Knut Jägersberg's picture

Knut Jägersberg

KnutJaegersberg

AI & ML interests

NLP, opinion mining, narrative intelligence

Recent Activity

Organizations

LLMs's profile picture Blog-explorers's profile picture Qwen's profile picture Social Post Explorers's profile picture M4-ai's profile picture Chinese LLMs on Hugging Face's profile picture Smol Community's profile picture

KnutJaegersberg's activity

view reply

the more refined your intuition, and the 'smarter' your feeling of rightness, the less you need to explicitly spit out logical chains for reasoning. a problem here is the division between system 1 and system 2, while in reality, those are more on a continuum.

view reply

while reasoning is often understood as logical argumentation, what it actually means is using "good sense" which is broader than mere logic. a certain form of steeredness is required, but I don't think you can reduce it to cold rationality. In German we call reason "Vernunft", which has an important affective component.
You can also understand it as controlled thought, using some kind of affective marker to distinguish 'good' vs 'bad' thought.
The important part is that it does something and happens as part of fluid intelligence. It's changing knowledge, that is key, whilst changing it according to some standard, which can be logic and often is.

upvoted 2 articles 2 days ago
view article
Article

🦸🏻#10: Does Present-Day GenAI Actually Reason?

By Kseniase
6
posted an update 4 days ago
reacted to BlinkDL's post with 🔥 5 days ago
view post
Post
3978
RWKV-7 "Goose" 0.4B trained w/ ctx4k automatically extrapolates to ctx32k+, and perfectly solves NIAH ctx16k 🤯 100% RNN and attention-free. Only trained on the Pile. No finetuning. Replicable training runs. tested by our community: https://github.com/Jellyfish042/LongMamba