Seohong Park

@seohong_park

Reinforcement Learning | CS Ph.D. Student @berkeley_ai | Physical Intelligence

5 781Followers

585Following

527Posts total

154.9KViews on collected posts

Against accounts of the same size

4 posts from the last 90 days, next to the under 10K follower range. below its peers on both reach and engagement.

Median views1 510this account3 830median for under 10K

Reach, %26.11%this account256.48%median for under 10K

Engagement, %0.77%this account1.26%median for under 10K

Metric	This account	Median for under 10K	Ratio
Median views per post	1 510	3 830	0.39×
Reach (views ÷ followers)	26.11%	2.6× audience	0.10×
Engagement rate	0.77%	1.26%	0.61×

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

151K24 Aug

1.3K

966

1.7K26 Aug

Last 4 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.79%24 Aug

0.75%

0.83%

0.54%26 Aug

Reactions — likes, reposts, replies and quotes — divided by views. Median for under 10K accounts is 1.26%.

What the audience does

Likes48.4%1 027 in total

Reposts6.1%129 in total

Replies1.9%40 in total

Quotes1.0%22 in total

Bookmarks42.6%905 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

@seohong_park Very insightful post! Five years ago, we noticed exactly the same phenomenon mentioned in this blog in autonomous driving -- the imitation agent tends to copy the previous actions. We called it "copycat problem". 1.7K views · 6 likes · 0 reposts · 3 replies 26 Aug 2026 @seohong_park Definitely vibe with a compounding errors perspective. Part of my work with @max_simchowitz (https://t.co/VPFrtszFIi) precisely shows that reducing frequency of feedback (eg action chunking) is key to stable deployment. This can be true even if your demos are Markov 966 views · 7 likes · 0 reposts · 1 replies 24 Aug 2026 @seohong_park Good paper on why action chunking works: https://t.co/B7dQIyoBGd It seems to me it's because human generated trajectories are non-Markovian. Maybe because humans plan and execute at different frequencies. But curious if the same phenomenon occurs on synthetic data. 1.3K views · 10 likes · 0 reposts · 0 replies 24 Aug 2026 Behavioral cloning mystery https://t.co/VqxzvcfGSx I wrote a new blog post about "mysteries" in behavioral cloning that appear with real-world robot data (e.g., overfitting is "good"). I also tried to demystify them and shared my thoughts! https://t.co/BF1ZeYPpLH 151K views · 1K likes · 129 reposts · 36 replies 24 Aug 2026