sudosays
@sudosays.com
AI researcher and skeptic.
Trust but verify.
Opinions subject to change, beliefs updated according to learning rate schedule.
Trust but verify.
Opinions subject to change, beliefs updated according to learning rate schedule.
On emacs it’s easy just: M-x kill-puffer
March 20, 2025 at 3:11 PM
On emacs it’s easy just: M-x kill-puffer
After reading this my priors suggest your job is most likely not to teach people about bayesian statistics.
March 20, 2025 at 10:20 AM
After reading this my priors suggest your job is most likely not to teach people about bayesian statistics.
Basically, LLMs are great. I think they are part of a whole.
March 6, 2025 at 7:46 AM
Basically, LLMs are great. I think they are part of a whole.
I think LLMs are only part of the whole.
We need to understand human cognition more clearly and cognition reasoning in general.
Instead of focusing on “human” thinking as the pinnacle, what if we thought about creating more robust forms of cognition?
This should be the AGI goal.
We need to understand human cognition more clearly and cognition reasoning in general.
Instead of focusing on “human” thinking as the pinnacle, what if we thought about creating more robust forms of cognition?
This should be the AGI goal.
March 6, 2025 at 7:46 AM
I think LLMs are only part of the whole.
We need to understand human cognition more clearly and cognition reasoning in general.
Instead of focusing on “human” thinking as the pinnacle, what if we thought about creating more robust forms of cognition?
This should be the AGI goal.
We need to understand human cognition more clearly and cognition reasoning in general.
Instead of focusing on “human” thinking as the pinnacle, what if we thought about creating more robust forms of cognition?
This should be the AGI goal.
It feels like LLMs bypass our disbelief/skepticism by capturing language really really well. Our instinct/conditioning is to ascribe intelligence to high quality language.
But it is not perfect.
Reasoning strategies and RAG are good approaches towards making LLMs more robust.
But it is not perfect.
Reasoning strategies and RAG are good approaches towards making LLMs more robust.
March 6, 2025 at 7:46 AM
It feels like LLMs bypass our disbelief/skepticism by capturing language really really well. Our instinct/conditioning is to ascribe intelligence to high quality language.
But it is not perfect.
Reasoning strategies and RAG are good approaches towards making LLMs more robust.
But it is not perfect.
Reasoning strategies and RAG are good approaches towards making LLMs more robust.
> it’s been 10 years since the go/rust cohort came up.
a close up of a man 's face with a gray shirt on
Alt: Gif of Matt Damon aging in Saving Private Ryan, demonstrating how the poster feels suddenly aged by the knowledge of time.
media.tenor.com
February 26, 2025 at 4:17 PM
> it’s been 10 years since the go/rust cohort came up.