With Manus chat mode, you can now explore and brainstorm your ideas before using agent mode to transform them into a deep-dive research site. Check out this case where Manus turns a string of simple chats into actionable web insights on Indie Hackers.
Total credits used: 0 (chat mode) + 817 (agent mode)
Sonnet 4 is INSANE on LoCoDiff
it gets 33/50 on the LARGEST quartile of prompts (60-98k tokens) which is better than any other model does on the SMALLEST quartile of prompts (2-21k tokens)
Exploration strategies in deep RL are such a critical topic. I almost immediately regretted it when I started writing on this big subject because it has so much more content than I expected.
But here it comes, phew:
https://t.co/FUcmM3veog
It is very funny and encouraging to find that minimizing external acting is also the goal of world model (https://t.co/LQIfXzhCl2) proposed by @ylecun
Actually, we independently discover that minimizing external acting actually is maxmizing internal reasoning of the model in our latest work: OTC-PO (https://t.co/MQvVL5fvp2). In other words, our goal is same and it is believed OTC-PO is the foudamental step to learn the turly intelligent model that capture the abstraction of the world as mush as possible and only take actions when necessary.
Stay tuned, we will release our position paper soon and maybe 2nd version of OTC-PO.