alex zhang ✓

@a1zhang · USA · joined 24 Dec 2015

phd student @mit_csail @nlp_mit, previously undergrad @princeton 🫵🏻 go participate in the @GPU_MODE kernel competitions!

38 007Followers

1 021Following

1 230Posts total

370KViews on collected posts

Against accounts of the same size

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

Median views9 525this account2 043median for 10K–100K

Reach, %25.06%this account6.16%median for 10K–100K

Engagement, %0.88%this account1.59%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	9 525	2 043	4.66×
Reach (views ÷ followers)	25.06%	6.16%	4.07×
Engagement rate	0.88%	1.59%	0.55×

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

6.4K24 Aug

9.5K

236.1K

5.1K

4.3K

2.3K

17.6K25 Aug

9.8K

16.4K

10.5K26 Aug

9.5K28 Aug

8.9K

33.5K3 Sep

Last 13 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.78%24 Aug

1.05%

0.88%

0.70%

1.40%

0.90%

0.57%25 Aug

0.85%

0.60%

1.17%26 Aug

0.25%28 Aug

0.94%

0.93%3 Sep

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.59%.

What the audience does

Likes52.1%2 762 in total

Reposts4.9%262 in total

Replies1.6%86 in total

Quotes1.1%60 in total

Bookmarks40.2%2 132 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

actually coolest thing I've read in a long time (and + points for being well written and to the point!) I guess a natural question in the language domain is whether a natural clustering of "tasks" is obvious enough to a human s.t. doing search over new archs has some clear goal 33.5K views · 300 likes · 8 reposts · 2 replies 03 Sep 2026 A bit different than my usual research content but a friend who has poured a lot into her indie game has finally released it! They don't have a huge Twitter following but I wanted to shout them out! Go check it out, it's a really engaging game :) 8.9K views · 76 likes · 3 reposts · 4 replies 28 Aug 2026 Ventreville: A Cure for Sorrow is AVAILABLE NOW! We’ve come a long way from friends playing an original Dungeons & Dragons campaign to an indie studio crafting our mystery into a video game. Check out Ch. 1 on Steam 🕵️🥐🧪 https://t.co/dYJ54juugU https://t.co/6L7RoHL308 9.5K views · 20 likes · 2 reposts · 0 replies 28 Aug 2026 alphaXiv had this up day 1, they now support blogs! https://t.co/USLkKLrsx5 10.5K views · 113 likes · 7 reposts · 3 replies 26 Aug 2026 “Speculative Programmatic Tool Calling” Code agents waste time waiting for the LLM to finish generating code before tool calls can start. So this paper speculatively executes predictable tool calls from partial generations, while a shadow REPL tracks dependencies safely. This 16.4K views · 84 likes · 8 reposts · 3 replies 25 Aug 2026 sneak peek of a chat with @CShorten30 :) 9.8K views · 72 likes · 7 reposts · 4 replies 25 Aug 2026 Speculative Programmatic Tool Calling (sPTC) Explained! 🔥 👉 https://t.co/ALWTSyO7q7 https://t.co/ru324hkFO4 17.6K views · 82 likes · 11 reposts · 5 replies 25 Aug 2026 @a1zhang https://t.co/kbbGunxX0F 2.3K views · 17 likes · 1 reposts · 3 replies 24 Aug 2026 The codebase is very simple, and you can easily build on top of it for your own plugins. We provide an example of patching the RLM, which will be included in the RLM codebase. https://t.co/1UHINaT2M9 4.3K views · 57 likes · 3 reposts · 1 replies 24 Aug 2026 We talk a bit in the blog about what you can safely speculate on, and how. We run a "shadowed" REPL with its own namespace to resolve input dependencies when necessary. These rules will likely evolve over time, but are as simple as possible for now. https://t.co/IAp1MOnqUJ 5.1K views · 33 likes · 2 reposts · 1 replies 24 Aug 2026 Introducing Speculative Programmatic Tool Calling (sPTC)! A general class of technique for speculating on tool calls during code generation in a harness and queuing them early to overlap with token generation + REPL execution time. Blog: https://t.co/0nzkLvTXNy https://t.co/SVH 236.1K views · 1.8K likes · 204 reposts · 58 replies 24 Aug 2026 We want to speculate in two cases: 1. To overlap with the LLM streaming outputs, especially when thinking. 2. To overlap with actual REPL execution time, which can be expensive. This can be thought of as JIT compiling when you have stronger priors about tool calls. https://t.co/z 9.5K views · 94 likes · 4 reposts · 1 replies 24 Aug 2026 We run a small experiment over the RLM on information-dense tasks, and focus on the sub-LLM call as the tool of interest. These runs are high variance, and over 5 runs we find 1-1.2x gains in speed on similar trajectories. We also test over more deterministic examples in the ht 6.4K views · 47 likes · 2 reposts · 1 replies 24 Aug 2026