Karan Joisher

I turn coffee to code

[Deedy @deedydas]

The most impressive AI video generation research dropped yesterday.

1 minute long consistent Tom and Jerry clips can be created from training a 5B model on ~7hrs of data for 50hrs on 256 H100s.

Not even OpenAI, Google or Meta can do >20s!

Paper, code and examples are public:

[Deedy @deedydas]

NVIDIA's $7B Mellanox acquisition was actually one of tech's most strategic deals ever.

The untold story of the most important company in AI that most people haven't heard of

1/12

[Karan Joisher @karanjoisher]

@dreamingtulpa @hollamad_

[Karan Joisher @karanjoisher]

@hollamad_ @cursor_ai https://t.co/a1oqMRVPh4

[Riley Brown @rileybrown]

Try this, it works very well....
@tedx_ai

[Chamath Palihapitiya @chamath]

Can’t make this up. We’re living in a simulation.

[Łukasz | Wookash Podcast @wookash_podcast]

@rfleury, thank you so much for joining me!

@everyone, enjoy this technical deep dive on debuggers and programming in general!

[Kevin Joey Chen @kevxalchemy]

While everyone watches AI, Google broke time itself.

Their quantum chip just solved a problem that would outlast the universe — in 5 minutes.

And this is just the start...

What you need to know about Willow — and how it'll transform humanity in 2030+: 🧵

[gingerBill @TheGingerBill]

This trailer is just over 10 years old but it spawned a career for me and many others, as well as the many friends I made that I still have today.

Thank you for everything you did, @cmuratori.

https://t.co/r15C0i6rn1

[Not Jerome Powell @alifarhat79]

We’re saved

[Eskil Steenberg @EskilSteenberg]

Recurring C programmer conundrum: You want to allocate memory up front, but you don't know how much memory will be needed. If anyone have good approaches to this, please share them with me! Here are mine:

[Daniel Lemire @lemire]

Any code that can request the allocation of an undetermined amount of memory can fail, either directly or indirectly.

Period.

There is no way around it.

You can pick at most one of those...

1. Arbitrary memory budget.

or

2. Code that cannot fail.

[Elan Ruskin @despair]

@imcolbyhall @pedrothedagger @MatRopert - Different pools for different systems. Physics draws from separate heap than components, textures, etc. Hard budgets for each.
- Accounting. Tag every alloc explicitly so you can report where memory goes.
- Perf. new/malloc costs more than you'd expect, even w/o syscall.
🔚

[Casey Muratori @cmuratori]

I'll go one further and say everything should be committed. I commit the built binaries as well. Storage is cheap. Recreating old tool chains is not.

[Grant Slatton @GrantSlatton]

The junior dev asked the senior dev “why are you pushing this code with no abstraction? What if you want to change it in the future?”

The senior dev responded “then I will change it in the future”

In that moment the junior dev was enlightened

[Ritesh Oedayrajsingh Varma @rovarma]

Whenever we assume some kind of bounded N in Superluminal, we inevitably end up regretting it.

“How many DLLs could an application possibly load?”

Turns out the answer to that question is “at least 113440”, a bit higher than our estimate of “probably no more than a few hundred”

[Sebastian Schöner @s4schoener]

In today's episode of "What am exactly I waiting for here?" I waited for Visual Studio to open 477 projects in a Unity Engine generated solution. It took > 15min. Did I attach a profiler? Why of course I did, because I ran out of patience and @superluminal_sft is my only friend.

[Omkar Ranadive @RanadiveOmkar]

Deep networks exhibit class-selective neurons in all their layers. Intriguingly, it was recently shown that these neurons aren’t crucial for network function. So, why do they even exist? Our new study reveals the special role they play in early training
https://t.co/cqMBlHZLib
1/

[Omkar Ranadive @RanadiveOmkar]

Are you curious about how machine learning can be applied to perform early detection of wildfires?

Come check out our work at the @ClimateChangeAI workshop at #NeurIPS2022

Paper: https://t.co/wrNVuL3kcJ

w/ @alchera_ai, @FireScoutAI

[Karan Joisher @karanjoisher]

@JioCare I already have an active plan but since I will be travelling to USA, I bought an international roaming plan. In the app I can see this international roaming plan as an upcoming plan with no option to activate it. How do I activate this international roaming plan?