Hacker News Reader

zinccaton Building a personal, private AI computer on a budget

P40 don't support fp16 well, buy 3090 instead

zinccaton Building a personal, private AI computer on a budget

I feel that you are mistaking the two bandwidth numbers

zinccaton PhD Knowledge Not Required: A Reasoning Challenge for Large Language Models

That make sense

zinccaton PhD Knowledge Not Required: A Reasoning Challenge for Large Language Models

I see, but still there's a lot of reasonings just for counting the letters. And ridiculous reasonings like:

FedEx"? Five letters? F-E-D-E-X? No, that's six. Wait, FedEx is two syllables: Fed-Ex. Five letters? F-E-D-E-X? That's five letters? F, E, D, E, X. Yes, five letters. FedEx.

zinccaton PhD Knowledge Not Required: A Reasoning Challenge for Large Language Models

My feeling is that a lot of challenge could come from the tokenizer used by the model, similar to r in strawberry problem.

zinccaton Llama3V is suspected to have been stolen from the MiniCPM-Llama3-v2.5 project

previous post of llama3v on hackernews: https://news.ycombinator.com/item?id=40504827

zinccaton LLMs can't do probability

The author used the temperature of 0.7, which is wrong

zinccaton LLMs can't do probability

I have retried the experiment with temperature=1, the result for 20 left (0.8)/right (0.2) is 17 lefts and 3 rights. I doubt why it is different from the blog.