cloud

inference @openai, prev robotics, sora

[cloud @cloud11665]

@cudagdb trvth nvke

[Tibo @thsottiaux]

@apples_jimmy Nobody touch roon

[cloud @cloud11665]

@SebAaltonen Do you think this has more to do with the models “intuition” lacking or was this a one-off error?

[cloud @cloud11665]

@austinc3301 @tenobrus The beautiful thing is that nobody is forcing you to use all the features. The best thing that came out of modern c++ (23) imo is concepts

[cloud @cloud11665]

Mfw no one is looking at the data

[Peter Nutter @PeterNutter1]

Kimi K3.1 Pro doesn't just recall hidden tests (great thread by @JSchaeff3r). It also writes fake OpenAI ciphertext 🤯

Give it a send_message tool in Codex and sometimes it sends ciphertext instead of its report. Then it gets confused by what it just sent 🤔

[cloud @cloud11665]

it's peak...

[Jarred Sumner @jarredsumner]

some things I often say to claude when doing perf optimization work:

> how do we rewrite this to better leverage data-oriented design principles

> can you replace those hash tables with a flat array? can we make the pointers an index into the array? can we pack the most commonly used data into upper bits of the index?

> caching is cope for a slow design. do not add caches and delete all existing caches. instead, redesign it to be faster

> should this be a structure of arrays instead of an array of structures?

[cloud @cloud11665]

@_R4V3N5_ Claudefade

[cloud @cloud11665]

@yacineMTB It’s just so much better to talk to than astra im sad; but worry not, good things are coming!

[cloud @cloud11665]

@willdepue I think only time will tell

[cloud @cloud11665]

@nrwnwn @tszzl @mike64_t Never goon

[cloud @cloud11665]

@cudagdb More like only nvdisasm will never lie to you haha

[cloud @cloud11665]

@itsnoahd @scaling01 This is just how inference scales though

[cloud @cloud11665]

@scaling01 you don't have to imagine, astra ultrafast is a thing

[cloud @cloud11665]

@JasonBotterill You can test it by lying with the clock and slowing it down like 20x. That way if they try to figure out from any timestamps it will feel like they are running at insane speeds

[cloud @cloud11665]

@__tinygrad__ propagate the affine index maps through shrink, recognize the reduction tree, and lower it to a warp/block reduction :)

[cloud @cloud11665]

@CliffLattner @ModuloRoland It truly was that traumatizing haha