Incredibly cringe combination of flags.
Something only a fellow white person living in Japan could achieve. Well done.
lowink retweeted
"Anthropic told investors on Wednesday distillation accounted for 5 per cent to 7 per cent of revenue"
If true, and Q2 revenue was $11.5B, then that's more than half a billion dollars spent by other labs to distill Anthropic's models!
ft.com/content/11632d03-358c…
ALT The start-up has also clamped down on "distillation" - a controversial practice whereby rival Al labs train their models on the outputs of more advanced ones. When rival companies "distil" their models from Anthropic's models, that involves heavy usage and spending. Anthropic told investors on Wednesday distillation accounted for 5 per cent to 7 per cent of revenue and that they had cut it back "materially" in the third quarter of the year, said four people with direct knowledge of the matter.
Actually it's because Islam is a cancer that has rotted your culture.
Just look at what has become of Iran since the Islamic Revolution.
lowink retweeted
これ事実上のリストラだよね。入社してから営業や簡単な事務作業しかしてない明らかにITリテラシー低い人を突然AI開発者に転換することで、殆どの人は適性がないからアウトプットが出ず人事評価で気持ちよく低評価付けれるから、本人メンタル病んで自己都合退職して人件費削減。現代の合法リストラだ。
みずほが数千人をAI開発者に転換 事務や営業、日本企業の働き方転機
nikkei.com/article/DGXZQOUB0…
みずほは傘下の銀行や証券会社の社員5万人を対象に11月からAIの開発担当への転換を始めます。
まず数十人を営業や事務職から配置換えします。
lowink retweeted
Hey Japanese learners! I built a font that renders Furigana or Romaji above Kanji, with ligatures to support words. You can apply it anywhere to get furigana notations for free.
Give it a try, and I hope it will be helpful :)
furifont.antfu.me/
lowink retweeted
Anthropic 初の日本向け Claude Code ハッカソンが発表された！入賞賞金はなんと総額 $50,000(約 800 万円)相当の Claude API クレジット。しかも参加者全員に Max 5x プランと $500 分のクレジットが配られる。これは太っ腹では。。日本在住ならエンジニアでなくても応募できて、完全オンライン、開発期間は 11/2〜11/10 の約 1 週間。応募締切は 10/26 とのこと。
自分は日本国内に住んでいないので応募資格はないかもしれませんが、agmsg を題材に使ってこれに応募する方を全力でサポートしたいと思っています！
それで、これに合わせて agmsg のプラグイン機能を出そうと考えています。GitHub に「自分のアカウント名/なんとか」というプラグインのリポジトリを作って、それをコマンドで指定すると誰でも取り込める、つまり本体にマージされなくても使いたい人が使える、よくあるあれです。これで作ったものは agmsg にいろいろ取り込める、というキャンペーンをやりたいなと思っています。
我こそはと思う方、agmsg 関連で応募したい方、ぜひ声をかけてください。一緒に盛り上げましょう！
#agmsg #claudejapan2026
The Claude Code hackathon is back, and we're bringing it to Japan!
エンジニアの方も、そうでない方も、日本在住のビルダーならどなたでも応募できます。世界に何を見せてくれるのか、楽しみにしています。
応募締切は10月26日 23:59（日本時間）：
cerebralvalley.ai/e/built-wi…
#claudejapan2026
It's amazing how much good will a simple "thank you" can earn you in the work place.
It's more amazing how few managers ever learn this.
On the weekend I had a delivery driver say to me that he could not deposit the parcel in the delivery boxes outside of my apartment building because he did not understand how to set it.
Neither did I, but in broken Japanese we worked it out together, and both learned something.
The technology itself is fine (it's a classifier for model requests) but the larger issue is in how Sakana presents their work.
Fugu was presented as if it outperformed, but if you obscure the hidden cost, you can always present yourself as the winner.
China has never been much to care about the legality of copying others. I don't know why Japan ought to restrict itself so hard given that this is something Japan could, in theory, excel at.
Of course, to the original argument, Sakana AI's 'model' is not a model, but..
Even if you're co-writing a paper with AI, you should absolutely fix these kinds of figures so they're presentable.
we show that iterative self-refinement with LLMs, no matter the prompt, usually makes things worse. the model hallucinates flaws to satisfy critique prompts, each pass expands scope unchecked, and models almost never decline to make changes even when they should
Autoreason fixes this by structuring each iteration as a three-way tournament: the unchanged original (A), an adversarial revision (B), and a synthesis of the two (AB), judged by a blind panel of fresh agents via Borda count. if A (ie "do nothing") survives two consecutive rounds, the loop stops
