jake

[jake @UKYOUKAY]

@pika2761 https://t.co/NQaOd5xZ4I

[jake @UKYOUKAY]

@pika2761 量子コンピュータはノイズ耐性や量子アルゴリズム開発に課題が多く、実用化には時間を要します。また、量子コンピュータが得意とする計算は限定的で、AGIが完成したとしても、GPUは既存AIやシミュレーション、グラフィックス処理など広���な分野で必要とされ、その役割がなくなることは考えにくいです。

[jake @UKYOUKAY]

@mamas16k

[jake @UKYOUKAY]

@6de7472mo74 @umiyuki_ai 主任AI研究者だとそれくらいある

[jake @UKYOUKAY]

@umiyuki_ai https://t.co/QzQgmcMjQD

[jake @UKYOUKAY]

@masa_0083 Stashipの残骸の大気圏突入についてどう思われますか
次回飛ぶかどうかの話になってくるかと

[jake @UKYOUKAY]

@POLITICS_202405 @hirotaturou @YahooNewsTopics では一度話を戻ってタラレバの話をしても無駄だという根拠を示していただきたいです

[jake @UKYOUKAY]

@hirotaturou @POLITICS_202405 @YahooNewsTopics @politics 今回の議論は、感情的に意見をぶつけ合う場ではなく、論理的に意見を交換し合う場だと認識しています。もし、勝ち負けを決めることが目的であれば、この議論はあま���意味がないかもしれません。

[jake @UKYOUKAY]

@POLITICS_202405 @hirotaturou @YahooNewsTopics 単なる「タラレバ」ではなく、現実的な可能性や懸念に基づいていますよね

[jake @UKYOUKAY]

@POLITICS_202405 @hirotaturou @YahooNewsTopics 北朝鮮が外国で稼いでいる事実に基づいての可能性の話をしているのですよ，推測であることは自明です．

[jake @UKYOUKAY]

@POLITICS_202405 @hirotaturou @YahooNewsTopics 北朝鮮が外国で稼いでいるのは事実
https://t.co/9qnLuTvGCw
https://t.co/VnO6g6a8BI

[Jason Wei @_jasonwei]

One thing in AI research that I have finally recognized with clarity is the idea of “inertia bias”: continuing to do something when it’s not the best option.

The most basic instance of inertia bias is the feeling of “I already spent time implementing X, so let me continue trying to make it work, even though something better came up”. I find that this happens to me rather frequently—an idea that I spend a few days on is rendered obsolete for some reason (e.g., stronger experimental setting, ingenuity from a collaborator, etc). In these cases I usually just document what I did, reflect on how I could have done better, and move on.

In the medium term, at OpenAI I once spent three months on a project, only to find that another team had implemented a surprisingly better solution. It didn’t feel great to abandon my work and not deploy anything. But I recognized that I needed to not waste more time, so I decided to start something new (and luckily my project had other positive outcomes, e.g., I learned a lot).

When I first started in AI, I was clouded by inertia bias for almost a year. In 2019, I wrote a successful paper in text data augmentation. At the time I had an academic mindset of creating a cohesive body of work, and so I tried to milk my success. I wrote four more data augmentation papers, all of which were accepted to top conferences. But the impact turned out to be nothing but a few dozen citations to pad my CV.

The defining bet of my career so far was on the power of scaling language models, which I made in 2021. Obviously this is the current paradigm, and large language models have been hugely impactful. But this won’t be the last paradigm in AI. So when the time comes, I hope I will have the wisdom to fight this inertia and pivot to the next big thing.

(Knowing when to pivot after finding success is tricky, and I really admire those that successfully do it. A few years back I had the pleasure of visiting Luke Zettlemoyer’s lab at the University of Washington. Luke had done seminal work in the early 2010s on semantic parsing, and again in the late 2010s on language pre-training. When I asked his students what made him so successful, the most common answer I heard was that he doesn’t try to milk his successes, but rather he adapts quickly to the front of the field and tries to innovate.)

[Jason Wei @_jasonwei]

As a kid I loved whiteboard lectures way more than slides, so for Stanford’s CS25 class I gave a whiteboard lecture!

My goal was to simply and clearly explain why language models work so well, purely via intuitions.

Youtube video: https://t.co/rgeh6TtfrN (w/ @hwchung27)

[jake @UKYOUKAY]

@cat_maroon Excuse me if I'm being rude, but what was the cause of death?

[jake @UKYOUKAY]

@VRIyfmn2dCDxKuV ビショップ本が片��しかない？

[jake @UKYOUKAY]

@bioshok3 使ってみました～！

[jake @UKYOUKAY]

@umiyuki_ai かなり難しいとは思うけど，裸体検出技術を作成するぐらいはない気がするな，防犯カメラに関しては顔以外はモザイク処理されるようにするとか法執行機関のみが解除できるようにするとか，あとはデータ処理をローカルLLMで行い(一人ひとりのLLMとして対応)流出を防ぐとかかな

[jake @UKYOUKAY]

@rockfish31 由来https://t.co/1f4zb3NKTC

[jake @UKYOUKAY]

@ctgptlb ソースもお願いします