Nasim Rahaman (@nasim_rahaman) 's Twitter Profile
Nasim Rahaman

@nasim_rahaman

Deep Learning and espresso slurping @MPI_IS + @Mila_Quebec. Previously, Physics & research fellow @UniHeidelberg, and research intern at @MetaAI & @awscloud.

ID: 25835564

linkhttp://nasimrahaman.super.site calendar_today22-03-2009 16:04:29

5,5K Tweet

886 Takipçi

644 Takip Edilen

Nasim Rahaman (@nasim_rahaman) 's Twitter Profile Photo

Honestly I find it bewildering that people get away without ascribing moral value to large models. Yeah, it’s a bunch of matrix multiplies. We’re a bag of chemicals, what’s your point?

Nasim Rahaman (@nasim_rahaman) 's Twitter Profile Photo

I’ve built search agents from scratch and benchmarked against perplexity along several axes. I can tell you there’s more than meets the eye, and it’d be nowhere as fast if they didn’t maintain their own web index. Same applies to you.com.

Nasim Rahaman (@nasim_rahaman) 's Twitter Profile Photo

High quality take by Ate-a-Pi. Zuck is not afraid of an AI Apple (my old thinking), Zuck is afraid of AI boy/girlfriends, i.e. character.ai. If everyone and their dog makes AI girlfriends (and lonely people spend money on them), Zuck wants the distribution.

Nasim Rahaman (@nasim_rahaman) 's Twitter Profile Photo

The monkey brain wants to optimize for empowerment. Empowerment in multiagent systems is scarce by nature, abundance will not solve that. Luckily, we built elaborate proxies of empowerment to trick our monkey brains (= trophies). That’ll be the only thing left for money to buy.

Nasim Rahaman (@nasim_rahaman) 's Twitter Profile Photo

Seeing through 𝔊𝔴𝔢𝔯𝔫’s commoditize your complement lens — Nvidia will try their darned best to make sure that LLMs are commoditized (open source?). If LLMs are not commodities and the big boys follow Google’s lead in cooking their own chips, Nvidia is cooked.