pulkit

Sr Staff SWE @Google, building intelligent devices, Pixel ❤️ Gemini

prev, Gemini Nano, TensorFlow Lite, MS@CMU, ThoughtWorks, ...

[pulkit @pulkitbe]

@_arohan_ my father in law left for India today, else would've asked him 🤣

[pulkit @pulkitbe]

@DynamicWebPaige temptation 😈

[pulkit @pulkitbe]

@_arohan_ so, now ant would've bought compute from every lab other than openai? ;-)

[pulkit @pulkitbe]

@ChanduThota Congrats on the role, Chandu!

[pulkit @pulkitbe]

@darshil himsagar

[pulkit @pulkitbe]

@garrytan Sorry about the trouble. Happy to get more details over DM, and share with the right group.

There's definitely a strong desire to get the personal/work profile experience right.

[pulkit @pulkitbe]

@DynamicWebPaige @DeepLearningAI @AndrewYNg Fun fact. I took Andrew's ML course on Coursera, and then decided to pursue an MS in ML after that in 2013, which led to cmu, then google, then tensor flow and now pixel AI.

The power of learning democratization is real :)

[pulkit @pulkitbe]

@jenzhuscott @octonion If you work in tech, you will be completely unsurprised by this.

[pulkit @pulkitbe]

@anmol01gulati Congrats Anmol!

[pulkit @pulkitbe]

@ponnappa agreed. we're staggeringly more advanced than 2 decades ago, so the next generation will hopefully get there.

[pulkit @pulkitbe]

this sounds nice, but has glaring gaps.

India is significantly behind US, China (and arguably other countries) when it comes to foundational hard tech, which drive tech generation cycles.

[Megh Updates 🚨™ @MeghUpdates]

🗣️ Interviewer: “There are so many talented Indians in Silicon Valley startups. Why are there not more Indian startups in India itself?”

🇮🇳 Chief Economic Adviser: “I don’t think you’ve been to India lately. India’s startup ecosystem is second only to Silicon Valley.” 🔥

[pulkit @pulkitbe]

@petmongrels I agree. We're hopefully no more than one generation behind, foundational tech requires a deep capital and talent pool.

I just think we should be aware of it, and honest about it.

[pulkit @pulkitbe]

my guess: the best PyTorch and Jax programmers are equally good. it's just that Jax users are likely to skew domain, or ML experts.

similar to how the median OCaml, Haskell and Rust programmer is likely more skilled than the median Java, and JS programmer.

[François Chollet @fchollet]

When looking at deep learning profiles, one of the most obvious tells between a mediocre and great candidate is whether they list PyTorch or JAX.

[pulkit @pulkitbe]

@jocarrasqueira @GoogleDeepMind @Google Yeah, it was wonderful! Berlin has such a different vibe :)

[pulkit @pulkitbe]

@GauravML also, other factors-

(1) Google is so large most criticisms can be true, and localized,

(2) "big tech" is absent from X discourse (other than devrel for top programs), so absence of commentary seems absence of capability.

plus, no VCs to hype all day :)

[pulkit @pulkitbe]

@_arohan_ @MillionInt @arb8020 congrats, if true Rohan.

[pulkit @pulkitbe]

rise up, pull up

[pulkit @pulkitbe]

@thevirdas CMU campus, see if you can catch a theatre performance. their drama school is incredible.

(sneak into some seminars on AI / Robotics if you want to glimpse the future :))

UPitt tower has some enchanting classrooms.

[pulkit @pulkitbe]

@DynamicWebPaige @Android @PyTorch Ah yes, brings back memories.

So many perfetto traces during Gemini nano!