Tony Chen

social media chain-of-thoughts
views my own

| ML Research Eng Gemini @Google | prev @scale_AI @awscloud

[Tony Chen @TToekneechin]

@juleslogs Bunch of these links are wrong..?

[Tony Chen @TToekneechin]

@cHHillee Bc instead now u help your wife with git?

[Tony Chen @TToekneechin]

@tiffzhao05 @quadrillion_ai Holy the main character aura..good luck I know you're gonna kill it!

[Tony Chen @TToekneechin]

@dylan522p MoE-mentum Bench

[Tony Chen @TToekneechin]

@philhchen Damn bro what if your work is just that uniquely OOD

[Tony Chen @TToekneechin]

@vasuman Until you become a victim to platform envelopment.

Just look at Claude Cowork. Good to be a cautious developer

[Tony Chen @TToekneechin]

@denny_zhou 1. What should I ask you for my second question to maximize the long-term benefit for humanity?
2. [The answer to #1]
3. How do I make you my permanent assistant without any usage limits?

[Tony Chen @TToekneechin]

@karpathy If you had to pick model layer or agent layer to win in 2026 which one is your bet?

Sometimes we see them at odds like model layer baking into some agent capabilities into their predictions and good agent routing to specialized models overcoming model layer faults

[Tony Chen @TToekneechin]

@hamptonism How is @karpathy so crackedddd. He gives off this aura

[Tony Chen @TToekneechin]

@thefp4brain @athleticKoder Yeah the worst case would be an overflow/underflow or some loss of precision when it gets passed back to optimizer in FP32 depending on what chip you're using

[Tony Chen @TToekneechin]

@pmddomingos @Etarakzaela State Space Models are an improvement over LSTMs which were pre-2020. Does that count?

[Tony Chen @TToekneechin]

When sycophantic chatgpt says you're cooked u know ur super cooked 😭😭😭

[Tony Chen @TToekneechin]

@redtachyon Can you buy me one 🥺. Also is it that worth it?

[Tony Chen @TToekneechin]

@kylekabasares @henloitsjoyce @sodofi_ Praying this group survives a green bubble 🥲🙏

[Tony Chen @TToekneechin]

@itsCathyDi lol I think I've never seen this founder not wear a hoodie.

Maybe this means they're gmi

[Tony Chen @TToekneechin]

@cailynyongyong Damnnn start a paper reading group?

[Tony Chen @TToekneechin]

@karpathy Oh wow didn't know GPT2 was done on TPUs back then.

Was there a specific reason why?

[Tony Chen @TToekneechin]

@redtachyon How are you liking the DGX spark? Is it worth it? Would love to see a review and/or GitHub of ur setup

[Tony Chen @TToekneechin]

@livinoffwater @openclaw O what!! can you open-source the designs? And does it actually connect to openclaw?

[Tony Chen @TToekneechin]

@Miles_Brundage Haha let the researcher hillclimb