"Judgement" is perfect for now. I use CC and Codex every day, and I agree they spit out something that will work on the first run, but they go wild about some system design choices. For now, I agree with you that judgement and system design/architecture is the distinction.
What I worry about and think will happen is we will see emerging platform that does full systems, and replit is a good example of that. I don't think these tools are enterprise ready yet (I might be wrong), as enterprise is always about integration with other systems including legacy ones and data silos.
I think we will be like COBOL developers now; we will maintain pre-AI systems until we retire.
I worry about that too, but I'm not sure we are close to that yet. Maybe we'll end up working on legacy systems where AI would break more than it would fix, or maybe we would become prompters. Idk!
> AI can write the code. It can’t architect the system. It can’t decide which tradeoffs to make, or know that the elegant solution it just generated will fall apart at scale, or understand why the team chose a boring technology stack on purpose.
I would add: "Yet."
Just as I've been completely astonished at the advancements AI has made in writing code, I can detect a trajectory at AI becoming an expert architect as well, likely within a shorter period of time than we'd all expect.
Not even yet - ask it to give you a research and plan for an easily maintained, highly scalable architecture and run a few adversarial agents against your plan- it will 100% do that today effectively. Like anyone if you don’t ask the right questions, you don’t get the right answers.
I've read the Tao Te Ching dozens of times. Every few years I'll re-read one new passage, daily, for three months (there aren't too many words within this semi-spiritual text).
My most recent read is the first, post-ChatGPT. From Verse Thirteen, three lines finally jumped out at me (which never have, before):
>>"I suffer because I'm a body; if I weren't a body, how could I suffer?" [1]
Already LLMs have shown me connections that no other human could endure/conjure from me (I've paid for a few attorney/therapists in my few decades living). Currently I'm the plaintiff in a lawsuit which I began with LLM counsel, and now have human counsel — this arrangement has saved lots of prep time, and led to interesting discussions with both counsel, human ¬.
One interesting conversation led to my human attorney recommending Neal Shusterman's Scythe Trilogy, which I've since read and absolutely re-recommend. Written in 2016 (same year as Attention is All You Need), it eerily hypothesizes many of the SciFi complexities that omnipotent general AIs now already-do ("Thunderhead" in scythespeak).
[1] Ursula K. LeGuin ~translation~, similar to Buddhist concept of "life is suffering"
yeah, I wrote it myself, but since I'm not a native English speaker, I do use AI to "fix" and "polish". I think AI made me a worst writer in a way lol.
it's my bad, I should have been more careful at keeping the content how I wrote it, without much of the fine tuning GPT did.
Wow, a guy who works at an AI company posting a blog post hyping AI to the moon.
Wow, the top comment on HN (currently) is from a user who works at company which sells AI tooling, talking about how amazing AI is going to get.
Wow.
Yeah people are starting to catch on ive noticed. About time!
"Judgement" is perfect for now. I use CC and Codex every day, and I agree they spit out something that will work on the first run, but they go wild about some system design choices. For now, I agree with you that judgement and system design/architecture is the distinction.
What I worry about and think will happen is we will see emerging platform that does full systems, and replit is a good example of that. I don't think these tools are enterprise ready yet (I might be wrong), as enterprise is always about integration with other systems including legacy ones and data silos.
I think we will be like COBOL developers now; we will maintain pre-AI systems until we retire.
I worry about that too, but I'm not sure we are close to that yet. Maybe we'll end up working on legacy systems where AI would break more than it would fix, or maybe we would become prompters. Idk!
> AI can write the code. It can’t architect the system. It can’t decide which tradeoffs to make, or know that the elegant solution it just generated will fall apart at scale, or understand why the team chose a boring technology stack on purpose.
I would add: "Yet."
Just as I've been completely astonished at the advancements AI has made in writing code, I can detect a trajectory at AI becoming an expert architect as well, likely within a shorter period of time than we'd all expect.
Not even yet - ask it to give you a research and plan for an easily maintained, highly scalable architecture and run a few adversarial agents against your plan- it will 100% do that today effectively. Like anyone if you don’t ask the right questions, you don’t get the right answers.
When that happens, we'll having nothing else to do.
I've read the Tao Te Ching dozens of times. Every few years I'll re-read one new passage, daily, for three months (there aren't too many words within this semi-spiritual text).
My most recent read is the first, post-ChatGPT. From Verse Thirteen, three lines finally jumped out at me (which never have, before):
>>"I suffer because I'm a body; if I weren't a body, how could I suffer?" [1]
Already LLMs have shown me connections that no other human could endure/conjure from me (I've paid for a few attorney/therapists in my few decades living). Currently I'm the plaintiff in a lawsuit which I began with LLM counsel, and now have human counsel — this arrangement has saved lots of prep time, and led to interesting discussions with both counsel, human ¬.
One interesting conversation led to my human attorney recommending Neal Shusterman's Scythe Trilogy, which I've since read and absolutely re-recommend. Written in 2016 (same year as Attention is All You Need), it eerily hypothesizes many of the SciFi complexities that omnipotent general AIs now already-do ("Thunderhead" in scythespeak).
[1] Ursula K. LeGuin ~translation~, similar to Buddhist concept of "life is suffering"
20 years left if health plays along, and I feel the same pain as the author.
We are all on the same boat now, whatever it takes us!
well, the author has no issues using an LLM to write the article.
And that’s the thing… it’s good at writing down my ideas in way that I can’t as a non native English speaker.
But it needed the judgement, the experienced and suffered input to know what to help me write
[flagged]
And yet this article itself has all the hallmarks of AI slop. Fitting.
yeah, I wrote it myself, but since I'm not a native English speaker, I do use AI to "fix" and "polish". I think AI made me a worst writer in a way lol.
it's my bad, I should have been more careful at keeping the content how I wrote it, without much of the fine tuning GPT did.
In a way, you did what you preached. You used judgement to determine what to write about and then had AI touch it up for you.
I don't view it as a bad thing.
That's true! that make me feel better lol. The internet is so sensitive to AI slop, it's hard to know what's the right balanced of usage when writing.
i would have preferred your real writing over the ai-ified slop
[dead]