This is the reason why I tell people who write code that AI/LLMs can't
ever replace them like this. AI tools can only be a tool that you use.
To write code, you'd still need to know something by yourself at the end
of the day.
On 03/06/2025 17:06, Ar Rakin wrote:
This is the reason why I tell people who write code that AI/LLMs can't
ever replace them like this. AI tools can only be a tool that you use.
To write code, you'd still need to know something by yourself at the end
of the day.
As the sage said, never is a very long time.
I started my career writing assembler. Nobody uses it any more for
system work - it's much easier and cheaper to use a higher level
language. Those skills I learned back then will never be performed by an >API. But nor are they performed by humans any more (at least on that >obsolete ISA!)
In recent years I used C++. I understand a lot of code now is written in >languages like Python. You could regard them merely as a detailed spec
for the processes you need the computer to carry out.
Get that spec right, and the computer behaves. Perhaps one day the AI
will be able to read a spec in English - but it will probably have to be >lawyer's English to avoid ambiguities.
Or maybe we'll have an AI that is truly intelligent...
On 03/06/2025 17:06, Ar Rakin wrote:
This is the reason why I tell people who write code that AI/LLMs can't
ever replace them like this. AI tools can only be a tool that you
use. To write code, you'd still need to know something by yourself at
the end of the day.
As the sage said, never is a very long time.
I started my career writing assembler. Nobody uses it any more for
system work - it's much easier and cheaper to use a higher level
language. Those skills I learned back then will never be performed by an API. But nor are they performed by humans any more (at least on that obsolete ISA!)
In recent years I used C++. I understand a lot of code now is written in languages like Python. You could regard them merely as a detailed spec
for the processes you need the computer to carry out.
Get that spec right, and the computer behaves. Perhaps one day the AI
will be able to read a spec in English - but it will probably have to be lawyer's English to avoid ambiguities.
Or maybe we'll have an AI that is truly intelligent...
Andy
On 03/06/2025 17:06, Ar Rakin wrote:
This is the reason why I tell people who write code that AI/LLMs can't
ever replace them like this. AI tools can only be a tool that you use.
To write code, you'd still need to know something by yourself at the end
of the day.
As the sage said, never is a very long time.
On 6/5/25 10:23 PM, Vir Campestris wrote:
On 03/06/2025 17:06, Ar Rakin wrote:
This is the reason why I tell people who write code that AI/LLMs
can't ever replace them like this. AI tools can only be a tool that
you use. To write code, you'd still need to know something by
yourself at the end of the day.
As the sage said, never is a very long time.
I started my career writing assembler. Nobody uses it any more for
system work - it's much easier and cheaper to use a higher level
language. Those skills I learned back then will never be performed by
an API. But nor are they performed by humans any more (at least on
that obsolete ISA!)
Saying *nobody* uses assembler today would be wrong - many low level projects still need to write assembly code. They may be not as relevant
as before, but it is still used. For example, compiler developers still need to have a very good understanding of the assembly languages of the systems they target.
In recent years I used C++. I understand a lot of code now is written
in languages like Python. You could regard them merely as a detailed
spec for the processes you need the computer to carry out.
Get that spec right, and the computer behaves. Perhaps one day the AI
will be able to read a spec in English - but it will probably have to
be lawyer's English to avoid ambiguities.
Or maybe we'll have an AI that is truly intelligent...
Andy
Maybe. Nothing is impossible, but I wouldn't like the idea of AI/LLMs taking over. Wouldn't want to see another THERAC-25 incident, caused by AI/LLMs writing code.
I asked: "Write a program that is valid C90 but invalid
C99."
ChatGPT said:
Certainly! To illustrate this, I'll write a C program that
is valid in C90 but invalid in C99.
Richard Heathfield:
I asked: "Write a program that is valid C90 but invalid
C99."
ChatGPT said:
Certainly! To illustrate this, I'll write a C program that
is valid in C90 but invalid in C99.
This AI cannot speak English and answer questions as they
are asked:
1. You did not ask it /whether/ it were possible to write
such a program, so its answer "Certainly" either makes
no sense, or means agreement to fullfil your request.
2. If it means agreement, the following sentence makes no
sense, for what is it going to illustrate (if not the
possiblity of such a program)?
3. And then it cocludes with a blatant tautology,
promising to "illustrate" the wrting of a C90-but-not-
C99 program, by writing a C90-but-not-C99 program.
Its diction is like that of schoolchild taught to start an
answers by repeating the question, but not understanding how
to do it.
After serious thinking Anton Shepelev wrote :
Richard Heathfield:
I asked: "Write a program that is valid C90 but invalid
C99."
ChatGPT said:
Certainly! To illustrate this, I'll write a C program that
is valid in C90 but invalid in C99.
This AI cannot speak English and answer questions as they
are asked:
1. You did not ask it /whether/ it were possible to write
such a program, so its answer "Certainly" either makes
no sense, or means agreement to fullfil your request.
2. If it means agreement, the following sentence makes no
sense, for what is it going to illustrate (if not the
possiblity of such a program)?
3. And then it cocludes with a blatant tautology,
promising to "illustrate" the wrting of a C90-but-not-
C99 program, by writing a C90-but-not-C99 program.
Its diction is like that of schoolchild taught to start an
answers by repeating the question, but not understanding how
to do it.
+ 1
Thanks to underlining it.
| Sysop: | DaiTengu |
|---|---|
| Location: | Appleton, WI |
| Users: | 1,064 |
| Nodes: | 10 (0 / 10) |
| Uptime: | 159:14:25 |
| Calls: | 13,691 |
| Calls today: | 1 |
| Files: | 186,936 |
| D/L today: |
7,069 files (2,119M bytes) |
| Messages: | 2,411,313 |