Where my gorillas at?

@ChombaBupe Who tf are you?

@buffaly_ai @deedydas PDF pages in JPEG.

@TheoPhysiJ @jerryjliu0 @deedydas What do you mean by interpret? You mean analyse or extract structured information?

@KavadeNitin @AniBaddepudi @deedydas @reductoai Did you try this with Gemini 2.0 Flash?

@Informcosmos She’s allergic to clothes.

@kimmonismus He’s been saying that for more than a year now.

@btibor91 I’ll use my cousin’s ID if you they can give pro for $20 but I know they will never offer that lmao

@mark_k @btibor91 I have seen other models reaching pretty high performance when trained on reasoning chains of R1 with very little data. I wonder if it’s because of this.

Transformers, as great as they’re, still not valuable enough to be used in most high economic use-cases due to inherent issues like hallucinations and limited context. Transformers are still an unproven tech mainly because of this and that’s the number reason why I think Nvidia isn’t investing their capital and focus on building supply chain for transformers specific GPUs and rather focus on hardware that’s more generally useable. I am sure they’ve prototypes that are on par with something like Groq perhaps better but I think they simply don’t feel from a business standpoint it warrants focus and capital in building transformers specific hardware until these issues are solved because anyway their current hardware does a pretty decent job however inefficient they are.

It’s not just about having a nice hardware, it’s the ecosystem, reliable supply chain and ability to handle scalable needs and most of all predictability. These hardwares are very new and unestablished and doesn’t make sense to take that leap of faith when you already have something like NVIDIA. Pretty sure Nvidia already has prototypes that are transformer specific and offer same or perhaps better performance than these hardware. I guess they’re waiting for transformers tech to mature enough to warrant heavy investment and focus from Nvidia.

@SpencerKSchiff @OpenAI They just fooling us man.

@jordanwjdurand @RepLuna I know right she’s just fucking with us.

@ashtom Finally

@_kickface_ @_lilpoptart What do you infer?

@ashtom What’s the context length on paid plans for this model?

How tf Sonnet so good that it always ranks on top without much updates? When 4 Sonnet coming?

@OfficialLoganK Hi, is there any benchmark on multi lingual performance of flash 2.0?

This is a very important chart. We are still not there yet but are on the trajectory of getting there soon.

@ColbyHawker @AviSchiffmann None. It’s just a gimmick.