Comments by "Vitaly L" (@vitalyl1327) on "Claude 3.7 is the best model for devs." video.

  1. I suspect people keep doing it wrong. End-do-end coding models are a dead end. Training those models on too much data result in overfitting, you're getting the most average, the "always do it the most popular way" solution. Slow down. We have perfectly capable coding models that do not even know the programming languages they're asked to code in. And they work way better than Claude 3.5 did (did not try 3.7 yet).
    4