Deepak Singh

bio/acc

[Deepak Singh @watdipak7]

There are faint stars in the night sky that you can see, but only if you look to the side of where they shine. They burn too weakly or are too far away to be seen directly, even if you stare.

[Nate McGrady @natemcgrady]

me and my main agent watching a subagent do all the work

[Chris Lakin @chrislakin]

what your risk-aversion costs you

[Beff (e/acc) @beffjezos]

Deep Tech founders on SaaS era VCs:

[Peter H. Diamandis, MD @PeterDiamandis]

Curiosity weighs nothing, costs nothing, and compounds. Use it!

[Drew Hahn @drewhahn]

“the agent is working in the sandbox”

meanwhile the agent:

[Charlie Marsh @charliermarsh]

It used to be the case that reviewing a contributor's PR was a form of "investment" in that contributor. They'd learn from your feedback, grow, take it into account for future contributions, etc.

Unfortunately with agents I feel that's largely gone.

[Deepak Singh @watdipak7]

Direct store link: https://t.co/iaXJmjhy95. It's on the official Rockstar page. Currently at 83% download.

[Deepak Singh @watdipak7]

I thought the GTA 6 beta was complete bs. Went to the store today, clicked "Play Beta", and the installer actually started downloading. No tricks. Link in replies. 👇

[Deepak Singh @watdipak7]

@samarsheoran 100% agree

[Beff (e/acc) @beffjezos]

If you're in nothing ever happens mode, pivot to everything all at once

[Deepak Singh @watdipak7]

I've always been both curious and suspicious about the still high accuracy even after skipping most of the labor work while trying to learn something. It's called "distillation", but what do you still miss and don't want to miss in training and learning?

[tender @tenderizzation]

two schools of thought

[w̸͕͂͂a̷͔̗͐t̴̙͗e̵̬̔̕r̴̰̓̊m̵͙͖̓̽a̵̢̗̓͒r̸̲̽ķ̷͔́͝ @anthrupad]

when mathematics gets complicated enough the primitives will be more mind-like than number-like, and the proof techniques will have to adapt where neo-mathematicians can then go..
"am I... the counterexample?" and then attempt to be the counterexample (or example)

[Association for Computing Machinery @TheOfficialACM]

#OTD in 1945, Edmund Melson Clarke was born. Clarke received the 2007 #ACMTuringAward with E. Allen Emerson and Joseph Sifakis, for their role in developing Model-Checking into a highly effective verification technology that is widely adopted in the hardware and software industries.

Model checking is widely used to verify designs for integrated circuits, computer networks, and software. It has made the technology that underpins modern life possible. https://t.co/kUZuchxd0A

#TuringAward #computerscience #ModelChecking

[Han @HanchungLee]

犯其至难而图其至远者，发之以勇，守之以专，达之以强。

man these kimi pinyin bros got taste

[Deepak Singh @watdipak7]

@Madisonkanna Agree a 100%, @oneill_c knows his stuff well

[ali @waterloo_intern]

every research team needs to spend some time learning how their modeling code lowers down to gpu kernels

example: rms norm vs layernorm

industry assumes rms norm is cheaper. you don't need the x̂, it looks simpler, so it must be faster, so it got adopted

but that's not true. both rms and layernorm are memory bound kernels (the amount of time it takes to get the data to gpu cores is longer than the amount of time it takes to do the computation on said cores)

both take the same amount of time e2e

so can probably train the model right with either, but maybe it makes a difference (e.g why diffusion models still use norms with an affine shift e.g adaLN)

i get why. as open-source models get better and, inevitably, commoditized across the inference providers, overall serving speed determines user experience determines which model gets adopted

but there's no reason to superstitiously avoid free things

[Tyler @TylerDurden]

Two weeks to film. $3mil budget.

The greatest financial movie of all time.

[Deepak Singh @watdipak7]

Experience is just fine-tuning. Real intelligence is far more than training data + experience.

Replacing doctor's “experience” with Lymph will show what evolution does to each new generation. The young will surprise the experienced.

e/acc bio/acc

[François Chollet @fchollet]

Eventually, much of AI will converge towards intuition-guided symbolic world modeling, i.e. deep learning-guided program synthesis. It is inevitable. Symbolic modeling lets a system construct a compact, reusable, highly generalizable mental model of a problem space using minimal data.