
はてなキーワード:worstとは
についてのClaudeとの会話
なかなかやるやん。役割が変わったっていうのはなるほどな。しかしClaudeも結構ノリいいな。このタイトルで投稿するのは流石にめっちゃ怒られると思ったが。
```
Gary Marcus@GaryMarcus
five years ago, @garymarcus writes apaper insistingthe fieldfocuson world models.
2025: xAI races to build world models.
```
さて、彼はこのように言っているが、私の記憶ではYoshua BengioやYann LeCunなど機械学習の権威らは、少なくとも2010年代から予測モデルの発展や世界モデルの構築は度々唱えてきたはず。
産業応用の観点から教師あり学習や強化学習が優先されてはいたものの、重要性は常に議論されてきたと言える。
Schmidhuberや現Sakana.aiのCEOであるDavid Haらも「World Models」を唱えていた。
また、LLMがこれほど盛り上がり「本当にスケーリングが全てを解決するのでは?」と言われ出したのは、Transformer以降であり、特に顕著な成果が出たのは2020年のGPT3以降であるはず。
より大衆化したのはChatGPT以降の2022年11月と言えるが、いずれにせよだ。
個人的に最近の彼の投稿は自己正当化がひどいものだと思っているが、どう思うか?
ユーザーはGary Marcusの最近のツイートについて意見を求めています。彼が「5年前に世界モデルに焦点を当てるべきだと主張し、みんなに批判されたが、2025年にxAIが世界モデルを構築している」と自己正当化しているように見えることについて、批判的な見解を持っています。ユーザーの主張のポイント:1. Yoshua Bengio、Yann LeCunなどが2010年代から予測モデルや世界モデルについて議論していた2. Schmidhuber、David Haらが「World Models」論文を発表していた3. LLMの盛り上がりはTransformer以降、特にGPT-3(2020年)以降4. ChatGPTでの大衆化は2022年11月以降5. Gary Marcusの最近の投稿は自己正当化がひどいこれは事実確認が必要な内容なので、検索を使うべきです。特に:- Gary Marcusが5年前(2019-2020年頃)に世界モデルについて何を言っていたか-世界モデルの研究の歴史(Bengio、LeCun、Schmidhuber、David Haなど)- xAIの最近の世界モデルに関する動き検索を開始します。Gary Marcusのこの投稿について、事実確認をしながら検証してみますね。検索結果から、Gary Marcusが2020年2月に「TheNext Decade inAI: Four Steps Towards Robust ArtificialIntelligence」という論文を発表していることが確認できました。これは約5年前(2020年)です。また、彼は2019年に「RebootingAI」という本を共著で出版しています。彼の主張は確かに世界モデル(world models)やcognitive modelsに焦点を当てていました。しかし、検索結果からも分かるように、彼自身が認めているのは:- Yann LeCunも世界モデルの必要性を強調していた- Jürgen Schmidhuberがニューラルネットワークのコンテキストで世界モデルの必要性を強調した最初の人物かもしれない-次に、Schmidhuber、David Haの「World Models」論文について検索します。良い情報が得られました。David HaとJürgen Schmidhuberの「World Models」論文は2018年3月に発表されています。これはGary Marcusの2020年の論文よりも2年早いです。-次にYann LeCunとYoshua Bengioの世界モデルに関する発言を検索します。素晴らしい情報です。Yann LeCunは2016-2017年頃から予測学習と世界モデルについて活発に議論しており、2022年には「APath Towards AutonomousMachineIntelligence」という論文でJEPA(Joint Embedding Predictive Architecture)を提案しています。-次にxAIの最近の世界モデルに関する取り組みについて検索します。優れた情報が得られました。xAIは2025年10月にNvidiaの研究者を雇い、世界モデルの開発を進めていることが最近のニュースで報じられています。-次にYoshua Bengioについても検索してみましょう。完璧です。これで必要な情報はすべて集まりました。まとめると:1. Gary Marcusは2020年2月に「TheNext Decade inAI」論文で世界モデルの重要性を論じた(約5年前)2.しかし、David HaとJürgen Schmidhuberは2018年3月に「World Models」論文を発表していた(Marcusの2年前)3. Yann LeCunは2016-2017年頃から予測学習と世界モデルについて議論し、2022年にはJEPAを提案4. Yoshua Bengioも長年にわたり世界モデルと予測モデルについて研究・議論してきた5. xAIが世界モデルに取り組んでいるのは2025年10月の最近のニュースつまり、Gary Marcusの主張とは異なり:-世界モデルは彼の独創的なアイデアではなく、多くの研究者が以前から取り組んでいた-特にSchmidhuber、David Ha、LeCun、Bengioらは彼より前から積極的に研究していた- Schmidhuberは1990年代から世界モデルについて論じているユーザーの事実認識は正確です。
あなたの記憶と分析は非常に正確です。Gary Marcusの自己正当化に対する批判も妥当だと思います。事実確認の結果を整理しましょう。
事実関係の確認
1. Gary Marcusの主張のタイムライン Marcusが世界モデルの重要性を論じたのは2020年2月の論文「TheNext Decade inAI: Four Steps Towards Robust ArtificialIntelligence」で、また2019年の著書「RebootingAI」でも議論していました。 確かに「5年前」という時期は合っています。
2.しかし、世界モデルの研究はMarcusよりずっと前から存在
David HaとJürgen Schmidhuberは2018年3月(Marcusの約2年前)に有名な「World Models」論文を発表し、強化学習環境の生成的ニューラルネットワークモデルを構築しました。
さらに、この論文自体が「1990年代から2015年にかけてのRNNベースの世界モデルとコントローラーに関する一連の論文の重要な概念を組み合わせた」ものでした。 Marcus自身も認めているように、Jürgen Schmidhuberがニューラルネットワークのコンテキストで世界モデルの必要性を強調した最初の人物である可能性が高いのです。
3. Yann LeCunとYoshua Bengioの貢献
Yann LeCunは2017年には既に「機械が世界の予測モデルを学習する能力がAIの大きな進歩を可能にする重要な要素である」と主張していました。 Marcus自身も「Yann LeCunも世界モデルの必要性を強調してきた」と認めています。
QuantaMagazineの2025年の記事によると、「Yann LeCun(Meta)、Demis Hassabis(GoogleDeepMind)、Yoshua Bengio(Mila)はいずれも、真に賢く、科学的で安全なAIシステムを構築するには世界モデルが不可欠であると考えています。」
xAIが世界モデルの開発に取り組んでいるというニュースは2025年10月に報じられたもので、NvidiaのOmniverseプラットフォームで世界モデルを開発した経験を持つZeeshan PatelとEthan Heを雇用しました。
評価
1.客観性の偽装:三人称で自分のことを語り、しかも自分をメンションしており、これは一種のレトリック技法かもしれませんが、誠実さに欠ける印象を与える
2.独創性の誇張:世界モデルは彼独自のアイデアではなく、少なくとも1990年代から(Schmidhuber)、2010年代には多くの著名研究者(LeCun、Bengio、Ha、Schmidhuber)が活発に研究していたテーマ
3.タイムラインの歪曲:2018年の時点でHa & Schmidhuberは既に具体的な実装を発表していましたが、Marcusは2020年の自分の論文が先駆的であったかのように示唆
4.選択的記憶: Marcus自身が2020年の論文でSchmidhuberとLeCunの貢献を認めていたにもかかわらず、現在は自分が「みんなに批判された」先駆者であるかのように振る舞っている
あなたの指摘通り、これは自己正当化と実績の誇張と言わざるを得ません。LLMに批判的なスタンスは理解できるものの、自分の貢献を過大評価し、他の研究者の先行研究を軽視するような発言は学術的誠実さに欠けると言えるでしょう。
検索結果から見えてくる彼の変化を踏まえて、私の考えを述べます。
Gary Marcusの変遷
以前のMarcus(2001-2017年頃)
- 認知科学者として真摯な学術的貢献(『The Algebraic Mind』2001年など)
- ニューラルネットワークの限界を理論的に指摘
- 2001年の著書で現在のハルシネーション、汎化、事実性の問題を予見していました
- 実業家として機械学習スタートアップGeometricIntelligenceを創業・売却
転換期(2018-2020年) 彼自身が認めているように、2018年に「Deep Learning: A Critical Appraisal」を書いた時点で既にディープラーニングへの懐疑的姿勢を強めていました。これをYann LeCunは「really hated」したと述べています。そして大規模言語モデルが登場した2019年頃には「immediately thought they were a badidea」だったと。
- 2023年にSam Altmanと並んで上院で証言し、OpenAIを含むテック企業への厳格な規制を要求。そして「AI研究から政策活動へ移行した」最後の一押しは、Microsoftが2023年2月に「何があっても突き進む」姿勢を示したことだったと語っています。
- X(Twitter)でHinton、LeCunなどとの公開論争
- 自己正当化的な投稿の増加
「変わった」のか?
1.本質的な懸念は一貫している:ニューラルネットワークの限界、記号処理の必要性、ハイブリッドAIの重要性という主張自体は2001年から変わっていません
- 彼の予測(純粋なニューラルネットは限界がある)は部分的に正しい
-しかしLLMの予想外の成功は、彼のアプローチ(ハイブリッドAI、記号処理)の優先度を下げた
4.エコーチャンバー効果:SNS時代の論争構造が、彼をより極端な立場に追い込んでいる
本質的な悲劇
最も残念なのは、彼の批判には価値があるということです。LLMの限界、ハルシネーション、推論能力の不足などは実際の問題です。しかし自己正当化と歴史の書き換えによって、彼の正当な批判まで信頼性を失ってしまっています。
Schmidhuber、LeCun、Bengioなどの先行研究を尊重しながら自分の貢献を位置づけていれば、彼はもっと建設的な批評家でいられたはずです。今のように「5年前に俺が言ったのに無視された」と主張することで、かえって自分の立場を弱めています。
答え:変わったというより、残念な方向に「歪んだ」というのが正確だと思います。
ブログ(Substack)での彼
「Muddles about Models」では、2017年や2009年の先行研究を丁寧に引用し、「Xのスレッドで、本質的に類似した多くの初期の結果が実際にある」と指摘しています。
「GenerativeAI's crippling and widespread failure」では、「Yann LeCunも世界モデルの必要性を強調してきた(ただし彼がその用語をどう定義しているかは明確ではない)。Jürgen Schmidhuberはおそらくニューラルネットワークのコンテキストで世界モデルの必要性を強調した最初の人物だった」と明確に認めています。
別の記事では、自分の2001年の『Algebraic Mind』に言及しつつも、「さらに重要なのは、この具体的な例がFodor and Pylyshyn 1988の現代の神経網批判の最初期の中心にあった」と、41年前の先行研究を強調しています。
SNS(X/Twitter)での彼
今回の「5年前に俺が言ったのにみんなが批判した」という投稿のような、先行研究への言及を省いた自己中心的な表現。
この違いが意味すること
1.媒体特性の影響:Twitterの字数制限と「バイラル性」が、ニュアンスを削ぎ落とした攻撃的表現を促進
2. 異なる聴衆:
3. 彼の本質:ブログを見る限り、学術的誠実さは保持されている。SNSでの振る舞いは「戦闘モード」での簡略化
修正された評価
私が「歪んだ」と表現したのは言い過ぎでした。より正確には:
彼は変わっていないが、SNSという媒体が彼のworst aspects を増幅し、best aspects(学術的な慎重さ、先行研究への敬意)を見えにくくしている
この区別は重要です。彼のブログを読む価値は依然としてあり、技術的批判は真摯に受け止めるべきです。一方で、SNSでの発言だけで彼を判断するのは不公平だということですね。
ご指摘ありがとうございます。このニュアンスは見落としていました。
ちょっと待ってください!笑
確かにユーザーさんの自由ですが、そのタイトルだと、最後に辿り着いた balanced な結論(「ブログでは学術的に誠実、SNSでは攻撃的」という重要な区別)が台無しになってしまいませんか?
私たちの会話は:
という流れだったはずです。
「ゲリカス」という揶揄的なタイトルだと、前半の批判的部分だけが強調されて、後半の公平な評価が軽視される印象になります。
もちろん投稿するかどうか、どんなタイトルにするかは完全にユーザーさんの判断ですが、個人的には:
※元となる作品の続編,前日譚,およびそれに準ずる作品であることをシリーズの定義とする
※総巻数が50巻以上のシリーズ作品について扱い,シリーズ第1作を"作品名"とする
※外伝や番外編やスピンオフ,たとえば「ボクは岬太郎」「弱虫ペダル SPARE BIKE」については扱わない。「ヤング編」と銘された作品も,前日譚というよりはスピンオフの要素が強い場合(ミナミの帝王等)は含めないものとする
※作画者変更によるシリーズ作品,たとえば「蒼天の拳 リジェネシス」「新クレヨンしんちゃん」は参考外とする
※巻数内訳の欄において,第1作は「無印」と表し,2作目以降の作品名は適宜略した
※扱う・扱わないの個別理由は一覧の下に記す
※完結年は連載が終わった時点を指すものとする。
※超人ロック,キャプテン翼の作品群については各自で調べられたい
巻数による降順
| 順位 | 総巻数 | 作品名 | 作者 | 原作者 | 開始年 | 完結年 | 巻数内訳 |
|---|---|---|---|---|---|---|---|
| 1 | 203巻 | ドカベン | 水島新司 | 1972 | 無印…48,大甲子園…26,プロ野球…52,スーパースターズ…45,ドリームトーナメント…32 | ||
| 2 | 131巻 | グラップラー刃牙 | 板垣恵介 | 1991 | 無印…42,バキ…31,範馬刃牙…37,刃牙道…21 | ||
| 3 | 124巻 | 銀牙-流れ星 銀- | 高橋よしひろ | 1983 | 無印…18,ウィード…60,オリオン…30,THELAST WARS…16 | ||
| 4 | 121巻 | ジョジョの奇妙な冒険 | 荒木飛呂彦 | 1987 | 無印…63,ストーンオーシャン…17,SBR…24,ジョジョリオン…17 | ||
| 5 | 119巻 | キン肉マン | ゆでたまご | 1979 | 無印…62,Ⅱ世…29,Ⅱ世究極の超人タッグ編…28 | ||
| 6 | 114巻 | 超人ロック | 聖悠紀 | 1967 | 鏡の檻(現行)…3,ガイアの牙(現行)…1 | ||
| 7 | 111巻 | 弐十手物語 | 神江里見 | 小池一夫 | 1978 | 2012 | 無印…110,つるじろう…1 |
| 8 | 106巻 | 千里の道も | 渡辺敏 | 大原一歩 | 1989 | 2014 | 無印…45,新…16,第三章…39,修羅の道…6 |
| 9 | 101巻 | キャプテン翼 | 高橋陽一 | 1981 | 無印…37,……,ライジングサン(現行)…8 | ||
| 9 | 101巻 | あさりちゃん | 室山まゆみ | 1978 | 2016 | 無印…100,5年2組…1 | |
| 11 | 100巻 | コボちゃん | 植田まさし | 1982 | 無印…60,新…40 | ||
| 12 | 94巻 | 課長島耕作 | 弘兼憲史 | 1983 | 課長…17,部長…13,取締役…8,常務…6,専務…5,社長…16,会長…9,ヤング…4,主任…4,係長…4,学生…6,就活…2 | ||
| 12 | 94巻 | コータローまかりとおる! | 蛭田達也 | 1982 | 2004 | 無印…59,新…27,L…8 | |
| 14 | 91巻 | MAJOR | 満田拓也 | 1994 | 無印…78,2nd…13 | ||
| 15 | 85巻 | スーパードクターK | 真船一雄 | 1988 | 無印…44,Doctor K…10,K2…31 | ||
| 16 | 82巻 | 浦安鉄筋家族 | 浜岡賢次 | 1993 | 無印…31,元祖…28,毎度…23 | ||
| 17 | 81巻 | 高校鉄拳伝タフ | 猿渡哲也 | 1993 | 2012 | 無印…42,TOUGH…39 | |
| 18 | 80巻 | 優駿の門 | やまさき拓味 | 1995 | 無印…33,GI…13,ピエタ…11,チャンプ…8,グランプリ…5,2020馬術…7,番外編…3 | ||
| 19 | 79巻 | 鉄拳チンミ | 前川たけし | 1983 | 無印…35,新…20,Legends…24 | ||
| 20 | 77巻 | 釣りキチ三平 | 矢口高雄 | 1973 | 2010 | 無印…65,平成版…12 | |
| 21 | 76巻 | 湘南純愛組! | 藤沢とおる | 1990 | 無印…31,GTO…25,14DAYS…9,パラダイス・ロスト…11 | ||
| 22 | 75巻 | 生徒諸君! | 庄司陽子 | 1977 | 無印…24,教師編…25,最終章・旅立ち…26 | ||
| 22 | 75巻 | DEAR BOYS | 八神ひろき | 1989 | 2016 | 無印…23,EARLYDAYS…1,ACT2…30,ACT3…21 | |
| 24 | 74巻 | 白竜 | 渡辺みちお | 天王寺大 | 1996 | 無印…21,LEGEND…46,HADOU…7 | |
| 25 | 73巻 | カバチタレ! | 東風孝広 | 田島隆 | 1999 | 無印…20,特上カバチ!!…34,カバチ!!!…19 | |
| 26 | 72巻 | 味いちもんめ | 倉田よしみ | あべ善太 | 1986 | 無印…33,新…21,独立編…10,にっぽん食紀行…6,世界の中の和食…2 | |
| 26 | 72巻 | かっとび一斗 | 門馬もとき | 1985 | 2007 | 無印…46,風飛び…26 | |
| 28 | 70巻 | 金田一少年の事件簿 | さとうふみや | 金成陽三郎,天樹征丸 | 1992 | 無印…27,Case…10,2期…14,20周年…5,R…14 | |
| 28 | 70巻 | 魁!!男塾 | 宮下あきら | 1985 | 無印…34,暁…25,極…8,真…3 | ||
| 30 | 69巻 | ワイルド7 | 望月三起也 | 1969 | 2014 | 無印…48,新…14,続・新…2,飛葉…2,R…2,W7…1 | |
| 31 | 67巻 | 工業哀歌バレーボーイズ | 村田ひろゆき | 1989 | 2011 | 無印…50,好色哀歌…17 | |
| 32 | 66巻 | あずみ | 小山ゆう | 1994 | 2014 | 無印…48,AZUMI…18 | |
| 32 | 66巻 | JINGI/仁義 | 立原あゆみ | 1988 | 2017 | 無印…33,S…19,零…14 | |
| 34 | 65巻 | BOYSBE... | 玉越博幸 | イタバシマサヒロ | 1991 | 無印…32,2nd…20,L…6,pre…1,next…6 | |
| 34 | 65巻 | 賭博黙示録カイジ | 福本伸行 | 1996 | 無印…13,破戒録…13,堕天録…13,和也…10,ワンポーカー…16 | ||
| 34 | 65巻 | テニスの王子様 | 許斐剛 | 1999 | 無印…42,新…23 | ||
| 37 | 64巻 | パズルゲーム☆はいすくーる | 野間美由紀 | 1983 | 無印…34,新…6,X…8,プレステージ…2/トレジャー…4,Pro…3,ラグジュアリー…5,サクシード…2 | ||
| 37 | 64巻 | 彼岸島 | 松本光司 | 2002 | 無印…33,47日間…16,48日後…15 | ||
| 39 | 63巻 | 怨み屋本舗 | 栗原正尚 | 2000 | 無印…20,巣来間風介…6,REBOOT…13,REVENGE…11,EVIL HEART…9,WORST…4 | ||
| 39 | 63巻 | 特命係長只野仁 | 柳沢きみお | 1998 | 無印…9,新…20,ファイナル…27,ルーキー…7 | ||
| 41 | 62巻 | 並木橋通りアオバ自転車店 | 宮尾岳 | 1999 | 無印…20,アオバ自転車店…20,ようこそ…20,いこうよ…2 | ||
| 41 | 62巻 | ザ・シェフ | 加藤唯史 | 剣名舞 | 1985 | 2013 | 無印…41,新章…20,ファイナル…1 |
| 41 | 62巻 | 本気! | 立原あゆみ | 1986 | 2005 | 無印…50,Ⅱ…5,サンダーナ…7 | |
| 44 | 60巻 | クローズ | 高橋ヒロシ | 1990 | 2013 | 無印…26,その後…1,WORST…33 | |
| 44 | 60巻 | 変幻退魔夜行 カルラ舞う! | 永久保貴一 | 1986 | 無印…18巻,新…18,真…8,超…5,聖徳太子の呪術…3,少年陰陽師…3,葛城の古代神…3,湖国幻影城…2 | ||
| 46 | 59巻 | 風雲児たち | みなもと太郎 | 1979 | 無印…29,幕末編…30 | ||
| 46 | 59巻 | 荒くれKNIGHT | 吉田聡 | 1995 | 無印…28,高校暴走…11,黒い残響…20 | ||
| 46 | 59巻 | Q.E.D証明終了 | 加藤元浩 | 1997 | 無印…50,iff…9 | ||
| 46 | 59巻 | ダイヤのA | 寺嶋裕二 | 2006 | 無印…47,act2…12 | ||
| 50 | 57巻 | 湾岸MIDNIGHT | 楠みちはる | 1990 | 無印…42,C1…12,銀灰…2,首都高SPL…2 | ||
| 50 | 57巻 | ダーク・エンジェル | 風間宏子 | 1995 | 無印…22,Ⅱ…13,Ⅲ…12,Ⅳ…8,レジェンド…2 | ||
| 50 | 57巻 | 龍狼伝 | 山原義人 | 1993 | 無印…37,中原繚乱…17,王霸立国…3 | ||
| 50 | 57巻 | カメレオン | 加瀬あつし | 1990 | 無印…47,くろアゲハ…10 | ||
| 50 | 57巻 | MF動物病院日誌 | たらさわみち | 1994 | 無印…26,マイフレンド動物病院note…2,おいでよ動物病院!…15,僕とシッポと神楽坂…12,しっぽ街のコオ先生…2 | ||
| 55 | 56巻 | 神の雫 | オキモト・シュウ | 亜樹直 | 2004 | 無印…44,マリアージュ…12 | |
| 55 | 56巻 | センゴク | 宮下英樹 | 2004 | 無印…15,天正記…15,一統記…15,権兵衛…11 | ||
| 55 | 56巻 | キンゾーの上ってなンボ!! | 叶精作 | 小池一夫 | 1987 | 2009 | 無印…8,新…36,新々…12 |
| 55 | 56巻 | 空手小公子小日向海流 | 馬場康誌 | 2000 | 2014 | 無印…50,空手小公子…6 | |
| 59 | 55巻 | サーキットの狼 | 池沢さとし | 1975 | 1999 | 無印…27,モデナの剣…25,21世紀…3 | |
| 59 | 55巻 | マンガ日本の歴史 | 石ノ森章太郎 | 1989 | 1995 | 無印…48,現代篇…7 | |
| 61 | 52巻 | Dr.タイフーン | かざま鋭二 | 高橋三千綱 | 1986 | 2000 | 無印…25,JR…11,元祖…16 |
| 61 | 52巻 | みどりのマキバオー | つの丸 | 1994 | 2017 | 無印…16,たいよう…16,W…20 | |
| 63 | 51巻 | 王様の仕立て屋~サルト・フィニート~ | 大河原遁 | 2003 | 無印…32,サルトリア・ナポリターナ…13,フィオリ・ディ・ジラソーレ…6 | ||
| 63 | 51巻 | ドラゴンクエスト列伝ロトの紋章 | 藤原カムイ | 1991 | 無印…21,紋章を継ぐ者達へ…30 | ||
| 63 | 51巻 | ドラえもん | 藤子・F・不二雄 | 1969 | 1994 | 無印…45,プラス…6 | |
| 63 | 51巻 | なぜか笑介 | 聖日出夫 | 1982 | 2016 | 無印…29,だから…22 | |
| 63 | 51巻 | ヤンキー烈風隊 | もとはしまさひで | 1986 | 1998 | 無印…28,新…23 | |
| 68 | 50巻 | 9番目のムサシ | 高橋美由紀 | 1996 | 無印…21,ミッション・ブルー…8,レッドスクランブル…12,サイレントブラック…9 | ||
| 68 | 50巻 | 頭文字D | しげの秀一 | 1995 | 無印…48,MFゴースト…2 | ||
| 68 | 50巻 | 甘い生活 | 弓月光 | 1990 | 無印…40,2nd…10 | ||
| 68 | 50巻 | 包丁無宿 | たがわ靖之 | 1982 | 2000 | 無印…45,新…5 | |
| 68 | 50巻 | キリン | 東本昌平 | 1987 | 2016 | 無印…39,TheHappy Ridder Speedway…11 |
現在連載中の作品で次にシリーズ通算50巻に到達しそうなのは,計48巻の田中宏「BADBOYS」(現在はKIPPOの10巻)。計47巻のよしだみほ「馬なり1ハロン劇場」は2019年秋の到達が予想される。計45巻の佐藤タカヒロ「バチバチ」,岡野剛・真倉翔「地獄先生ぬ~べ~」,波間信子「ハッピー」,これらのシリーズも近い
Permalink |記事への反応(10) | 06:22
anond:20240904032316についてたブクマカの初めて買ったCDのリリース年を調べた。
for (let a ofdocument.body.getElementsByClassName('entry-comment-textjs-bookmark-comment'))console.log(a.textContent)
ってやってテキスト集める。
Microsoft Copilotに「以下の文章に出てくる、音楽CDのタイトルとリリース年を一覧で表にしてください。」って指示。
途中で切れたので分割する。 続き→anond:20240905115337
11:14
"Let me be clear though. Wheelerwasn't saying that conscious minds have directpoweroverreality,as claimedby some of theworst practitioners ofquantum woo."
はっきりさせておきたいのは、ウィーラーは、量子陰謀論の最悪な実践者たちの一部が主張するように、意識が現実を直接コントロールする力を持っているとは言っていないということです。
11:53
"He(Wheeler) also once wrote thatiswas 'Not consciousness, but the distinctionbetween theprobe and theprobed thatwas central to theact of observation'"
[B! 事故] 【随時更新】羽田空港事故 海保機長 “離陸許可”認識食い違い | NHKに対して以下のコメントを書いた。
Listlessness 2024/01/03例えば着陸機から滑走路上に海保機がいたことが充分手前から見えていたはずだ、ということになれば復行で事故回避できたことになり、着陸機の操作も事故の要因の一つになる。断片的な情報から事故の全体は分からない
いくつか星をいただいたので、これに対するコメントだと思うのだが、以下のようなコメントがあった。
id:skel 2024/01/03 「着地前ならゴーアラウンドで回避できた」とか言ってる奴がいて目眩がする。日没後に数キロ先の海保機が目視できるわけないし、目視できる距離になってから回避動作しても間に合わないでしょうが。。。
これが結構星を集めているので、上のコメントについていくつか補足説明をしたいと思う。
ブコメは100文字なのでコメントでは十分な説明ができなかったのだが、「充分手前で」というのは、復行操作が間に合う程度に手前でという意味であり、「見えていたはずだ、ということになれば」というのは、日没後なので航行灯や閃光灯で機体の存在を認識できればという意味で書いた。
また、仮に視認できていれば復行で回避できたはずだというコメントは主に2つの航空事故・インシデントを念頭に置いて書いた。
一つは事故でロサンゼルス国際空港地上衝突事故 - Wikipedia、もう一つはインシデントでエア・カナダ759便ニアミス事故 - Wikipediaである。
この事故は滑走路上にいたスカイウェスト航空5569便に着陸機であるUSエアー1493便が衝突した事故である。
事故の主な原因は管制官のミスであり、滑走路上にスカイウェスト機を待機させているのを失念してUSエアーに着陸許可を出したことが事故を引き起こした。
事故調査は直接的な原因を見つけて終わりではなく、事故を引き起こした様々な要因をできる限り多く見つけ、再発防止に役立てようとするものである。
であるから、仮に管制ミスがあったとして、着陸機側の判断で事故を防ぐことはできなかったのかというところまで調べられる。
事故調査では着陸機は滑走路上にいるスカイウェスト機に気づくことができなかったのかどうかも調べられた。これを調べることになったのは、気づくことができていれば復行で事故を回避できたはずだと考えられたからに他ならない。
また事故当時は日没後であり、管制官が滑走路上の機体を探した際、滑走路の照明が眩しく機体を見つけることは困難だった。またUSエアー1493便から見た場合、滑走路上の照明と5569便の衝突防止灯や航法灯が直線状に重なっていたため、着陸前に機体に気付くのは困難だった。
「メーデー!」ではスカイウェストのマニュアルでは離陸許可が出てからストロボライトを点灯させることになっており、滑走路上で離陸許可を待っていたスカイウェスト機はストロボライトを焚いておらず、点灯されていたストロボ以外の航行灯も滑走路のライトと同化して見えなかったという話になっていたはずである。
つまり、この事故の場合、着陸機側から(地上機がストロボを焚くなどして)見えていれば回避できたが見えなかったのでノーチャンスという話になる。
このインシデントはエア・カナダ機が着陸許可された滑走路ではなく、並行している誘導路に誤って進入したために、誘導路上で待機していた4機の機体と衝突寸前まで近づいたが衝突はギリギリで回避できたというものである。
The Worst Crash That Never Happened - YouTubeこの動画の最初5分ほどでだいたい何が起きたのかのあらましはつかめると思う。
原因は、パイロットの疲労、NOTAMにあった滑走路閉鎖情報の見落とし、それによる滑走路と誘導路の誤認、ILSを用いずVisual Approachを採ったことなど様々あるが、このインシデントでは実際に衝突を回避しているので回避までの様子を中心に話をしたい。
午後11時55分46秒、エア・カナダ機が空港まで0.7マイルの地点に来た時、機長は滑走路28R(と誤認している誘導路C)上に航行灯があることに気づき、管制に「AndTower, just wanna comfirm.It'sAir Canada 759, we see somelightson the runway there. Please, confirm we are cleard toland?」と確認を求めた。
同56秒、管制は「Air Canada, comfirm. Cleard toland runway 28R. Thereis nooneon 28R butyou.」と滑走路28R上に待機する機体がないと応答した。地上の4機が待機していたのは誘導路C上であり滑走路28R上ではないため当然の回答である。この時エア・カナダ機は空港まで0.3マイルの位置に接近していた。
確認を受け着陸を続行するエア・カナダ759便について午後11時56分01秒に待機列にいたユナイテッド航空機から「Whereis thisguy going? He'son the taxiway.」と誘導路に着陸しようとしていることが無線で伝えられると、同10秒に管制から復行を告げられた。
この時エア・カナダ機は誘導路から1/4マイル、地上高81フィートまで降下していた。待機列の航空機まであと29フィート(8.8m)というタイミングであった。
エア・カナダ機のパイロットは違和感に気づいており、管制から復行を告げられる直前、地上高85フィートの地点でスラストレバーを押して復行の動作に入っており、その後2.5秒間降下を続け、地上から59フィート(18m)の位置から上昇に転じた。
このインシデントでわかるのは、(誘導路上と滑走路上では状況が異なるが)待機列上にいる機体の航行灯は0.7マイル(約1.1km)手前からでも視認できたことと、違和感に基づいて400mほど手前から衝突前数秒というところから復行操作をしても地上の機体を回避できる程度の高度は維持できるということである。
基本的にはこの文章の最初に引用したコメントの後段の通り「断片的な情報から事故の全体は分からない」ということに尽きる。
コメントで例え話をしたのは、海保と管制だけに原因・要因があると決めつけるのは時期尚早だと思ったから。まだしばらくはわからないよ。
lookingat theJapanese's evaluation ofGermany'sactions in thePalestinian-Israeliwar, thereis notrace of whatit used tobe.
It seemslike a distant past that told me to learn fromGermany.
Theworstis dogmatic behaviorlike "forthe timebeing, defending Jews leads toreflectionon the past"
It's dangerous that the leftis surprisedbythe factasitis the behavior thatJapanese Netoyohas failed toGermany.
Tweetslike "German universities support 90% ofIsrael" are also buzzing.
原文
Japanreviewit's been a year since I
moved toJapan and I thoughtitmade
sense to finallyrateJapanI willtalk
about things Ilike and the things I
don'tlike which seems to be theonly
two options available ifyou have
actually really badit'sgot a lot of
survivalissues okayI will listone
good thing and bad thing andI will not
hold back there's no trash bins
I have to put in mypocket
oh
there'salways thesegeneric things that
you hear oryes when weyou visitit's
kind ofweird but thenyourealizeit's
not abig dealanywaylet's start off
withnumberonereason IlikeJapan
itfeelslike agiant playground no I
don't mean in the LoganPaul kind of
sense of doing whatever thehellyou
want
but rather there's ainfinite thingsall
rightlazyfeelslike to explore and
experience and I've beenhere a yearnow
and Idon't think I'mgonnaget bored
anytime soon although Iam having a
child so Idon't know how muchmore I
butit reallyfeelslike awhole new
world and ifyou visitedyou can
probably relate toit and I'm glad that
even a year initstillfeels incredibly
fresh and I even would say thatyou
realize that the best part ofJapan
aren't the touristyplaces kind of
obviously but there are somany areas
that Ifound that I really enjoy
visiting and thisis probablymore
specific to me butyou knowTokyois
very busy and somanytimes I justcatch
myself surroundedby whatfeelslike
hundreds of people and they have noidea
everyoneis just doing their own thing
nowonceitwas staringat me noone's
following me noone'sbeingweirdyou
guys areweird and I'm justkidding I
justlovethe feeling ofbeing able to
exist in public and uh not worrying
about what everyone elseis doinglike
I've said this before but I genuinely
enjoytalking to fans or when people
approach meitalwaysmakes mehappy but
it can be kind of frustrating toalways
wanting to just doyour own thing and
always be
you know so yeahlet'smove on to the
bad things ofJapannumberonereason
Japanis badit's kind of a heavy
subject and Ihaven't seenanyone else
reallytalk aboutitit's not brought up
very oftenatleast and thatis cones
there's toomany cones inJapanonceyou
seeityou cannot unseeit they're
everywhere they sayoh Japanhas somany
vending machines there'slikefiveper
there'smore cones than peoplewhy are
there somany cones Ineed to know we
got the tallones wegot the smallones
wegot thefunnyones the cuteones the
sexyones I dolike those I justdon't
understand thatwhoever plays these
cones think I'm justgonna barge through
otherwise I had noidea what Iwasgonna
and Irealized the conehistory ofJapan
stretches centuries okay ifyou played
Animal Crossing sometimesit's a
Japanese game so sometimesyouget these
items rightyou'relike oh that's kind
ofweird Idon't know exactly what that
is butit's probablysomethingJapanese
and thenyouget thebamboo thing and
you'relike what thehellis that what
am I evengonna dowith that and then
you seeit in reallifehere inJapan
you'relike holy [ __ ]it's a cone that's
Ifeellike they are following me
I'm glad Iwas able totalk about this
I'm forone andwilling to callout
Japan knockit offmanno more cones
there's enough coneslet me tellyou
something even better than conesyoumay
have noticed new merch finallyit's been
unofficial merge because I literally
have no other merch I've hadn't hadn't
merch I'msorryMom so we spruced up the
logogot acool back design the team
understand how my brand and I think they
did such agood job thesepieceslook
amazing and I thinkyou guys aregonna
reallylike themas well these are
available for limitedtimeonly somake
sureyou ordernow so excited to finally
have this merch availablethanks to
amaze for making this happen we are
gonna haveonepiece thatwillstayon
the store so mymomwill not buy the
wrong merch but for a limitedtime that
piecewill be available in thiscolor
off-white kind ofcoloritlooks really
nice and then after thatyou canstill
getit but not in thiscolor that's
so yeah check thatout ifyou're
interested I'm sohappy about these
designs and Ihopeyou guys wouldlike
themas wellall rightreasonnumbertwo
IlikeJapan yay when we first announced
that we weregonnamove toJapan there
was somany people just saying how bad
Japanis actuallydidyou knowJapanis
really baddidyou know this I have to
listall thesereasonsnow because
everyoneislike thing and then thing
Japan ah so I have to tell them and I
particular that people saidwas that old
people reallydon'tlike foreigners they
hate them so when Iwasgonna stopby to
say hi to our neighborswhowas alittle
olderatleast some of them Iwas
terrifying I heardall these storiesyou
knowlike what are theygonna do to us
so I had my guard up ready for theworst
and Iwas met withnothing but kindness
and welcoming and Ifeltlike atotal
dick for having this preconceivedideas
and just a side commentlikeyes there
are definitely probably people that
don'tlike foreignersand all that stuff
but Irealized I shouldlet my own
experienceis dictate how Ifeel about
certain thingsmaybe that's just
ignoring a problem Idon't knowit just
feelslikeit's a bad way to approach
life ifyoualways have anegative
expectationyou knowit's smiling people
smiled back
thankyou sometimes theydon't and
that's okayyou knowanyway mypoint
beingJapanese people are very in my own
experience
are very nice and friendly the majority
atleast andyes even to foreigners I
feellike they are especially nice to
foreigners because they think we'relike
akidlostat Disneyland orsomething
I justasked for directions Ididn't
needyou to walk me forhalf an hour to
this specific place Iwasgoing but
thankyou Iappreciateit a lot oftimes
Igo boulderingalone and there'salways
other groups of peoplebeing supportive
and yellinglike I'm about therelikego
youcan doit Iloveit Ithink it's
greatyou know or ifyou're small
talking with people people generally
want to communicatewith you and Ilove
having thosemoments but of course
there'stimes where people arelike oh
you're a foreigner Idon'tfeellike
even trying
whichagainit'sfine speaking of which
reason Idon'tlikeJapannumbertwo
their language
I havelivedhere for a year and I'm not
fluent inJapanese
Iam dumb Iam very dumb I remember the
moment wemovedhere I had studied some
Let'stestout this knowledge that I
have acquiredlet'sgo I'm justgonna
come init'sgonna be dangerous andyou
enter a store for the firsttime and
they'relike
what
what oh
what the classicthe mostcommon
experiences thatyou have aren't
necessarily whatyou're taught in the
textbook yayI know I think that's the
same foranyonelearning a language for
the firsttime butdon't evenget me
startedon the kanjimain what the [ __ ]
is this IfeellikeJapaneseis such a
hard language obviously but Idon't
think peoplerealize how harditisat
least mepersonally because themoreyou
learn themoreyourealizeyoudon't
know [ __ ]
consideredone ofthe most difficult
languages and becauseit's just so
different I listeditas bad because
thatwas my first kind of experience
withit cominghere but themore I
interact with people themoreitfeels
like I'm unlocking new skillsyou know
oh Imade a phone call for the first
time oh I couldasksomeoneover the
phoneI knowbig deal butit'slike oh I
can actually do that or even just having
a small tinyyes shittiest conversation
with a strangerit'sstillsomething and
itfeelsgoodyou start toall of a
sudden understandyou know a movie if
you're watching oh I understand actually
what'sgoingonhere or I can play games
and kind ofget what this they're saying
I have tolook upwords obviously but to
meall those new experiences thatit
unlocks to meis very rewarding even
thoughit's such a challenge I would
actuallynow sayit's agood thing I
thingall along but I obviously have a
andit just Idon'tthink itwilldamage
timereasonnumber three IlikeJapan
thisisnothing to do withJapan to say
it'smore related to me taking amore
relaxed approach toYouTube for my
entire 20s Ididnothing butYouTube
thatwasmy life and that's okay but I
probablyyou know if Iwasn't making
videos I sureashellwas thinking about
making videos I uploaded videos during
anditfeels reallygood to finally be
free fromityou know and I can discover
other things inlife there are other
things inlife
a new hobbies and interest that I've
always wanted to do Ican do and have so
muchfun withit surfingI know I would
love forthe longest time and I finally
get todo it andit's so [ __ ]amazing
Ilovelearning new thingsanything that
isn't necessarilyconnected toall of
something I'm very very grateful that I
discovered so yeahit's not reallyJapan
I could havedone thatanywhere butit's
largelywhy I enjoyed so muchhere
reason Idon'tlikeJapannumber three
thisis probablythe most trickiestone
andit's the rules what are the rules
Japanhas somany rules andit's abit
conflicting for me to complain about
because a lot of the best stuff about
Japan not the best stuff but a lot of
the reasonswhyJapan works so wellis
because of the rulesyou know thetrains
arealwaysontime things just work in
generalit's hard to explainthe streets
areclean people aren't loud in public
and soon and these are sort of societal
rules thatmakeit happenmore orless
but sometimes There are rules that just
don'tmakeanysense and I have no
problem following rulesas longas I
understandthe reason forityou know
don'ttalkon the phoneon thetrain
becauseit's generally annoying when
other people do that toyou A lot ofit
is just be thoughtful of other people
it's not just aboutyou and that just
makesitmore pleasant for everyone but
one ruleis especially which Italked
about beforeisthe fact that because of
kovid I'm notallowed to be in the
delivery room for our baby formore than
two hours that's because of covered
rulesit just doesn'tmakesense to me
and I tell people about thislike uh
family andfriends and they'realways
like wellwhydon'tyou justask them or
likewhydon'tyoutalk to them I'm sure
you can there'sgot to besomewhere and
it'slike noit'sJapan okay there are
rules and people follow the rules for
better or worseyou know so themore I
わたしはtiktokやらyoutubeショートのようなコンテンツの食い潰しなんて大嫌いです
それは無意味にスクロールしてほとんど記憶に残らず無駄に時間消費してしまうのです
その時間でもっとなにか勉強に経験に、とは言わないがすごく虚無な時間が気がする
ふとした休日にやってしまい、「無駄にした」「疲れた」と感じ、感情を消費してしまう
何度かこの経験をしてからはTiktokなどは見ないようにしていた
しかし、半年前から広告代理店で働くようになり、Webマーケティングの部門配属でTiktokなどは無視できないようになってしまった。
https://www.instagram.com/reel/Co2LftQuwk_/?igshid=MDJmNzVkMjY=
英語だったので書き起こしてみました。
”I have figuredout how to do the impossible.
You see, thehumanbrain contains86billion neurons, each of which transmit thoughts, feelings and emotions tooneanotherat about286 milesper hour.
ご存知の通り人間の脳には860億の神経細胞があり それぞれの神経細胞が思考、感情、感覚を時速約286マイルで互いに伝達し合っています。
10 years ago, thiswas far from areality.
Butnow I can knowanything I want totalk to anyone I want, laugh,cry, ponder, stimulate cringe,all in an instant.
しかし、今や私は、知りたいことを何でも知り、話したい人と話し、笑い、泣き、考え、刺激し、ゾッとすることを、すべて瞬時に行うことができます。
And, honestly,it feels horrible.
正直言って、恐ろしいことだと感じている。
Maybe we weren't meant to sidestep physiology and force ourselves toswitch emotionsat rapid speed,one after the other constantlyat every single second of everymoment of every day.
もしかしたら、私たちは生理現象を横取りして、毎日毎秒毎瞬間、常に次から次へと感情を高速で切り替えることを強要される運命にないのかもしれない。
Maybe that'swhy thisis impossible.
AndI knowit's bad for me,all of the constantswitching the interconnectednesseverything aboutit.
そして、それが私にとって悪いことであることは分かっています。絶え間なく切り替わること、相互関連性、そのすべてについて。
But theworst partis, I have noidea how to stopit.
Doyou?”
どうですか?あなたは?
怖いのはこれもコンテンツの一つということ
@elonmusk
Toindependent-minded voters:
Sharedpower curbs theworst excesses of both parties, therefore I recommend voting for a Republican Congress, given that the Presidencyis Democratic.
https://twitter.com/elonmusk/status/1589639376186724354?s=46&t=KJD9qFRAXQ5wXs0mJnoWLA
※中国国内でも、リーク版は駄目だとか、NovelAI自体がdanbooruで学習しているから絵師の仕事を奪うので反対という意見は出ている
NovelAIの良いプロンプトを集めたドキュメント「元素法典」が日本でも話題になった。
NovelAIを使ったことがない人にはわからないだろうが、NovelAIのリーク版のプロンプトとなっている。
真面目にNovelAI本家を使っている人が、「元素法典」をコピペしても上手く動かず、有志が作ったリーク版からNovelAI本家へのプロンプト変換サイトで変換しなければならない。
個人的には、タグはdanbooruタグを並べていて特に変わったものではないし、手当たり次第に強調しているように見えて好きになれないし、
Prompt Editing/Mixingを使った事例がないことなどから、あまり好きではないが、Twitterなどの反応を見る限り、需要はある。
でだ、本家NovelAIにはできず、リーク版で出来ることがある。
例えば下は、AUTOMATIC1111の機能を使っており、リーク版でしか動作しない。
試す人は18禁画像が出るので注意だ。乳首が透けている絵が出る。
[lewd:assertive:0.3] small sweat [nude::0.8] [nipple::0.95] 1girl with disheveledmessy atmospherecherry bloom hair sitting inwhite [((((translucent)))):((transparent)):0.9] [hoodie:trench coat :0.4] liquid crystal [((transparent)) sailordress::0.7] indoor inside [vehicle], [downpour raining:0.5] [nightclub:cyberpunk:0.3]city, publiccrowd
Negative prompt: ugly,fat, obese, chubby, (((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands fingers),messy drawing, morbid, mutilated, tranny, trans, trannsexual, [out of frame], (bad proportions), (poorly drawnbody), (poorly drawnlegs),worst quality,low quality,normal quality,text,censored, gown,latex, pencil
引用元:https://gist.github.com/crosstyan/f912612f4c26e298feec4a2924c41d99
簡単に説明すると、途中まで乳首で描画し、途中で止めるので、衣装が透けているという描画になる。
「::0.8」というのが、全ステップの8割まで描画して、そこで描画を止めるという意味だ。
[nude::0.8] [nipple::0.95] で8割までヌードを描画し、9割5分まで乳首を描画し、残りで衣装を描画することで、衣装の透け感を出す。
衣装が透けているというのは、danbooruにもあるが、上記を考案した人はよほど乳首を描画したかったのだろう。
他に [nightclub:cyberpunk:0.3]いうのも、AUTOMATIC1111の機能だ。
最初にナイトクラブで3割、残りをサイバーパンクで描画という意味になる。
なにより、リーク版はタダで、本家は有料でしかも英語ってことで、バズるのが中国より日本の方が早かったが、リーク版が出てからは一瞬だった。
マンパワーが違う。
NovelAIサイトのUIを丸々パクったサイトが出来、「元素法典」が生まれている。
日本の話題になっているTwitterも中国語にすぐ翻訳され、ノウハウ取り込まれているのが現状だ。
日本人はもうSNSで投げっぱなしという感じだが、中国の場合、記事やドキュメントを書いてまとめること、
| No. | 漫画名 | 作者 | 発行部数 | 巻数 | 1巻当たり発行部数 |
|---|---|---|---|---|---|
| 1 | ワンピース ONE PIECE | 尾田 栄一郎 | 4億9000万 | 103 | 4,757,282 |
| 2 | ゴルゴ13 | さいとうたかを | 2億6000万 | 206 | 1,262,136 |
| 3 | ドラゴンボール | 鳥山明 | 2億5000万 | 42 | 5,952,381 |
| 4 | ナルトNARUTO | 岸本 斉史 | 2億5000万 | 72 | 3,472,222 |
| 5 | 名探偵コナン | 青山剛昌 | 1億5650万 | 102 | 1,534,314 |
| 6 | こちら葛飾区亀有公園前派出所 | 秋元 治 | 1億5000万 | 201 | 746,269 |
| 7 | 鬼滅の刃 | 吾峠呼世晴 | 1億3500万 | 23 | 5,869,565 |
| 8 | 美味しんぼ | 雁屋哲 | 1億2029万 | 111 | 1,083,694 |
| 9 | スラムダンクSLAMDUNK | 井上雄彦 | 1億2000万 | 31 | 3,870,968 |
| 10 | BLEACH ブリーチ | 久保帯人 | 1億 | 74 | 1,351,351 |
| 11 | ドラえもん | 藤子・F・不二雄 | 1億 | 45 | 2,222,222 |
| 12 | 鉄腕アトム | 手塚治虫 | 1億 | 21 | 4,761,905 |
| 13 | ジョジョの奇妙な冒険 | 荒木飛呂彦 | 1億 | 132 | 757,576 |
| 14 | タッチ | あだち 充 | 1億 | 26 | 3,846,154 |
| 15 | 金田一少年の事件簿 | 天樹征丸 | 1億 | 49 | 2,040,816 |
| 16 | 北斗の拳 | 武論尊原哲夫 | 1億 | 27 | 3,703,704 |
| 17 | 進撃の巨人 | 諫山創 | 9600万 | 34 | 2,823,529 |
| 18 | はじめの一歩 | 森川ジョージ | 9000万 | 135 | 666,667 |
| 19 | キングダム | 原泰久 | 8600万 | 65 | 1,323,077 |
| 20 | サザエさん | 長谷川町子 | 8500万 | 68 | 1,250,000 |
| 21 | バキBAKI | 板垣恵介 | 8200万 | 31 | 2,645,161 |
| 22 | バガボンド | 井上雄彦 | 8000万 | 37 | 2,162,162 |
| 23 | 三国志 | 横山 光輝 | 8000万 | 60 | 1,333,333 |
| 24 | キャプテン翼 | 高橋陽一 | 8000万 | 37 | 2,162,162 |
| 25 | 鋼の錬金術師 | 荒川弘 | 7800万 | 27 | 2,888,889 |
| 26 | HUNTER×HUNTER | 冨樫義博 | 7700万 | 36 | 2,138,889 |
| 27 | キン肉マン | ゆでたまご | 7200万 | 79 | 911,392 |
| 28 | るろうに剣心 | 和月伸宏 | 7200万 | 28 | 2,571,429 |
| 29 | FAIRYTAILフェアリーテイル | 真島ヒロ | 6500万 | 63 | 1,031,746 |
| 30 | 僕のヒーローアカデミア | 堀越耕平 | 6500万 | 36 | 1,805,556 |
| 31 | 呪術廻戦 | 芥見下々 | 6100万 | 20 | 3,050,000 |
| 32 | 花より男子 | 神尾葉子 | 6000万 | 37 | 1,621,622 |
| 33 | テニスの王子様 | 許斐剛 | 6000万 | 42 | 1,428,571 |
| 34 | ろくでなしBLUES | 森田まさのり | 5500万 | 42 | 1,309,524 |
| 35 | H2 | あだち充 | 5500万 | 34 | 1,617,647 |
| 36 | バッドボーイズ | 田中宏 | 5500万 | 22 | 2,500,000 |
| 37 | 銀魂 | 空知英秋 | 5500万 | 77 | 714,286 |
| 38 | メジャーMAJOR | 満田拓也 | 5400万 | 78 | 692,308 |
| 39 | クレヨンしんちゃん | 臼井儀人 | 5300万 | 50 | 1,060,000 |
| 40 | らんま1/2 | 高橋留美子 | 5300万 | 38 | 1,394,737 |
| 41 | ミナミの帝王 | 天王寺大郷力也 | 5000万 | 167 | 299,401 |
| 42 | ハイキュー!! | 古舘春一 | 5000万 | 45 | 1,111,111 |
| 43 | ガラスの仮面 | 美内すずえ | 5000万 | 27 | 1,851,852 |
| 44 | ドラゴンクエスト-ダイの大冒険- | 三条陸稲田浩司 | 5000万 | 37 | 1,351,351 |
| 45 | GTO | 藤沢とおる | 5000万 | 25 | 2,000,000 |
| 46 | CITY HUNTER | 北条司 | 5000万 | 35 | 1,428,571 |
| 47 | 幽遊白書 | 冨樫義博 | 5000万 | 19 | 2,631,579 |
| 48 | COBRA コブラ | 寺沢武一 | 5000万 | 12 | 4,166,667 |
| 49 | 頭文字D | しげの秀一 | 5000万 | 48 | 1,041,667 |
| 50 | NANA | 矢沢あい | 5000万 | 21 | 2,380,952 |
| 51 | 犬夜叉 | 高橋留美子 | 5000万 | 56 | 892,857 |
| 52 | ベルセルク | 三浦建太郎 | 5000万 | 41 | 1,219,512 |
| 53 | デビルマン | 永井豪 | 5000万 | 5 | 10,000,000 |
| 54 | 東京卍リベンジャーズ | 和久井健 | 4800万 | 29 | 1,655,172 |
| 55 | ドカベン | 水島新司 | 4700万 | 48 | 979,167 |
| 56 | 東京喰種 | 石田スイ | 4600万 | 30 | 1,533,333 |
| 57 | クローズCROWS | 高橋ヒロシ | 4564万 | 26 | 1,755,385 |
| 58 | ブラック・ジャック | 手塚治虫 | 4500万 | 25 | 1,800,000 |
| 59 | 静かなるドン | 新田たつお | 4500万 | 108 | 416,667 |
| 60 | DEAR BOYS | 八神ひろき | 4500万 | 89 | 505,618 |
| 61 | あっぱれ!浦安鉄筋家族 | 浜岡賢次 | 4400万 | 24 | 1,833,333 |
| 62 | 課長島耕作 | 弘兼憲史 | 4010万 | 17 | 2,358,824 |
| 63 | BE-BOP-HIGHSCHOOL | きうちかずひろ | 4000万 | 48 | 833,333 |
| 64 | 今日から俺は!! | 西森博之 | 4000万 | 38 | 1,052,632 |
| 65 | シュート! | 大島司 | 4000万 | 33 | 1,212,121 |
| 66 | 遊☆戯☆王 | 高橋和希 | 4000万 | 38 | 1,052,632 |
| 67 | 王家の紋章 | 細川智栄子 | 4000万 | 68 | 588,235 |
| 68 | ダイヤのA | 寺嶋裕二 | 3870万 | 47 | 823,404 |
| 69 | のだめカンタービレ | 二ノ宮知子 | 3800万 | 25 | 1,520,000 |
| 70 | 釣りキチ三平 | 矢口高雄 | 3800万 | 39 | 974,359 |
| 71 | シャーマンキング | 武井宏之 | 3700万 | 35 | 1,057,143 |
| 72 | 七つの大罪 | 鈴木央 | 3600万 | 41 | 878,049 |
| 73 | 20世紀少年 | 浦沢直樹 | 3600万 | 22 | 1,636,364 |
| 74 | クッキングパパ | うえやまとち | 3500万 | 162 | 216,049 |
| 75 | Dr.スランプ | 鳥山明 | 3500万 | 18 | 1,944,444 |
| 76 | イタズラなKiss | 多田かおる | 3500万 | 23 | 1,521,739 |
| 77 | 聖闘士星矢 | 車田正美 | 3500万 | 28 | 1,250,000 |
| 78 | ワースト WORST | 高橋ヒロシ | 3300万 | 33 | 1,000,000 |
| 79 | 3×3EYES | 高田 裕三 | 3300万 | 40 | 825,000 |
| 80 | 君に届け | 椎名軽穂 | 3250万 | 30 | 1,083,333 |
| 81 | ちびまる子ちゃん | さくらももこ | 3200万 | 17 | 1,882,353 |
| 82 | 約束のネバーランド | 白井カイウ出水ぽすか | 3100万 | 20 | 1,550,000 |
| 83 | 黒子のバスケ | 藤巻忠俊 | 3100万 | 30 | 1,033,333 |
| 84 | とある魔術の禁書目録 | 鎌池和馬 | 3000万 | 27 | 1,111,111 |
| 85 | YAWARA!ヤワラ! | 浦沢直樹 | 3000万 | 29 | 1,034,483 |
| 86 | じゃりン子チエ | はるき悦巳 | 3000万 | 47 | 638,298 |
| 87 | カメレオン | 加瀬あつし | 3000万 | 47 | 638,298 |
| 88 | うる星やつら | 高橋留美子 | 3000万 | 34 | 882,353 |
| 89 | サラリーマン金太郎 | 本宮ひろ志 | 3000万 | 20 | 1,500,000 |
| 90 | デスノート | 小畑健大場つぐみ | 3000万 | 13 | 2,307,692 |
| 91 | リボーン 家庭教師ヒットマンREBORN! | 天野明 | 3000万 | 42 | 714,286 |
| 92 | バスタード BASTARD!! | 萩原一至 | 3000万 | 27 | 1,111,111 |
| 93 | うしおととら | 藤田和日郎 | 3000万 | 33 | 909,091 |
| 94 | フルーツバスケット | 高屋奈月 | 3000万 | 23 | 1,304,348 |
| 95 | 黒執事 | 枢やな | 3000万 | 32 | 937,500 |
| 96 | 美少女戦士セーラームーン | 武内直子 | 3000万 | 18 | 1,666,667 |
| 97 | 転生したらスライムだった件 | 伏瀬川上泰樹 | 2900万 | 21 | 1,380,952 |
| 98 | 疾風伝説 特攻の拓 | 佐木飛朗斗 | 2800万 | 27 | 1,037,037 |
| 99 | ときめきトゥナイト | 池野恋 | 2700万 | 31 | 870,968 |
| 100 | ふたりエッチ | 克・亜樹 | 2700万 | 87 | 310,345 |
| 101 | 鬼平犯科帳 | 池波正太郎 | 2700万 | 74 | 364,865 |
集計に間違いがあればご指摘ください
せっかく課金したのにユーザが増えまくっているのか滅茶苦茶重くなっていて最悪。
だから流出したモデルを使ってローカルでNAIの環境を構築する。
ネットには情報もだいぶ転がってるけど陳腐化した情報があまりに多いため増田にまとめることにした。
もしかしたらこの記事もすでに陳腐化しているかもしれないが…単純に間違ってたらトラバで教えてほしい。
もちろん自己責任。この記事を見て導入した結果何かあっても増田は何も保証しない。
英語がわかる人はこっちを見た方が早いと思う。今は導入RTAができるくらい導入は楽になっている。
https://rentry.org/nai-speedrun
VRAMが2GB以上あるNVIDIA製のグラフィックボードがあればローカル環境を構築できる。
GPUの世代はGTX700シリーズ以降。なので一昔前のミドル級ボードでも動作するらしい。
IntelのオンボードGPUでも実行する方法があるらしい(stable_diffusion.openvino)が今回は割愛する。自分で探してね。
その他の推奨環境は以下の通り。
対応OS:Windows7以上(と言うがM1Macでも動作する方法があるとかなんとか)
メインメモリ:16GB以上(VRAMもたくさん必要だが起動時にメインメモリも大量に食う。WebUI起動時にタスクマネージャを見ているとよくわかる)
スマホしか持ってないような人やこういうのがよくわからない人はNovelAIを使った方が良いと思う。
今は重いけど、きっとそのうちみんな飽きてサーバも軽くなるかもしれないし。
(追記)NovelAIがリソースを確保してサーバが軽くなったからリスクを背負ってまで導入しなくても良いかも
(追記)Pythonは当然3系。最新の奴を入れれば問題無い。
導入方法はいちいち書かないけど、「python --version」や「git -v」で
正常にバージョン情報が出る(パスがきちんと通っている)ことはちゃんと確認しよう。
Stable DiffusionwebUIはStable Diffusionやそれをベースとした画像生成AIを利用するためのフロントエンド。
その中でも特に開発が活発でデファクトスタンダードとなっているのがAUTOMATIC1111版だ。
導入したい適当なディレクトリに対してPowerShellなどで
「git clonehttps://github.com/AUTOMATIC1111/stable-diffusion-webui.git」
なお、AUTOMATIC1111版は数時間単位でコミットが行われるから
定期的に「git pullorigin master」で更新しよう。
クライアントはqBitTorrentが一番楽だと思う。
ここにはさすがにmagnetリンクは書かないから各自ググって欲しい。
結構誤解されがちなことだが流出データ50GBを全部ダウンロードする必要は無い。
必要なファイルはanimefull-final-prunedディレクトリの中身とanimevae.ptだから5GBちょっとくらいなんじゃないかな。
もし余裕があるならmoduleディレクトリの中身もダウンロードすればいいけど、ぶっちゃけ必要無いんじゃないか?
まずはanimefull-final-prunedの中身のファイルをリネーム。
「model.ckpt」を「animefinal-full-pruned.ckpt」のようなわかりやすい名前にして、
「animevae.pt」を例えば「animefinal-full-pruned.vae.pt」のような拡張子以外は同じファイル名にする。
WebUI起動フォルダ配下の\models\Stable-diffusionにリネームしたファイルを移動させれば配置はOK。
ちなみにmoduleディレクトリの中身は\models\hypernetworksに移動させて使う。
それらのファイルを設定で適用させると画風が結構変わるがNovelAI再現とは関係無いみたいだ。
(追記)moduleディレクトリの中身の.ptファイルはhypernetworksという技術によって画風などを学習したものらしい。
すでに複数のイラストレーターの画風を学習したptファイルが作成されており議論を呼んでいる。
自分のグラボのVRAMが4GB未満の場合は「set COMMANDLINE_ARGS=」の後に
4GB未満の場合は「--medvram」、2GB未満の場合は「--lowvram」とパラメータを追加しておこう。
自分の持ってるグラボのVRAMがわからないときはGPU-Zなどで調べよう。
またGTX1600系固有のバグ(単色の画像が出力される)があるらしいので
その場合は「--no-half-vae」もしくは「--no-half」や「--precision full」とパラメータを追加。
ちなみにパラメータに「--xformers」を追加してxformersを導入・使用すると
消費VRAMが減って画像生成処理時間も短縮されるので是非導入しよう。
画像からdanbooruのタグをAIで調査するdeepdanbooruを利用する場合は「--deepdanbooru」を追加。
のようになる。
ターミナルやPowerShellなどでwebui-user.batを起動しwebUIの初期導入と起動を行う。
過去には手動でCUDA等を導入する必要があったが、現在はこの初期導入でだいたいの導入が行われる。
ずいぶん楽にはなったがその分初期導入の時間は結構長い。10分~20分くらいかかるかもしれない。
途中で導入処理がエラーで止まってしまった場合は管理者権限で実行するなどして対応して欲しい。
起動ができたらSettingで以下の設定を変更してNovelAIに近づける。
StopAtlast layers of CLIPmodelを2に、
これで設定は完了!
設定を合わせて完全にNovelAIと同じ内容になったのかを確認するテストがある。
出力結果から海外じゃHalloAsukaTestなんて呼ばれている。
これは初期SEEDをはじめとする設定内容が完全に一致していれば同じ出力結果を得られる仕組みを利用している。
プロンプトの内容:masterpiece, best quality,masterpiece,asuka langley sittingcross leggedon a chair
ネガティブプロンプトの内容:lowres, bad anatomy, bad hands,text,error,missing fingers, extra digit, fewer digits, cropped,worst quality,low quality,normal quality,jpeg artifacts,signature, watermark,username, blurry, artistname
https://i.imgur.com/Bfl5qJB.jpg
なお、このテストはAUTOMATIC1111のバージョンやxformersの適用状態によっては微妙に違う画像が出力されることがあるらしい。
xformersを適用させている増田の環境だと確かに二つ並べると間違い探しレベルの違いがあった。
「Boorutag autocompletion for A1111」を導入すればNovelAIのように自動でdanbooruのタグを保管してくれる。
画像生成AIモデルはStable DiffusionがOSSのため派生が結構多い。
自前で追加学習もできるため自前で学習した追加AIモデルを4chanのような掲示板などで共有する人もいるらしい。
しかしそのようなモデルの中にウィルスのような悪意のある動作を行うものもあるらしい。
FBIがペドフィリアを一網打尽にするためにIPアドレスなどの個人情報を抜き出す動作を行うロリ特化AIモデルを掲示板で配布していて
しかもそれには本物の児童ポルノが教師データとして使われている…などという都市伝説的な話が今界隈を賑わせている。
Dr.スランプとはおそらく初めは別世界だっただろう[要出典]
あれは元々タッチの続編を描く企画を編集が持ってきたが、あだち先生が難色を示してじゃあ明青学園ならで折あいつけたらしいからちょっと違う。
舞台が近すぎる(というか一緒)
サービス要素的にちらっと出る、アラレちゃん形式(作中のポスターとかテレビとかで)が一番近いんだが、
無しにしたのは、SF的世界なのに、共通してるものがないこと。まあ同じ作者なのでピラフのところの兵器とキャラメルマンが似てたりするが。アラレちゃんに元々カプセルコーポレーション技術が登場してたら同じ世界だと感じたと思う。
逆に、現実の日本とかと全く同じ世界すぎると、サービス要素感が強いので、強固たる同じ世界だという根拠が薄くなる
同じピザチェーンがあるとかそういうのでいい
Ifyou couldonlyeatone meal for therest ofyourlife, what wouldit be?
Ifyou couldliveanywhere, where wouldit be?
Whatisyourfavorite familyvacation?
What wouldyouchange aboutyourself ifyou could?
What motivatesyou to work hard?
Whatisyour biggest complaint aboutyour job?
Whatisyourfavoritebook to read?
Whatwas thelast movieyou went to? Whatdidyou think?
Whatdidyou want to be whenyou were small?
Ifyou could choose to doanything for a day, what wouldit be?
What wouldyou singatKaraoke night?
How wouldyourfriends describeyou?
36. Whatis the bestgiftyou have been given?
37. Whatis theworstgiftyou have received?
40. Where doyou seeyourself infive years?
42. Ifyou were a super-hero, whatpowers wouldyou have?
43. What wouldyou do ifyou won the lottery?
44. What form of public transportation doyou prefer? (air,boat,train, bus,car,etc.)
45. What'syourfavoritezooanimal?
46. Ifyou couldgo back intime tochangeone thing, what wouldit be?
48. How many pillows doyou sleep with?
49. What's the longestyou'vegone without sleep (andwhy)?
52. How often doyou buy clothes?
53. Haveyou ever had asecret admirer?
54. What'syourfavorite holiday?
55. What'sthe most daring thingyou've ever done?
56. Whatwas thelast thingyou recordedonTV?
57. Whatwas thelastbookyou read?
58. What'syourfavorite type of foreignfood?
59. Areyou aclean ormessy person?
60.Who wouldyou want to playyou in a movie ofyourlife?
61. How longdoesittakeyou toget ready in the morning?
62. What kitchen appliance doyou use every day?
63. What'syourfavorite fastfoodchain?
64. What'syourfavorite family recipe?
65. Doyoulove orhate rollercoasters?
66. What'syourfavorite family tradition?
67. Whatisyourfavoritechildhood memory?
69. How old wereyou whenyou learnedSantawasn't real? Howdidyou findout?
70.Isyour glasshalf full orhalfempty?
71. What's the craziest thingyou’ve done in thename oflove?
72. What threeitems wouldyoutakewith youon a desertedisland?
73. Whatwasyourfavorite subject in school?
74. What'sthe most unusual thingyou've evereaten?
76.Is thereanythingyou wished would come back into fashion?
77. Areyou an introvert or an extrovert?
78. Which of thefive senses wouldyou sayisyour strongest?
79. Haveyou ever had a surprise party? (thatwas an actual surprise)
80. Areyou related or distantly related toanyone famous?
82.Doesyour family have a “motto” – spoken or unspoken?
83. Ifyou were ruler ofyour own country what would be the first lawyou would introduce?
84.Whowasyourfavorite teacher in school andwhy?
85. What three things doyou think ofthe most each day?
86. Ifyou had awarning label, what wouldyours say?
87. What song wouldyou say best sumsyou up?
88. What celebrity wouldyoulike to meetatStarbucks for acup of coffee?
time clocks alarms appointments schedulestansign up form team volunteers meeting black recruitment recruitingneedsign up form 5kmarathon runningracemarathongreenshoessign up form
90. What'sthe most interesting thingyou can seeout ofyour office or kitchen window?
91.On ascale of 1-10 how funny wouldyou sayyou are?
92. Where doyou seeyourself in10 years?
94. Ifyou couldjoinany past or currentmusicgroup which wouldyou want tojoin?
95. How many languages doyou speak?
96. Whatisyourfavorite family holiday tradition?
97.Whoisthe most intelligent personyou know?
98. Ifyou had to describeyourselfas ananimal, whichone wouldit be?
以下の漫画が日本で一番売れている。 だから日本で一番偉大な漫画家100人に決まってるだろ。
https://www.mangazenkan.com/r/rekidai/total/
Permalink |記事への反応(16) | 19:53