elie ✓

@eliebakouch · joined 12 Jan 2024

training llm @PrimeIntellect (prev: @huggingface) anon feedback: https://t.co/JmMh7Sg3mL

24 114Followers

4 540Following

8 164Posts total

858.6KViews on collected posts

Against accounts of the same size

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

Median views31 519this account2 896median for 10K–100K

Reach, %130.71%this account9.51%median for 10K–100K

Engagement, %0.92%this account1.56%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	31 519	2 896	10.9×
Reach (views ÷ followers)	130.71%	9.51%	13.7×
Engagement rate	0.92%	1.56%	0.59×

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

588.5K15 Aug

105.8K

105.8K2 Sep

16.5K

31.5K

3.8K

6.7K

Last 7 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.39%15 Aug

0.92%

0.80%2 Sep

1.28%

0.47%

2.20%

1.22%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.56%.

What the audience does

Likes75.5%3 925 in total

Reposts6.8%352 in total

Replies3.9%201 in total

Quotes2.5%129 in total

Bookmarks11.4%591 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

long context benchmark (MRCR) going crazy, very interesting would be great to have other benchmarks such as graphwalks for long context and release test time scaling curves and not just raw benchmark scores in general https://t.co/yfLD2RDlk3 6.7K views · 75 likes · 0 reposts · 7 replies 02 Sep 2026 it has always been clear that when you scale model size, you need to scale depth (with "recurrence" or not) which makes the model more reasoning efficient. openai and (a bit less) anthropic are also clearly training models to be more reasoning efficient and not less for obvious h 3.8K views · 71 likes · 5 reposts · 7 replies 02 Sep 2026 note that the latest big model by oai with architecture details was already very deep by today's standards, here is the number of layers for a few models Llama-3.1 405B: 126 GPT3: 96 Kimi K3: 93 Qwen 3.8 max: 92 GLM-5.3: 78 DeepSeek-V4-Pro: 61 gpt-oss 120b: 36 gpt-oss 20b: 24 ht 31.5K views · 135 likes · 4 reposts · 7 replies 02 Sep 2026 this is not a "bitter lesson" pilled way to scale transformers, you're introducing a structural bias by looping blocks instead of having more unique weights and letting them specialize as they want 16.5K views · 189 likes · 6 reposts · 16 replies 02 Sep 2026 a few thoughts on "recurrent depth" transformers the main question: why recurrent depth instead of just scaling depth? it's not faster at inference or training* since you still go through the full "effective depth", the advantage is storage (for kv cache storage btw you could do 105.8K views · 743 likes · 48 reposts · 39 replies 02 Sep 2026 to my knowledge this is the largest open experiment on autonomous agents iterating on a research environment we scaled runtime, compute, diversity of models and harnesses. as a comparison, similar tasks on oai/anthropic system cards are anthropic "optimizing an llm training on h 105.8K views · 831 likes · 76 reposts · 51 replies 15 Aug 2026 We ran the largest open experiment on how frontier models do AI research. 100+ autonomous runs across 10+ models, sandboxed on 8xH200s for up to 8 days, iterating on the nanoGPT optimizer track. Best runs closed 82% of the gap to a record built by dozens of humans over months. 588.5K views · 1.9K likes · 213 reposts · 74 replies 15 Aug 2026