What is even worse is that this isn't even a proper article, this is just a thinly concealed advertisement.
I have found the opposite to be true, starting this year. You can commiserate with people who are interested in cryptography.
- "Dexter" is no longer "Dexter's Lab" (It's Darkly Dreaming Dexter)
- "Dark Descent" is no longer "Amnesia" (It's some other game)
- "Server" no longer means "server" (It means a Discord guild)
- "Serverless" now means "On a server"
- "Operating system" sometimes means "web app"
- "Powered by" has nothing to do with power
And Internet speeds are still measured in bits, in case anyone is not running the latest System/360
Jason isn't my friend anymore, but a data interchange format.
Neither is Kate, she's how I orchestrate my containers.
Ghosting doesn't involve poltergeists or a Halloween, Zoom isn't a function on my camera, and fishing steals my data and doesn't get me delicious salmon.
Here. Here's another article about it[0]. Like all language prescriptivists, you're simply wrong.
[0]https://www.thecut.com/2018/01/the-300-year-history-of-using...
Oh, and since we're apparently trading insults about linguistic views: like all linguistic descriptivists, you don't know what words mean. ;)
This feels like a JVM-land interpretation? I think it's quite domain-specific.
This is an odd take to me when LLMs are very, very good at generating code, which has got a lot of attention recently. Sure it may be a different beast from what we've currently identified as "codegen" but it remains a descriptive term for the code-generating technology.
Codegen's long history of macro-fied (or similar kind of scripting) source writing matters here. It has the very important property of having consistent/deterministic output from a process that can be verified with very high confidence by rudimentary human inspection.
It's like calling autocomplete "codegen".
I truly love LLM-assisted coding. I would never call it codegen, and think it can even be unethical to do so when the stakes are high, because it gives it the veneer of trustworthiness that lends one to carelessly not audit it.
LLM-aided code writing should, IMO, be called "code assist", not "codegen".
I'm basing my evaluation on Copilot. It functions as a highly context-sensitive and very useful autocomplete, so the "autocomplete" label is a complete, accurate, and precise description for it.
It does not fill the same role as writing repeatable macros, batch scripts in part of the build process, token manipulation, or joining a tabular dataset with templates...all components associated with traditional code generation.
In the most strict pedantic sense, yes, LLMs create lines code via their internal processes, and that could be called "code generation" at a technical level. But they serve different needs, with different techniques, and different interfaces.
In terms of common parlance, no PM or manager who is not pants-on-head dumb is going to suggest replacing scripts that generate code 1:1 with LLM output, unless the terminology itself has confused them into thinking they fill the same role.
Poor communicators, whether because they are overly pedantic or undertrained, may make this mistake. But in a team environment, if "codegen" is considered a valid name for LLM output, an effective communicator is always going to need to clarify which they mean, because the tasks are not interchangeable in the least.
Ummm.... Awful code that often looks right at first glance, maybe.
Maybe LLMs can generate the kind of code that's really shallow in its complexity, but for literally everything I would call interesting LLMs have produced hot garbage. From "it doesn't quite do what I want" to "it couldn't possibly work and it's extremely far from being sane," though it always looks reasonable.
> Maybe LLMs can generate the kind of code that's really shallow in its complexity, but for literally everything I would call interesting LLMs have produced hot garbage. From "it doesn't quite do what I want" to "it couldn't possibly work and it's extremely far from being sane," though it always looks reasonable.
None of this has any bearing.