1 Year = “Long Term Support”
I was perusing the GitHub Copilot documentation tonight, a kind of thing I’ll routinely do, something I call a ‘documentation soak’ where I go through practically everything in the documentation for a particular product, framework, etc. to soak in the various aspects. I do this to plant mental flags, follow rabbit holes and keep an eye out for weak spots (functionality that looks underused / un-loved and might be removed).
I found a bit of a humorous gem in the Github Copilot docs tonight I had to take a ‘snapshot’ in time. If you’ve been around in the business long enough, hopefully it’s a bit obvious why it’s humorous.
Do You Understand the ‘L’ in LTS?
This is a fast moving target, as is everything in the realm of AI, I get it - but 1 year?
LTS support is something that’s supposed to be for engaging enterprise customers who want stability and reliability. One year is NOT a long term support period, it’s a joke. It’s a tone-deaf offering to enterprise customers who are looking for a stable and reliable product to build their worklfows on.
This is not even mentioning the behind-the-scenes “tuning” we’ve seen from other companies like Anthropic on Opus models where the results and behaviors are being tweaked and changed on a regular basis, which is not something you want to deal when consistency is already a huge challenge.
Also: Risky Fallback Behavior
A questionable strategy: automatically falling back to an “inferior” model.
This isn’t an internet or cell phone data plan where you get throttled after a usage cap. It materially changes the already non-deterministic behavior in very meaningful ways…
Who’s in charge over there over this stuff?!