Phillip Isola

@phillip_isola · joined 04 Dec 2016

Associate Professor in EECS at MIT, trying to understand intelligence.

21 651Followers

178Following

899Posts total

278.2KViews on collected posts

Against accounts of the same size

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

Median views14 021this account1 195median for 10K–100K

Reach, %64.76%this account3.80%median for 10K–100K

Engagement, %0.73%this account2.00%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	14 021	1 195	11.7×
Reach (views ÷ followers)	64.76%	3.80%	17.0×
Engagement rate	0.73%	2.00%	0.37×

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

20.3K2 Jul

11.8K3 Jul

30.1K7 Jul

14K

14.3K

134.9K14 Aug

41.2K15 Aug

334

151

334

10.7K29 Aug

Last 11 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.56%2 Jul

0.78%3 Jul

0.48%7 Jul

1.70%

0.73%

0.47%14 Aug

0.98%15 Aug

0.60%

0.66%

1.50%

1.14%29 Aug

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 2.00%.

What the audience does

Likes65.9%1 642 in total

Reposts6.6%164 in total

Replies1.6%39 in total

Quotes0.9%22 in total

Bookmarks25.1%624 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

An interesting thing about living on an exponential is that "this time is different" is in fact always true. 10.7K views · 115 likes · 5 reposts · 1 replies 29 Aug 2026 @phillip_isola @CSProfKGD These are nice results but 75 hours is not a massive amount of data. Maybe it’s a better claim to say you did something fairly efficient? Using sim as pretraining could be a big lift in data collection efficiency if it works well. In my experience real d 334 views · 4 likes · 0 reposts · 1 replies 15 Aug 2026 @phillip_isola Amazing! Did you try deploying the pre-trained model directly? It would be interesting to know how model behavior changed after post-training. 151 views · 1 likes · 0 reposts · 0 replies 15 Aug 2026 @phillip_isola This is cool! Did you have to pre-train on lots of different textures or occlusions for this to work well in the real world? Or lots of different physical conditions (e.g friction, etc)? 334 views · 1 likes · 0 reposts · 1 replies 15 Aug 2026 There's a really simple recipe behind much of modern AI: 1. Pre-train on massive proxy data 2. Post-train on a small amount of the real thing Here we apply this recipe to robot hands: 1. Pre-train on lots of sim 2. Post-train on a bit of real Very simple and works! 41.2K views · 367 likes · 30 reposts · 7 replies 15 Aug 2026 Excited to share SPD: simulation pre-training for dexterity. We pre-trained a policy in simulation and fine-tuned with less than 2 hours of real data (with @sarthakkamat) https://t.co/1majN5fYj5 134.9K views · 535 likes · 71 reposts · 17 replies 14 Aug 2026 Two common q's about RandOpt: 1. What if you repeat the process? 2. Can you avoid K-pass inference? We tried: guess --> ensemble --> distill --> repeat It works, and results in a final 1-pass inference model with good preformance. Code here: https://t.co/inXmt4sC3Z 14.3K views · 98 likes · 7 reposts · 0 replies 07 Jul 2026 At ICML, I'll present our Neural Thickets work today: 2:30–4:15 PM KST Hall A #2602 Come by if you have questions or want to chat! https://t.co/BmuTHoFekS 14K views · 221 likes · 15 reposts · 2 replies 07 Jul 2026 An update to the Neural Thicket paper: We avoid the K× inference cost of RandOpt ensembling and make RandOpt iterative: a single model with strong performance. Like policy gradients, but with search in weight space and stepwise supervision. Like ES, but with updates through htt 30.1K views · 120 likes · 17 reposts · 5 replies 07 Jul 2026 It was a lot of fun to be on this podcast. Thanks for hosting me @ziv_ravid and Allen Roush! Contains some of my recent thoughts on representation learning, the platonic hypothesis, neural thickets, and a few newer directions I'm thinking about like LLMs as artificial life. 11.8K views · 82 likes · 7 reposts · 3 replies 03 Jul 2026 New episode of the Information Bottleneck with Phillip Isola @phillip_isola (MIT) is out 🥳🥳🥳 - What makes a good representation? Phillip has been a legend for me for many years. So much of how I think about representation learning comes from his work, so getting to ask him ht 20.3K views · 98 likes · 12 reposts · 2 replies 02 Jul 2026