is ready for them - the nearly simultaneous discovery of calculus for example.
Pick any thought of yours you truly believe is novel and do a serious literature search on the topic and adjacent fields. Ask an AI to help you with the search if necessary :-)
https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
> The bug gets fixed. Your mental model doesn’t move.
> The symptom vanishes. You ship.
> The tool didn’t determine the outcome. The posture did.
Here's a free prompt if you can't come up with one that avoids this awfulness yourself: https://news.ycombinator.com/item?id=48100213
Or use software that only lets the AI critique/give you feedback, but not write for you.
> Even as someone who uses a lot of AI, if you can't be bothered to at least give it a prompt like "Go through the documentation and comments in detail and remove any obvious AI shibboleths like emdashes, it's not x it's y, rule-of-three, 'delve', excessive grandiosity and flourishes, boldness, bullet points, etc", you should receive a brisk kick in the rear.
...is it that easy? I mean, wouldn't the AI providers just include this in their system prompts and/or harness?
I get good results doing that, but it's all about being the human who has the taste and making a tasteful result. change the prompt, try again, change the temperature, whatever it takes.
I agree with the premise, but it's quite painful in practice constantly probing and prodding for justification and explanation -- especially because _even with_ the justification, explanation, etc, one's mental map / the "topology" of the thing being built is only very loosely being populated as a result of the conversation. I say this having continuously tried to find a way to keep my learning rate comparable to if I was writing the code myself, and having somewhat failed.
I'm starting to wonder if the thing to address is the anxiety itself rather than the "fuzziness about the code" that creates the anxiety - and more explicitly model myself as an engineering and/or product manager counterpart to these things. I wonder how non-IC EMs/PMs do it - it seems maybe fundamentally anxiety-inducing? – but they _do_ do this already (tolerate the fact that the underlying technical system is not fully within their grasp).
A few different differences.
Unlike with an ai, a pm knows what it means to understand something. Just in general, they have had the part of the human condition which is gaining an understanding of something. Not memorizing a fact or learning how top perform some recipe steps, but understanding several dimensions of something.
They don't code, but there are other things they have learned and understand, and witnessed other people showing how they do or don't understand that thing. It could be about fishing or calligraphy or being careful what you say to kids or anything not just work stuff. From that they know how to recognize it in others on other topics. Modulo the effect of the exceptionally good and bad on both sides, a good bullshitter can fool a simple pm, but the exceptions to a generality are irrelevant.
And they know that some other people do understand the code. They haven't performed all the steps to reproduce someone else's reasoning and correct final result, but they don't have to do that every time to know that it's possible to do and would arrive at the same or equivalent result. It's not faith, or not blind faith like religion. It's just letting someone else do a job that you know if you wanted to you could do exactly the same and it would work exactly the same. You don't have to carry a load of grain from one pile to another to know for a fact that it's possible to do and roughly what it takes to do it.
They also know how to detect cues about consensus, or lack of it. When most people who understand a topic, you can tell, without relying on any simple rules about what words are spoken, you can tell when most of the people who have put in the time to understand something agree or disagree on some premis. And the people you are guaging that consensus from are again other humans who you share the human condition with. You have a power to understand and interpret them that you don't for anything else.
They may also understand some level of code. It will have been explained to them at some point to some degree. They will have some sort of simple example in their history where they were walked through something in some class or a meeting.
These things are all missing with an ai.
It makes me incredibly sad to see Osmani letting AI write his stuff for him.
I went to go find some of the stuff that he wrote pre-AI and found myself on his bio. Not only is it generated, it's incredibly clumsy and boastful.
Who would put these embarrassing brags on their own website? Did he even read this?> embarrassing
I'd go a step further, this is hypocritical and insulting of him considering the content of the blog post.
He has upper management potential.
Indeed, it's a bit sad that he writes his own blog posts with AI, which is very obvious
I wanted to avoid exactly this for me, so I made https://www.writelucid.cc/. It's free (bring your own API key), hopefully someone will find it useful.
It's a nice-looking app, but "solve good writing with AI" is exactly the thing I'm not interested in.
The way to avoid creating slop is to read and write things yourself, regularly. Be more selective about what you're trying to solve with AI.
You cannot escape the fact that AI has no taste, period. The more creative the medium, the more apparent this is. There's nowhere to hide from even a moderately literate reader.
"I’m not anti-AI. I use these tools daily and have shipped more with them in the last year than in the five years before it.
[...] Software Engineer at Google working on Google Cloud and Gemini."
The things he must have seen.
I dunno. People worry we give up vital skills doing this. I question if they'll be vital in the future? If the LLM can genuinely solve the bug today, why wouldn't it be able to tomorrow?
One thing that seems fairly certain is that llms aren't going to get any worse. They'll probably keep getting better, but there is 0% chance they'll get worse.
If you can get away with not fixing the bug yourself today, the very idea of doing it yourself will be laughable tomorrow.
I try to learn the skills that the LLMs struggle with. Some of those skills will be made irrelevant too, probably when Mythos gets released to the public. But also some of them won't. Probably. The skills that Claude has a handle on today? Waste of space in my brain!
thinking is vital. just like homework at school to improve your thinking. final output can be totally ignored .
one could argue that thinking itself would be not needed in future. But we are not there. I would love that future though. thinking is suffering. if ai can think for us then it would eliminate suffering.
What are we without thinking? What’s the point of our existence in that case? Just being a biological bag of soup consuming resources?
In fact, thinking is what makes humans interesting. What separates us from random bacteria if we don’t think anymore?
Your questions only make sense if you assume there's a "point" to existence.
its the Faustian bargain we have to make to eliminate suffering
But you can't successfully outsource all of "thinking" to a LLM today(or, likely, ever), so this is mostly a strawman.
Thinking isn't Boolean either. A LLM let's you not think about, say, rust lifetimes, as an example. You'll still need to think about other things.
I'm not abandoning all thinking, I just change what things I think about. Some things I delegate thinking about(like rust lifetime syntax, to stick with my example), which frees up time to think about other, more interesting things, like application design, or architecture!
whats the basis for your claim that its impossible in future.
I don't see how that would be a good future... how is thinking suffering?
'Suffering' is hyperbole. But one can't deny thinking is hard and painful. Otherwise we wouldn't have so many shortcuts for it.
suffering comes from movement of thought. doesnt it?
Something that only thinks of the things thought before isn't much of a tinker now is it?
I believe almost nobody thinks original thoughts. I never have. At best I applied an idea from one area to another, which is something AI can do.
Moreover, most novel advancements seem like they come when society is ready for them - the nearly simultaneous discovery of calculus for example.
Pick any thought of yours you truly believe is novel and do a serious literature search on the topic and adjacent fields. Ask an AI to help you with the search if necessary :-)
> thinking is suffering
then thinking is also joy and happiness :)
"I am one of these people who keeps pushing this trash, but I like to pretend that I care about you."
"And oh, if you got affected by all of this, then it is cognitive surrender, meaning it is your fault."
> That isn’t a conspiracy. It’s UX gravity.
This article has a satirical quality I'm quite enjoying. To write is to think. If you're not thinking, how are you learning.