Twitxy Saved

头雁

Verified account This account is verified. Learn more

@alacheng

-AI & RL
-Crypto/隐私技术/ZK/FHE
-具身智能
-研究/发现/分享 - 伟大的技术/产品/公司
-投资/技术/健康
Joined March 2011
3.2K Following 24K Followers 6.3K Posts

Latest posts 5

头雁 Verified account @alacheng
在MacBook 上微调Qwen2.5 0.5B 来实现一个使用用户的WhatsApp 5000+条个人聊天数据训练一个小模型,学会模仿该用户来发短信。

-以 4 位量化
-使用的是LoRA 与 mlx-lm,

github.com/Fedeparg/talk-…
github.com GitHub - Fedeparg/talk-like-me: Fine-tuning a 0.5B model on my own WhatsApp replies, and measuring... Fine-tuning a 0.5B model on my own WhatsApp replies, and measuring what it really learned - Fedeparg/talk-like-me
2 replies 1 reposts 13 likes 808 views View on X
头雁 Verified account @alacheng
Ilya 肯定认为超级安全就是什么也不发布最安全,现在的AI模型都太不安全了,都成了黑客的提款机了!
spor Verified account @sporadica 18h
What's Ilya up to these days?

@ssi has been around for 2+ yrs and we haven't heard a peep, no releases, no products, no papers, nothing

Are they a dud? Or are they still gonna come out of nowhere some day and just mic-drop safe superintelligence on the rest of the industry?
View quoted post on X