Rosanne Liu

@savvyRL · San Francisco, CA · joined 03 Mar 2013

Cofounded & running @ml_collective. Weekly talks at Deep Learning Classics & Trends (DLCT). Research at Google DeepMind. DEI/DIA Chair of ICLR & NeurIPS.

55 546Followers

1 070Following

4 944Posts total

243.9KViews on collected posts

Son gönderiler

Rosanne Liu@savvyRL2 Sep

@savvyRL The variable iteration depth is exactly the important distinction. For any fixed number of iterations, you can unroll the model into a deeper network with tied parameters. But a fixed unrolling has fixed computational depth, whereas a loop can use an input-dependent numb 1.3K views · 10 likes · 1 reposts · 3 replies Open on X →

Rosanne Liu@savvyRL2 Sep

We talked about this extensively in "The Topological Trouble With Transformers" (arxiv: 2604.17121) and in "Recirculation" (arxiv: 2608.17981) 3.2K views · 26 likes · 2 reposts · 1 replies Open on X →

Rosanne Liu@savvyRL2 Sep

However many loops you do, the transformer is still inherently feedforward, the only "recurrence" added here is "depth recurrence", which is the same as making the model deeper (with tied parameters). The only real "recurrence" is if you carry things *forward in time*. 3.4K views · 32 likes · 1 reposts · 1 replies Open on X →

Rosanne Liu@savvyRL2 Sep

Stop blowing up Looped Transformers — it's not magical, it doesn't add recurrence to the (still and always) feedforward network, it was done years ago (Dehghani et al., 2019). All it does is making your model deeper, the same as if you added more layers to begin with. 46.9K views · 346 likes · 15 reposts · 13 replies Open on X →

Rosanne Liu@savvyRL20 Aug

100%! The title we secretly want to publish it under is "Recirculation. OR: CoT is a cop out." 4K views · 20 likes · 0 reposts · 4 replies Open on X →

Rosanne Liu@savvyRL20 Aug

@mc_mozer Interesting distinction from chain-of-thought. CoT is the default for reasoning gains, but state tracking feels more fundamental. If feedforward depth limits state updates, inference-time recurrence is a natural fix. 7.7K views · 4 likes · 1 reposts · 2 replies Open on X →

Rosanne Liu@savvyRL19 Aug

18 years into paper publishing and I still sheepishly search for our new paper title to see if it's gotten picked up in the gap between arxiv posted and twitter announced 7.2K views · 90 likes · 1 reposts · 2 replies Open on X →

Rosanne Liu@savvyRL27 Jul

On a two week vacation aka full time childcaring with no pay no appreciation and 24h on call where your coworker is kinda a jerk but super duper cute 8.9K views · 40 likes · 0 reposts · 6 replies Open on X →

Rosanne Liu@savvyRL10 Jul 2025

@savvyRL @ml_collective @DeepIndaba I will forever be grateful for this noble effort. It is indeed a rare opportunity for the @ml_collective community 708 views · 6 likes · 0 reposts · 0 replies Open on X →

Rosanne Liu@savvyRL10 Jul 2025

This year we have 25 amazing researchers to sponsor. For the first time we ask them to film a video about themselves. It's nothing like the usual well-polished, professional fundraiser video you'd stumble upon, but give it a watch: https://t.co/zkX35g93mI 7.6K views · 40 likes · 15 reposts · 1 replies Open on X →

Rosanne Liu@savvyRL10 Jul 2025

We are raising $20k (which amounts to $800 per person), to cover their travel and lodging to Kigali, Rwanda in August, from either Nigeria or Ghana. Donate what you can here! https://t.co/ryCItIoxNs 14.7K views · 49 likes · 27 reposts · 0 replies Open on X →

Rosanne Liu@savvyRL10 Jul 2025

The opportunity gap in AI is more striking than ever. We talk way too much about those receiving $100M or whatever for their jobs, but not enough those asking for <$1k to present their work. For 3rd year in a row, @ml_collective is raising funds to support @DeepIndaba attendees.

138.3K views · 262 likes · 118 reposts · 22 replies Open on X →

Aynı büyüklükteki hesaplara karşı

Son 90 güne ait 8 gönderi, 10K–100K takipçi aralığıyla yan yana. geniş kitleye gösteriliyor, ama izleyenlerin azı tepki veriyor.

Medyan görüntülenme5 574bu hesap99510K–100K için medyan

Erişim, %10.03%bu hesap3.85%10K–100K için medyan

Etkileşim, %0.85%bu hesap1.52%10K–100K için medyan

Ölçüt	Bu hesap	10K–100K için medyan	Oran
Gönderi başına medyan görüntülenme	5 574	995	5.60×
Erişim (görüntülenme ÷ takipçi)	10.03%	3.85%	2.61×
Etkileşim oranı	0.85%	1.52%	0.56×

Bu aralıktaki diğer hesaplar → Başka bir hesapla karşılaştır → Bu kıyas değerleri nasıl kuruluyor →

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

138.3K10 Jul

14.7K

7.6K

708

8.9K27 Jul

7.2K19 Aug

7.7K20 Aug

46.9K2 Sep

3.4K

3.2K

1.3K

Last 12 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.29%10 Jul

0.52%

0.73%

0.85%

0.52%27 Jul

1.29%19 Aug

0.09%20 Aug

0.61%

0.80%2 Sep

0.99%

0.91%

1.06%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.

What the audience does

Likes68.5%925 in total

Reposts13.4%181 in total

Replies4.1%55 in total

Bookmarks14.0%189 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.

Son gönderiler

Aynı büyüklükteki hesaplara karşı

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Benzer hesaplar