George White
@its.stonehippo.com
Yes, they would really struggled with ChatGPT in 2003 😉 😂
November 28, 2025 at 6:05 PM
Yes, they would really struggled with ChatGPT in 2003 😉 😂
Oh, I know! And for those it works for, I know it's appreciated.
But can we agree that perhaps the problem sometimes *isn't* that what they need is coaching to get bette results? That perhaps the problem is that there are (possibly many) times an LLM is the wrong tool for the task?
But can we agree that perhaps the problem sometimes *isn't* that what they need is coaching to get bette results? That perhaps the problem is that there are (possibly many) times an LLM is the wrong tool for the task?
November 28, 2025 at 5:40 PM
Oh, I know! And for those it works for, I know it's appreciated.
But can we agree that perhaps the problem sometimes *isn't* that what they need is coaching to get bette results? That perhaps the problem is that there are (possibly many) times an LLM is the wrong tool for the task?
But can we agree that perhaps the problem sometimes *isn't* that what they need is coaching to get bette results? That perhaps the problem is that there are (possibly many) times an LLM is the wrong tool for the task?
Does LLM code generation fail 100% of the time? No, clearly not.
Does it fail a significant part of the time, often in subtle ways? Yes, it does.
If it works for you, great. There's a lot of people it won't work for, literally, because what they need isn't generated by the tools.
Does it fail a significant part of the time, often in subtle ways? Yes, it does.
If it works for you, great. There's a lot of people it won't work for, literally, because what they need isn't generated by the tools.
November 28, 2025 at 5:34 PM
Does LLM code generation fail 100% of the time? No, clearly not.
Does it fail a significant part of the time, often in subtle ways? Yes, it does.
If it works for you, great. There's a lot of people it won't work for, literally, because what they need isn't generated by the tools.
Does it fail a significant part of the time, often in subtle ways? Yes, it does.
If it works for you, great. There's a lot of people it won't work for, literally, because what they need isn't generated by the tools.
In many cases, they're right, the experience they're having is that an LLM is not a good fit for the thing they are doing. Telling them it can produce workable code under some other circumstance isn't all that helpful.
They'll often be told they're doing it wrong, which is unfair and often untrue.
They'll often be told they're doing it wrong, which is unfair and often untrue.
November 28, 2025 at 5:34 PM
In many cases, they're right, the experience they're having is that an LLM is not a good fit for the thing they are doing. Telling them it can produce workable code under some other circumstance isn't all that helpful.
They'll often be told they're doing it wrong, which is unfair and often untrue.
They'll often be told they're doing it wrong, which is unfair and often untrue.
For example: in my coder-viability correlation statement above, a caveat is that for some use cases, generated code will do well, no matter the skill of the human involved.
LLM code generation is better for some tasks than others. It's not a general solution to all coding needs.
LLM code generation is better for some tasks than others. It's not a general solution to all coding needs.
November 28, 2025 at 2:10 PM
For example: in my coder-viability correlation statement above, a caveat is that for some use cases, generated code will do well, no matter the skill of the human involved.
LLM code generation is better for some tasks than others. It's not a general solution to all coding needs.
LLM code generation is better for some tasks than others. It's not a general solution to all coding needs.
If you want people to accept the position that "LLMs are good at X", it would be helpful to be clear how and when that's true. And when it is not.
November 28, 2025 at 2:10 PM
If you want people to accept the position that "LLMs are good at X", it would be helpful to be clear how and when that's true. And when it is not.
It's a "flame war" because the statement is more conditional than implied.
LLMs *may* produce excellent code. Or they may not, which I can say based on the code my students are struggling with. Plus: the viability of any generated code is often correlated with the skill of the human "coder".
LLMs *may* produce excellent code. Or they may not, which I can say based on the code my students are struggling with. Plus: the viability of any generated code is often correlated with the skill of the human "coder".
November 28, 2025 at 2:10 PM
It's a "flame war" because the statement is more conditional than implied.
LLMs *may* produce excellent code. Or they may not, which I can say based on the code my students are struggling with. Plus: the viability of any generated code is often correlated with the skill of the human "coder".
LLMs *may* produce excellent code. Or they may not, which I can say based on the code my students are struggling with. Plus: the viability of any generated code is often correlated with the skill of the human "coder".
Yes, but they'll have to do it very, very slowly
November 28, 2025 at 2:18 AM
Yes, but they'll have to do it very, very slowly
Oh, hyper-focused procrastination!
I can be VERY busy about doing the thing I shouldn't be doing while avoiding the thing I should.
I can be VERY busy about doing the thing I shouldn't be doing while avoiding the thing I should.
November 26, 2025 at 2:47 AM
Oh, hyper-focused procrastination!
I can be VERY busy about doing the thing I shouldn't be doing while avoiding the thing I should.
I can be VERY busy about doing the thing I shouldn't be doing while avoiding the thing I should.
Reading through an AuDHD description and hmm that sounds very familiar.
I think the obsessive communication of things that interest me is one area of challenge (but also works in my favor in some cases). I've got my hobby horses and I ride them like every day is a rodeo.
I think the obsessive communication of things that interest me is one area of challenge (but also works in my favor in some cases). I've got my hobby horses and I ride them like every day is a rodeo.
November 26, 2025 at 2:45 AM
Reading through an AuDHD description and hmm that sounds very familiar.
I think the obsessive communication of things that interest me is one area of challenge (but also works in my favor in some cases). I've got my hobby horses and I ride them like every day is a rodeo.
I think the obsessive communication of things that interest me is one area of challenge (but also works in my favor in some cases). I've got my hobby horses and I ride them like every day is a rodeo.
The Mechanical Turk to Oblivious Recreation of an Existing Thing Pipeline is real
Also: who valued this at $1b?!
Also: who valued this at $1b?!
November 25, 2025 at 12:41 PM
The Mechanical Turk to Oblivious Recreation of an Existing Thing Pipeline is real
Also: who valued this at $1b?!
Also: who valued this at $1b?!
Hey, I got rid of half my obsolete cables
November 24, 2025 at 2:28 AM
Hey, I got rid of half my obsolete cables
I keep thinking of the Matrix. Humans "burned the sky" to cut the machines off from solar energy, and the machines made humans into batteries.
All of which is incredibly implausible and dumb and still seems more rational than tech-bro geoengineering.
All of which is incredibly implausible and dumb and still seems more rational than tech-bro geoengineering.
November 24, 2025 at 1:32 AM
I keep thinking of the Matrix. Humans "burned the sky" to cut the machines off from solar energy, and the machines made humans into batteries.
All of which is incredibly implausible and dumb and still seems more rational than tech-bro geoengineering.
All of which is incredibly implausible and dumb and still seems more rational than tech-bro geoengineering.
Reposted by George White
