Does OpenAI make their models worse as they get closer to launching a new model to accentuate the improvements of the new model?

Certainly feels that way to me

All these AI for software development companies making claims about % increase in productivity when nobody is able to reliably measure dev productivity.

Have they solved measuring dev productivity? I would pay for that!

Everytime OpenAI makes a new announcement they expose how weird their testing is: "stricter evaluation setting" "answer right in four out of four attempts", so previous testing allowed models to attempt a question 4 times?

The fact that every Elixir LSP constantly crashes really affects my confidence that Elixir is a stable language or a language you can write reliable code in

Business consultants constantly trying to sell bogus personality tests are a solution to your problems

BlueSky takes way too long to load