Ceshine Lee 💙💛 @ceshine_en

Data Geek @ https://t.co/zD08GsUC6Y blog.ceshine.net Taiwan Joined December 2008

Tweets

1K
Followers

248
Following

539
Likes

7K

Simon Brown @simonbrown

2 years ago

The slides for my "The lost art of software design" can be found at static.architectis.je/devsum2023-the… #devsum

Andreas Wänqvist @mobilemancer

2 years ago

The slides for my "The lost art of software design" can be found at static.architectis.je/devsum2023-the… #devsum

3 18 87 111K 10

Download Image

11 105 537 104K 286

Gary Marcus @GaryMarcus

3 years ago

Speaking of bullshit, anyone remember this?

175 690 5K 694K 265

Download Image

When using automatic mixed-precision training to accelerate model training, there are two common options: float16 and bfloat16 (16-bit "brain" floating points). What's the difference? Compared to float16, bfloat16 has the same dynamic range as float32 but lower precision. In my…

10 55 397 106K 141

Download Image

Christoph Molnar 🦋 christophmolnar.bsky.social @ChristophMolnar

3 years ago

@marktenenholtz "To make this tutorial as real as possible, we created a chatbot simulating Bob. Bob is working in the fictional IT department and you will spend the first few weeks of the tutorial convincing Bob to finally give you access to the database"

2 8 134 9K 2

Tim Dettmers @Tim_Dettmers

3 years ago

In the RTX 40 post, I introduce a GPU recommendation chart and discuss the new Tensor Memory Accelerator (TMA) and FP8 computation. Overall, RTX 40s are faster for inference and shine through their FP8 performance but are inefficient for 16-bit training. timdettmers.com/2023/01/16/whi…

35 165 859 221K 263

Download Image

Andrej Karpathy @karpathy

3 years ago

Potentially nitpicky but competitive advantage in AI goes not so much to those with data but those with a data engine: iterated data aquisition, re-training, evaluation, deployment, telemetry. And whoever can spin it fastest. Slide from Tesla to ~illustrate but concept is general

65 390 3K 0 626

Download Image

kirsten lum @machsci

3 years ago

the decision maker didn’t even know the project was happening, and they are asking for a lot of changes! By this time, both the ML and domain expert feel pretty jaded — ML must not work for their organization, right? This is a symptom of a missing coordination expert: <6/13>

1 2 20 3K 2

Download Image

steven t. piantadosi @spiantado

3 years ago

Yes, ChatGPT is amazing and impressive. No, @OpenAI has not come close to addressing the problem of bias. Filters appear to be bypassed with simple tricks, and superficially masked. And what is lurking inside is egregious. @Abebab @sama tw racism, sexism.

441 2K 8K 0 1K

Download Image

Christoph Molnar 🦋 christophmolnar.bsky.social @ChristophMolnar

3 years ago

There's a confusing amount of plots for visualizing SHAP. I put together a cheat sheet with the most important SHAP plots (+interpretation) for explaining machine learning models. Get it here: christophmolnar.gumroad.com/l/shap-plots-f… It's free!

3 27 102 0 58

Conrad Hackett @conradhackett

3 years ago

How gerrymandering can change election outcomes, in one chart washingtonpost.com/blogs/wonkblog…

7 97 276 0 31

Download Image

John Burn-Murdoch @jburnmurdoch

3 years ago

NEW: income inequality in US & UK is so wide that while the richest are very well off, the poorest have a worse standard of living than the poorest in countries like Slovenia ft.com/content/ef2654… Essentially, US & UK are poor societies with some very rich people. A thread:

992 19K 41K 0 5K

Download Image

Andrej Karpathy @karpathy

3 years ago

Reading through OpenAI Whisper paper github.com/openai/whisper some notes:

35 412 2K 0 808

Download Image

Andrew Gelman et al. @StatModeling

3 years ago

“You should always (always) have a Naive model. It’s the simplest, cleanest, most intuitive way to explain whether your system is at least treading water. And if it is (that’s a big IF), how much better than Naive is it.” statmodeling.stat.columbia.edu/2022/09/08/you…