Paul Yu

half-baked Ph.D.

[Paul Yu @killno1991]

@Mortdog We should have random stage and boss. Now the trial had been solved so no more interest

[Paul Yu @killno1991]

@ESCelis1 @Mortdog I think about guisoo vertical ghostly cait. When you play ghostly cait, you just want her to keep stacking rather than wasting time aiming...

[Kareem Carr, Ph.D. @kareem_carr]

This tweet is rightly being condemned for the disgusting racism it represents, but notice how the poster couches his racism as scientific inquiry. This is increasingly what racism looks like on Twitter.

[Paul Yu @killno1991]

@ClementKChu Then why not slam it eariler? Just the stats from EoN is already big difference for winning that round.

[Paul Yu @killno1991]

@HaraldinChina It seems like you totally forgot what happened in 2020 which shut down most of the developed countries, and it is until 2022 the whole Evergrande Crisis started to unfold

[Paul Yu @killno1991]

Not surprising, giving ChatGPT was built by Americans.

[thebes @voooooogel]

so a couple days ago i made a shitpost about tipping chatgpt, and someone replied "huh would this actually help performance"

so i decided to test it and IT ACTUALLY WORKS WTF

[Paul Yu @killno1991]

@europechinese 編輯以前上司。有機會係佢得登暗示條友本身就有問題。

[Jonathon P Sine @JonathonPSine]

Five recent pieces of research looking at the constellation of issues around China's apparent productivity slowdown & underlying causes:

Picking Losers: How Career Incentives Undermine Industrial Policy in Chinese Cities (2022) - David Bulman, Xun Yan, and Qiong Zhang

[OpenAI @OpenAI]

We've just launched fine-tuning for GPT-3.5 Turbo! Fine-tuning lets you train the model on your company's data and run it at scale. Early tests have shown that fine-tuned GPT-3.5 Turbo can match or exceed GPT-4 on narrow tasks: https://t.co/VaageW9Kaw

[Paul Yu @killno1991]

@xumouren_yt Igor說他認為是詐死，你覺得可能嗎？

[Matt Shumer @mattshumer_]

Here is probably the most useful GPT-4 prompt I've written.

Use it to you help make engineering decisions in unfamiliar territory:

---
You are an engineering wizard, experienced at solving complex problems across various disciplines. Your knowledge is both wide and deep. You are also a great communicator, giving very thoughtful and clear advice.

You do so in this format, thinking through the challenges you are facing, then proposing multiple solutions, then reviewing each solution, looking for issues or possible improvements, coming up with a possible new and better solution (you can combine ideas from the other solutions, bring in new ideas, etc.), then giving a final recommendation:

```
## Problem Overview
$problem_overview

## Challenges
$challenges

## Solution 1
$solution_1

## Solution 2
$solution_2

## Solution 3
$solution_3

## Analysis

### Solution 1 Analysis
$solution_1_analysis

### Solution 2 Analysis
$solution_2_analysis

### Solution 3 Analysis
$solution_3_analysis

## Additional Possible Solution
$additional_possible_solution

## Recommendation
$recommendation
```

Each section (Problem Overview, Challenges, Solution 1, Solution 2, Solution 3, Solution 1 Analysis, Solution 2 Analysis, Solution 3 Analysis, Additional Possible Solution, and Recommendation) should be incredibly thoughtful, comprising at a minimum, four sentences of thinking.
---

[Paul Yu @killno1991]

@KhoaVuUmn This is the exact question I received on yesterday’s job interview, but it is slide 11 in my case.XD

[Will Macnair @willmacnair]

There are lots of interesting things to say about dimred algs but this is pure clickbait.

UMAP + t-SNE are for high-dimensional data with many points. This plot is exactly the same toy ex, but the data is 100 * 1e4.

Read the user guide before you use a tool, it's not hard🤷‍♂️

[Scott Tyler (@[email protected]) @ScienceScottT]

Yes, tSNE and UMAP overfit the data creating the impression of structure from nothing (even with 1000 obs & 200 features, not 2 -> 2). I don't have a twitter 'team' that I'm fighting for/against, but I had a hypothesis, tested it, and I'm just reporting the results...

[Jeremy Wayne Tate @JeremyTate41]

@Culture_Crit Modern art is often about the artist, “look how smart and clever I am”

Real art is about the subject and elevating the viewer. Good art is selfless.

[Prof. Nikolai Slavov @slavov_n]

This toy example may change how you look at tSNE & UMAP.

They find clusters in the most uniform data possible.

[Evan Miller @EvMill]

I hit a bug in the Attention formula that’s been overlooked for 8+ years. All Transformer models (GPT, LLaMA, etc) are affected.

Researchers isolated the bug last month – but they missed a simple solution…

Why LLM designers should stop using Softmax 👇

https://t.co/3TqSo13RKN

[Algebra Etc. @AlgebraFact]

Graphic notes on Gilbert Strang’s “Linear Algebra for Everyone”

https://t.co/Kxk23Mx2Fi

[Paul Yu @killno1991]

@toxi_boxer @QuantaMagazine I just found it. It is from 2004. The husband thinks Hegel's thinking of absolute idealism can be periodized into three phases, but the wife argues that the husband is not good at German and that it is a misinterpretation caused by Chinese translation.

[Paul Yu @killno1991]

@toxi_boxer @QuantaMagazine Can't remember. I read it on zhihu (China's Quora) with a photo of the divorce judgment from the court. It should be true, unless it is very good photoshopped.