Miffy Pan
miffypan.bsky.social
Miffy Pan
@miffypan.bsky.social
Half-rabbit half-bread navigating life with caution

I'm tanned but I love wearing sunscreen
新しいデータを使ってモデルを再トレーニングしても、結果的にパフォーマンスが下がった

機械学習、学んでる時にありえないことは実用的にはありえる物だ
October 18, 2025 at 12:59 AM
とりあえず頑張ろう。そういう思いを込めて嫌なことを扱う

今日の嫌なことはコスト増加の原因だった
October 13, 2025 at 11:52 PM
機械学習とはモデルとデータの組立で意外な結果が出る時はどこから探り始めたらいいって聞かれたら、確実的に私はデータから疑う
October 10, 2025 at 10:46 PM
面接で候補者がコードの中で小さなシンタックスミスをしていた
今の時代では毎日プログラミングしてもオートコンプリートに頼る結果なのではないかと思った
October 9, 2025 at 12:15 AM
最近cursor-agentにハマってる
現在の形はミッフィーだけど今度こそぐでたまになるのがマジ怖い
October 8, 2025 at 12:58 AM
データを探る探るほど異変なことが必ず現れる。

何十分かけて、行数を確認した。あるフィルターのこと完全に忘れてた。そうしたら、サンプリング頻度が違うと気づいた。そう知らなかった私が幸せだった。
October 6, 2025 at 10:19 PM
反応早くしたいことが私の弱点だ

少し迷っても前へ進みたいという気持ちになって悪い決断に至る

そして必ず後悔する
October 2, 2025 at 12:20 AM
他のチームからの助っ人が来るって言われて、それがいいかどうかまだ分からない。不安な思いの中で明日を待ってる
October 1, 2025 at 1:23 AM
初めて新しいプロジェクトにuvを使った。便利で早くて感動したんだ
September 30, 2025 at 1:27 AM
午後3時から停電。スマホをホットスポットとしてインターネットを使い、無事に仕事が出来た。結果的にたった2時間以内に1.5GBを焼き尽くした
September 27, 2025 at 1:46 AM
データを見ながら楽しい、ありえないくらい説明を考えるのが好き

私の話を聞いて笑わない人、訳分からない
September 26, 2025 at 12:22 AM
ある問題があって、すごく頑張って他のチームの手を借りることができた。すぐ後に同僚が自分で問題を解決した。もちろん嬉しいけど力になれなくて寂しい
September 24, 2025 at 11:29 PM
Slackメッセージを早く読んだら、1つ目のパラグラフを完全にミスした。上手く分からなかった部分を確認しようと思ったら、同僚に1つ目のパラグラフを送られたため恥ずかしいな〜と思った
September 23, 2025 at 12:47 AM
Reposted by Miffy Pan
we gotta start building the AI that does the interviews for candidates fortune.com/2025/08/03/a...
AI is doing job interviews now—but candidates say they'd rather risk staying unemployed than talk to another robot
Job-seekers tell Fortune they’re outright refusing to do AI interviews, calling them dehumanizing and a red flag for bad company culture.
fortune.com
August 4, 2025 at 1:22 PM
Reposted by Miffy Pan
What are your favorite recent papers on using LMs for annotation (especially in a loop with human annotators), synthetic data for task-specific prediction, active learning, and similar?

Looking for practical methods for settings where human annotations are costly.

A few examples in thread ↴
July 23, 2025 at 8:10 AM
Reposted by Miffy Pan
Opinion: There is enormous potential for AI agents, but only if we deploy them responsibly.
Don’t let hype about AI agents get ahead of reality
There is enormous potential for this technology, but only if we deploy it responsibly.
www.technologyreview.com
July 3, 2025 at 12:15 PM
ChatGPT word at work for me is "enhance" - now I can't read it without thinking it was written by AI
Excellent story in @theverge.com today on the impact of AI-Mediated Communication. Quotes me but the best line is the closer from the writer, Sara Parker.
You sound like ChatGPT
AI isn’t just impacting how we write — it’s changing how we speak and interact with others. And there’s only more to come.
www.theverge.com
June 24, 2025 at 1:52 AM
Reposted by Miffy Pan
The company claims its ability to tackle complex, multistep problems paves the way for much more proficient AI agents.
Anthropic’s new hybrid AI model can work on tasks autonomously for hours at a time
The company claims its ability to tackle complex, multistep problems paves the way for much more proficient AI agents.
www.technologyreview.com
May 22, 2025 at 6:26 PM
Who else is sad that Pocket shutting down? :((((
May 22, 2025 at 11:57 PM
Reposted by Miffy Pan
This term I'll be teaching a graduate course on Human-AI Interaction. Check out the reading list and let me know if I am missing any of your favorites:

docs.google.com/document/d/1...

#HAI #Human-AI #HumanComputerInteraction
CS 889 - W25
[ CS 889 | W 2025] University of Waterloo © Anamaria Crisan (2025) Interfaces for Human-AI Interaction CS889 // Winter 2025 Instructor: Anamaria Crisan (ana.crisan@uwaterloo.ca) Location: TBD Off...
docs.google.com
January 3, 2025 at 10:45 PM
Reposted by Miffy Pan
some OpenAI guy says that o1 is better at most humans at most tasks

o1, make me a coffee? fail.
o1, brush my teeth? fail.
o1, drive me to work? fail.


o1 can do literally none of the things I do within the first hour of waking up. but, sure. you're great. never change 😀
December 10, 2024 at 3:17 AM
Everyone thinks modeling is cool but the truth is data scientists spend most of the time looking at data 🫠 It can really feel never-ending when it takes time to query datasets, to validate data, to do EDA -- rinse and repeat
November 23, 2024 at 1:21 AM
My cleaning robot lost its skills to go to other rooms because I re-arranged my furniture. The easy fix was to move its home to another room. Now I have a clean apartment again :)
November 18, 2024 at 11:21 PM
As someone working with AI, I dread the future in which AIs will be all over customer support, no matter the channel. The beauty of talking to random human customer support agents is that if an issue is solvable, someone will do it as long as if you try enough. With AI, I'm not so sure...
November 5, 2024 at 4:01 AM