Marvin Tong ✓

@marvin_tong · San Francisco, CA · joined 30 Sep 2015

do things for AI privacy : Trust Machines @PhalaNetwork / Privacy LLM @redpill_gpt / 🦞 icloud for AI agent: https://t.co/4MpkGHHOhb /

14 613Followers

4 415Following

4 779Posts total

20.6KViews on collected posts

Against accounts of the same size

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

Median views1 372this account1 266median for 10K–100K

Reach, %9.39%this account3.99%median for 10K–100K

Engagement, %0.83%this account1.98%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	1 372	1 266	1.08×
Reach (views ÷ followers)	9.39%	3.99%	2.35×
Engagement rate	0.83%	1.98%	0.42×

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

13.7K25 Apr

1.2K28 Aug

2K31 Aug

1.6K

1.4K2 Sep

8364 Sep

Last 6 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.74%25 Apr

0.51%28 Aug

0.67%31 Aug

0.83%

0.95%2 Sep

1.91%4 Sep

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.98%.

What the audience does

Likes75.3%128 in total

Reposts4.1%7 in total

Replies12.4%21 in total

Quotes3.5%6 in total

Bookmarks4.7%8 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

modern inference’s quality-efficiency frontier increasingly comes down to cache hit rate. Infra should be built around maximizing it. Encryption makes cache reuse extremely hard in private inference—but we’re breaking through that barrier. https://t.co/EuYTZ1ZL3L 836 views · 15 likes · 1 reposts · 0 replies 04 Sep 2026 Congrats! 1.4K views · 9 likes · 0 reposts · 4 replies 02 Sep 2026 If you are using H200 to inference GLM 5.3, try this model version! 1.6K views · 11 likes · 0 reposts · 2 replies 31 Aug 2026 Benchmark checks: 91.92 GPQA-Diamond (182/198), 82.2 on a 45-item BFCL subset, and 3/3 NIAH retrieval at ~930k-token prompts. Weights, serving config, and per-item evaluation artifacts: https://t.co/F8oBJf7zkp 2K views · 12 likes · 0 reposts · 0 replies 31 Aug 2026 Common misunderstanding. They are not subsidizing, it’s just very good infrastructure work from top inference teams. If you can make cache hit rate 99.5% with over 500m TPM with 99% online status, you could pricing low too. 1.2K views · 6 likes · 0 reposts · 0 replies 28 Aug 2026 I run 2 companies with 5 AI agents every day: Claude Code, IDE+CC, Codex, OpenClaw, Hermes. And I fucking hate it. https://t.co/eW0bJAJJtS 13.7K views · 75 likes · 6 reposts · 15 replies 25 Apr 2026