Sharon Zhou ✓

@realSharonZhou · Stanford, CA · joined 08 Nov 2016

Recursively self-improving | VP Eng & AI @AMD | Prev: Founder/CEO, Lamini. CS Faculty & PhD @Stanford. @Harvard | @MIT 35 under 35. Angel investor.

27 784Followers

1Following

1 506Posts total

332.3KViews on collected posts

Against accounts of the same size

4 posts from the last 90 days, next to the 10K–100K follower range. shown widely, but few of those viewers react.

Median views16 496this account964median for 10K–100K

Reach, %59.37%this account3.25%median for 10K–100K

Engagement, %1.24%this account2.01%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	16 496	964	17.1×
Reach (views ÷ followers)	59.37%	3.25%	18.3×
Engagement rate	1.24%	2.01%	0.62×

Others in this range → Compare with another account → How these benchmarks are built →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

243.9K10 Mar

2.9K

723

2.3K

21.8K5 Jul

5.4K22 Jul

44K14 Aug

11.2K1 Sep

Last 8 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.44%10 Mar

0.31%

0.55%

0.39%

0.43%5 Jul

1.00%22 Jul

1.48%14 Aug

1.59%1 Sep

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 2.01%.

What the audience does

Likes61.9%1 809 in total

Reposts5.0%145 in total

Replies3.3%96 in total

Quotes0.7%19 in total

Bookmarks29.2%854 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

there is about to be a large wave of algorithms, architecture changes, and performance optimizations that trade more flops for less memory because the relative cost of flops per token to hbm per token is dropping. excited to see what we invent! 11.2K views · 159 likes · 9 reposts · 9 replies 01 Sep 2026 · Open on X →

NVIDIA CUDA and AMD ROCm are software. And software is free. Anthropic just ran Claude for a weekend and it got a version of itself running on GPUs that it's never seen. Every compute customer (frontier lab, neolabs, etc.) will just generate their own kernels on the fly for their 44K views · 561 likes · 47 reposts · 38 replies 14 Aug 2026 · Open on X →

Anthropic is an AMD customer and will be making ~2GW of AMD GPUs go brrr! (That’s hundreds of thousands of GPUs) This is partly possible bc Claude can write Claude’s own kernels, and will only get better down the stack as many tasks to optimize chips are verifiable 5.4K views · 52 likes · 1 reposts · 1 replies 22 Jul 2026 · Open on X →

We need better GPU performance benchmarks for RLing capabilities into frontier models for RSI, and I have a few ideas. The goals would be to 1) help models improve their own compute efficiency (basically “living” RL envs with kernels that matter for new model generations AND 21.8K views · 74 likes · 11 reposts · 8 replies 05 Jul 2026 · Open on X →

@realSharonZhou Focus on composability of inference optimizations for ROCm stack in order to defeat the CUDA moat. (wide EP + disagg + FP4 + KV offloading + more) for maximum efficiency/scaling on huge models. AMD/ROCm lags in making those combinations smooth 2.3K views · 8 likes · 0 reposts · 1 replies 10 Mar 2026 · Open on X →

@realSharonZhou There goes Nvidia's moat! 723 views · 4 likes · 0 reposts · 0 replies 10 Mar 2026 · Open on X →

@realSharonZhou Kernel wins are real only if they hold across model families. What are the p95 speedup and numerical-drift deltas on held-out production workloads? 2.9K views · 6 likes · 0 reposts · 3 replies 10 Mar 2026 · Open on X →

It's here: We just hit superhuman performance on AI kernel optimization! Real customer models & production settings. Not toy problems (what I typically see). This is the year that Claude writes its own kernels, Codex its own kernels, for every new GPU that it wants to run o 243.9K views · 945 likes · 77 reposts · 36 replies 10 Mar 2026 · Open on X →