logoalt Hacker News

ThrowawayR2 • today at 4:02 PM • 7 replies • view on HN

Models only have information up to the date when they were generated. APIs get updated and expanded, new frameworks and libraries get introduced, bugs get fixed. How long is an existing model going to still be useful? A year or two at most?

As existing models become out of date, somebody has to pay the massive cost for training updated models to replace them, including, of course, the free models.


Replies

Arnt • today at 4:51 PM

I'm using Claude on APIs on which it was not trained, and it's doing well. The techniques it uses to debug and find solutions aren't that different from older APIs. "Let me dump all of this response and guess which field might be what" sometimes, "let me get a stack trace just before it goes wrong", many familiar development techniques.

I need to fill in timesheets, which uses a proprietary API. Claude guessed and iterated until it found out how to use that API, and filling in those sheets is now much less bothersome. I only need to enter a password every three days instead of four times per day.

jjav • today at 5:52 PM

> APIs get updated and expanded, new frameworks and libraries get introduced, bugs get fixed. How long is an existing model going to still be useful?

You don't rely on the model knowing about such things up front. You feed it the latest API documentation and then work off that.

➕ show 1 reply
tintor • today at 5:05 PM

Not even human's can keep track with all the latest APIs.

Top models can search the web and read the API docs.

Opus 5.5, for example, can write zig code very well, even though there is very little zig code repos out there.

Training models (of same size) gets more efficient over time from combination of software, hardware, and infrastructure improvements, so retraining that open model to update it, will be cheaper and cheaper.

➕ show 2 replies
jamienk • today at 6:04 PM

Why do we accept this "massive cost" phase all the time? I want to see the data and intelligent analysis of it. I think this is not clear yet.

InsideOutSanta • today at 5:49 PM

Current models do well with proprietary systems they were never trained on. Ar most you need a good AGENTS.md and a directory full of skills.

luckydata • today at 5:40 PM

Red herring. AIs can read an api spec and do just fine.