Karthik Prabhu

Microsoft Imagine Cup World Top 5. Grad student @ IIITH. ML & Theory Researcher

[Karthik Prabhu @karpieml]

@Raj_1843 It probably took him that long to transition from H1B to green card due to the waiting period and by then there was no point in switching. Cannot switch before then, because of H1B indentured servitude (Vance is right, like it or not)

[Karthik Prabhu @karpieml]

@Brandonknudsen @omarvvvr They nudge you. It’s not a hard cap and they try not to enforce it. But they’ll keep nudging you until you feel guilty enough to switch.

[Karthik Prabhu @karpieml]

@omarvvvr Try it.

[Karthik Prabhu @karpieml]

@kiragoldner I’m scared😔

[Karthik Prabhu @karpieml]

Learning Journey Day #16

Finished the first draft of my research proposal.

(I’m a research masters student, proposal is in the same area I am currently studying on this journey)

[Karthik Prabhu @karpieml]

@HaohanWang Thank you!

[Karthik Prabhu @karpieml]

@HaohanWang Thanks for the reply! Yes, the ideal student-advisor matching should be based on personalisation.

As a student though, the incentive is to target these conferences for prestige and credibility, then apply to as many top schools as feasible. Dunno how that loop will break.

[Karthik Prabhu @karpieml]

Learning Journey Day #14 and #15

Still measure theory. And working a bit on my research proposal today.

[Karthik Prabhu @karpieml]

This could take a while🫠

[Karthik Prabhu @karpieml]

Learning Journey Day #13

Still learning measure theory😨

[Karthik Prabhu @karpieml]

@antoniolupetti I am skeptical since the author compiles this monograph using pre-prints which appear to be research contributions from the author alone, which is very unusual and slightly misleading for a textbook on a modern field.

[Karthik Prabhu @karpieml]

@ThomasWolgast @EmmettStinson Followed!

[Karthik Prabhu @karpieml]

@ThomasWolgast @EmmettStinson This could work. Will incentivise supervisors to self select their student’s work. The only catch is independent researchers. Probably an exception can be carved out for them.

[Karthik Prabhu @karpieml]

Our brains naturally keep processing input and generating output and never really stop from the time we are born till we die. LLMs perform a stateless transformation of input to output only upon invocation. This is a likely (and maybe only) barrier to consciousness.

[Karthik Prabhu @karpieml]

LLMs could be exhibiting the building blocks for emotional inference but consciousness requires agency and state, an LLM cannot invoke itself because the concept of “itself” does not exist to the LLM until it is called (invoked).

[Veee @vikktorrrre]

Dario Amodei: We don’t know if AI models are conscious, but we’re open to the idea that they could be.

“We’ve taken a precautionary approach.”

“We gave the models an ‘I quit this job’ button where they can stop doing a task.”

“They very infrequently press it, usually around disturbing content.”

“We’ve even found activations associated with concepts like anxiety.”

“Does that mean the model is experiencing anxiety? That doesn’t prove that at all.”

[Karthik Prabhu @karpieml]

That’s why biological consciousness is likely different. Because there is no real reason for a biological being like a human to think, yet it does (Until it dies)

[Karthik Prabhu @karpieml]

And looping an LLM inside an agent suddenly does not make it stateful or truly have agency, because the core mechanism or neural network used for generation is inherently stateless.

[Karthik Prabhu @karpieml]

@togelius @peterbhase The real problem is the volume of submissions of which a large majority are non meritorious which is exhausting reviewers. The solution could be more reviewers or some form of triage for reviewers. It is a difficult problem to solve, that’s true.

[Karthik Prabhu @karpieml]

@togelius @peterbhase The whole point of the double blind peer review was to judge the merit of a paper without looking at the name attached. It has worked well for AI conferences.

Even before AI slop, there have been non meritorious papers. Reviewers should continue judging on the basis of merit.