News for hackers. Stuff that matters.

A comment on AMD MI300x GPUs with GEMM tuning improves throughput and latency by up to 7.2x. Parent comment

Re:AMD MI300x GPUs with GEMM tuning improves throughput and ... by idonotknowwhy on Saturday June 29, 2024 @09:43AM (#40830465)
Thanks for articulating something I've noticed happening all over the internet and even in YouTube video scripts.

Are claudisms different from gptisms?

Why can't these authors tell ChatGPT to write with a different prose and avoid "delve", "crucial", etc?

Wizardlm writes these same things in all its answers too

[ Parent | on HN ]
Re:AMD MI300x GPUs with GEMM tuning improves throughput and ... by robko on Saturday June 29, 2024 @10:23AM (#40830837)
> Are claudisms different from gptisms?

Sorry, no idea. I rarely use Claude.

> Why can't these authors tell ChatGPT to write with a different prose and avoid "delve", "crucial", etc?

The authors could do this, but that would contradict the reason for using ChatGPT, which is to do less work.

> Wizardlm writes these same things in all its answers too

WizardLM has inherited that from its instruction tuning dataset, which has been generated with ChatGPT: https://openreview.net/pdf?id=CfXh93NDgH

[ Parent | on HN ]