진용진 (Yongjin Jin) (@jjins) header
진용진 (Yongjin Jin) (@jjins) profile picture

😀

South Korea
Joined October 2008
Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)Photo by 진용진 (Yongjin Jin) (@jjins)
진용진 (Yongjin Jin) retweeted
What is effort really? When do you change it it and why not just use max effort for everything? I dove deep into this problem, looking into evals and doing my own tests and I was quite surprised by the results.
301
516
241
6,269
1,517,334
진용진 (Yongjin Jin) retweeted
Too many AI personal assistants, too little time to assess all of them. Nice work here: assistantbenchmark.com
218
271
141
3,355
1,036,147
진용진 (Yongjin Jin) retweeted
AI has collapsed the cost of building products, but not the cost of knowing what to build. - The old loop: spec, scope, build - The new loop: build, play, design, ship Prototypes are cheap. Great products still require taste, judgment, and a clear understanding of what users actually need. a16z's @joshelman on the artifact a product manager actually produces: a16z.news/p/product-manageme…
150
473
101
3,537
453,784
진용진 (Yongjin Jin) retweeted
Anthropic の 自己改善型 Agent の構築方法の Blog です。 ドメイン特化の base skill に対し、ユーザーの FB を収集し反映する improver skill を定期実行し、その際に FB を楽に与える仕組み (Issue/PRに直接コメント) を整備する等のベストプラクティスが紹介されてます。 claude.com/blog/how-warp-bui…
2
80
5
798
57,543
진용진 (Yongjin Jin) retweeted
Anthropic engineers just showed how they build a full app from scratch, using a loop of agents 40 minutes from the team behind Claude Code they used three agents: one to plan, one to build, one to judge, cycling until the app actually works the winners won't have the smartest model, they'll have the best loop watch it, then read the full guide on how to actually use loops below
217
1,136
52
8,871
1,937,108
今週GitHubで急上昇したAIリポジトリ10選: 1. headroom ツールの出力・ログ・ファイル・RAGの塊を、AIに渡す前に圧縮。トークンを60〜95%削っても回答は変わらない。ライブラリ・プロキシ・MCPサーバーとして使える。 github.com/chopratejas/headr… 2. hermes-agent 使うほど賢くなる自己成長型のAIエージェント(Nous Research製)。スキルを自分で作り、セッションをまたいで記憶を保持、200以上のモデルやTelegram・Slack連携にも対応。 github.com/NousResearch/herm… 3. ECC Claude Code・Cursor・Codexなどのエージェント環境をまとめて底上げ。スキル・記憶・セキュリティを1パックに、64エージェント+261スキル入り。 github.com/affaan-m/ECC 4. codegraph コードを先にナレッジグラフ化して、AIが毎回ファイルを探し回らずに済む。ベンチマーク平均でトークン47%減・ツール呼び出し58%減、完全ローカル動作。 github.com/colbymchenry/code… 5. last30days-skill どんなテーマでもReddit・X・YouTube・Hacker News・ウェブを横断調査し、根拠付きで要約。反応の大きい情報から拾うAIエージェント用スキル。 github.com/mvanhorn/last30da… 6. MoneyPrinterTurbo キーワードを入れるだけで短尺動画が完成。台本・素材・字幕・BGMまで自動生成、縦9:16も横16:9も出せる。 github.com/harry0703/MoneyPr… 7. VoxCPM 多言語対応の音声合成(読み上げ)エンジン。声のデザインから、本人そっくりの声まねまでこなす。 github.com/OpenBMB/VoxCPM 8. Agent-Reach AIエージェントにネットを"見る目"を与えるCLI。X・Reddit・YouTube・GitHub・Bilibili・小紅書を横断で読めて、APIキー課金なしで動く。 github.com/Panniantong/Agent… 9. Open-LLM-VTuber 好きなAIモデルと声で会話できるVTuber。ハンズフリー・割り込み発話に対応し、Live2Dの顔がローカルで動く。 github.com/Open-LLM-VTuber/O… 10. PaddleOCR PDFや画像を、AIがそのまま扱える構造化データに変換。軽量で100言語以上に対応するOCRツールキット。 github.com/PaddlePaddle/Padd…
22
240
10
1,556
128,639
진용진 (Yongjin Jin) retweeted
OMG these numbers
Replying to @claudeai
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision. The longer and more complex the task, the larger Fable 5’s lead over our other models.
23
15
3
467
45,086
진용진 (Yongjin Jin) retweeted
分享一篇文章:《How LLMs Actually Work》 0xkato.xyz/how-llms-actually… 好像是前几天 HackerNews 排名第一来着,类似的文章很多,但是这篇深入浅出和直观的例子非常适合有一定编程但没深入学Transformer的人阅读,里面的比喻也恰当,一看就是活人写的,没什么 AI 味道。 最近重新爱上了写东西,写了两篇技术文章,之后还会继续写,而且我的一个原则,活人写,绝对不用 AI,写作是一种乐趣,梳理逻辑,表达观点,不要让这种乐趣被 AI 剥夺。
23
133
3
635
84,902
진용진 (Yongjin Jin) retweeted
Claude Code creator: "I don't prompt Claude anymore. I create loops - and the loops do the work. My job is to create loops." In 18 minutes , Boris breaks down his real daily Claude Code setup. Claude + agents + loops + routines + dynamic workflow Worth more than a $500 vibe-coding course
43
73
3
601
162,445
진용진 (Yongjin Jin) retweeted
If you’ve been procrastinating on learning loops, read this.
28
161
7
1,547
525,383
진용진 (Yongjin Jin) retweeted
Excited to share how Anthropic's data team has automated 95% of business analytics queries with Claude. Blog post covers how we approach evals, ablations, and online validation!
59
118
6
2,962
884,750
진용진 (Yongjin Jin) retweeted
Once you internalize the auto-research concept and use /goal with it, you unlock ridiculously fast app improvement on anything that has a numeric rubric. We’re all still too rooted in the “labor is expensive” world.
68
68
12
1,488
296,697