"Nerf"ing models isn't real in the vast majority of reported cases. Benchmarks like this or the 100 other "let's see if nerfing is real" copies would have shown it by now if it was.
I made a graphic to explain why people feel like the models get nerfed:
The idea is that new models can handle up to a certain level of complexity, at which point they fall apart. Every new model can handle more complexity, so there's a wonderful time upon release when you feel like you can do anything, only for you to hit the complexity ceiling a few days later when you saturate it. Rinse and repeat for the next model.
Nerfing is certainly real and I don't see how you could argue it isn't.
A/B testing alone would result in a performance nerf for one group.
8-bit quantized models will barely show degradation on benchmarks. The performance is reliably at 99% of the non-quantized model. 4-bit quantization retains somewhere around 95-98% performance on benchmarks. But if you've ever used a 4-bit model, it feels lobotomized.
And just consider what a compny serving these models would do if they were at capacity. Would they stop serving the model altogether? Of course they wouldn't...
Denying that models experience purposeful degradation is gaslighting.
johnfn · · focus · HN ↗
I made a graphic to explain why people feel like the models get nerfed:
<a href="https://x.com/thesilenceturns/status/2103551351825543610" rel="nofollow">https://x.com/thesilenceturns/status/2103551351825543610
The idea is that new models can handle up to a certain level of complexity, at which point they fall apart. Every new model can handle more complexity, so there's a wonderful time upon release when you feel like you can do anything, only for you to hit the complexity ceiling a few days later when you saturate it. Rinse and repeat for the next model.
rybosworld · · focus · HN ↗
A/B testing alone would result in a performance nerf for one group.
8-bit quantized models will barely show degradation on benchmarks. The performance is reliably at 99% of the non-quantized model. 4-bit quantization retains somewhere around 95-98% performance on benchmarks. But if you've ever used a 4-bit model, it feels lobotomized.
And just consider what a compny serving these models would do if they were at capacity. Would they stop serving the model altogether? Of course they wouldn't...
Denying that models experience purposeful degradation is gaslighting.
applicative · · focus · HN ↗