tweetindex
RU

Yuchen Jin

@Yuchenj_UW · another Galaxy · joined 22 Nov 2016

Cooking fun AI systems & products @databricks. Prev: co-founder & CTO @ Hyperbolic, OctoAI (acquired by @nvidia) Apache TVM, PhD @ University of Washington.

287 590Followers
836Following
14 229Posts total
3MViews on collected posts

Последние посты

OpenAI’s “significantly more capable than GPT-6 Astra” model solving a Millennium Prize Problem is huge, an AlphaGo moment for math. AI is now smarter than a math PhD, but dumber than an intern in many tasks. “Jagged intelligence” needs to be solved. 9.2K views · 296 likes · 18 reposts · 22 replies Open on X →
OpenAI’s comeback is the biggest comeback after Meta’s comeback, after xAI’s comeback, after Google’s comeback. AI is an endless cycle of “it’s over” and “we’re so back.” 46.7K views · 1.3K likes · 52 reposts · 79 replies Open on X →
Sam and Jensen say “AGI has arrived.” But “AGI” means different things depending on who you ask. - Sam’s bar: outperform humans at most economically valuable work. - Dario and Demis’s bar: smarter than a Nobel Prize winner across most fields. These are wildly different bars. 54.5K views · 900 likes · 39 reposts · 112 replies Open on X →
If you want to understand how fast AI is improving, look at intelligence per dollar. 1.5 years ago: o1 Pro: $150 / $600 per M tokens Today: GLM-5.3 Flash: $0.15 / $0.50 That’s a ~1000x collapse in price in under 1.5 years, while GLM-5.3 Flash is more intelligent than o1 Pro. 411K views · 5.4K likes · 314 reposts · 133 replies Open on X →
@Yuchenj_UW It’s interesting that you can 3X the LR. You’d expect the original paper to be well tuned near what is tolerable. 44.3K views · 138 likes · 1 reposts · 3 replies Open on X →
@karpathy I'm unsure why the evaluation score dropped near the end. It might be due to some bad training data batch. @karpathy, would love to hear your insights. It reminds me of the lossfunctions Tumblr though lol: https://t.co/9MQivsrCPI. 10.3K views · 27 likes · 2 reposts · 0 replies Open on X →
Outperform GPT-3 with @karpathy's llm.c using just 1/3 training tokens ✨ Another day has passed, and I trained GPT-2 (124M) with llm.c for 150B tokens, achieving 35.5% accuracy on HellaSwag. This surpasses the GPT-3 paper’s 33.7% accuracy trained for 300B tokens. It matched the
2.4M views · 1.6K likes · 137 reposts · 126 replies Open on X →

На фоне аккаунтов своего размера

4 постов за последние 90 дней рядом с диапазоном 100K–1M подписчиков. показывается далеко за пределами своей аудитории, и она отвечает.

Медианные просмотры50 598этот аккаунт4 981медиана для 100K–1M
Охват, %17.59%этот аккаунт1.73%медиана для 100K–1M
Вовлечённость, %2.57%этот аккаунт1.22%медиана для 100K–1M
ПоказательЭтот аккаунтМедиана для 100K–1MОтношение
Медианные просмотры на пост50 5984 98110.2×
Охват (просмотры ÷ подписчики)17.59%1.73%10.2×
Вовлечённость2.57%1.22%2.10×

Другие в этом диапазоне →   Сравнить с другим аккаунтом →   Как считаются эти ориентиры →

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

2.4M1 Jun
10.3K
44.3K2 Jun
411K6 Sep
54.5K7 Sep
46.7K
9.2K8 Sep

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

0.08%1 Jun
0.28%
0.32%2 Jun
1.43%6 Sep
1.96%7 Sep
3.17%
3.65%8 Sep

Reactions — likes, reposts, replies and quotes — divided by views. Median for 100K–1M accounts is 1.22%.

What the audience does

Likes75.6%9 646 in total
Reposts4.4%563 in total
Replies3.7%475 in total
Quotes0.8%108 in total
Bookmarks15.5%1 973 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.

Похожие аккаунты