Open source clone of Claude Science shipped in 5 days. Anthropic launched Claude Science on June 30. By July 5, OpenScience was live. Apache 2.0, built by Synthetic Sciences , runs entirely on your machines. Same job. Literature, hypothesis, code, experiment, write-up. One agent runs the whole loop. Here's what's actually different: → Any model. Claude, GPT, GLM, Kimi, DeepSeek, your own fine-tune. Switch per request, so one lab's pricing never holds your research hostage. → 250 research skills across ML, comp bio, cheminformatics. Plain files you can read and edit, so the tool bends to your lab, not the other way around. → 30 scientific databases as agent tools. UniProt, PDB, ChEMBL, arXiv. Now zoom out. Claude Code got OpenCode. Cowork got OpenWork. Science got OpenScience in under a week. Every flagship gets a free twin before the launch tweet stops trending. The moat was never the product. It's how fast you ship the next one.

PDB apaan ih..

Apa kita gausa buka sirkel 00 tapi mjd trio PDB

En un PDB es posible

coba cek deh sumbangsih tokped terhadap pdb pas masa jaya2 berapa persen. Emg tolol regulatornya

Elle a tweeté depuis le local de la PdB

pdb jest popierdolone

Después, la obviedad que podemos criticar desde una mirada moralista que es "¿Por qué trabajaban en un medio corrupto con gente que lava guita?" la respuesta es que: - Todos los medios que cobran publicidad politica estan en esa. Apoyen a los medios independientes como PDB.

Menengah atas Vietnam ya mentok diatas kertas aja. Alias ya ga sesignifikan angka pdb , kek masih dibawah Indonesia Ama Thailand banget Lah Vietnam infrastrukturnya, apalagi Filipina.

Morrow listening on the Vox (he canonically killed his Commissar on Armageddon after the man fell to Khorne):

Pemerintah bersama Badan Anggaran DPR RI telah menyepakati asumsi dasar ekonomi makro dan indikator pembangunan yang akan menjadi dasar penyusunan Rancangan Anggaran Pendapatan Dan Belanja Negara (RAPBN) tahun 2027. Setelah kesepakatan ini, Pemerintah akan menjadikan rancangan awal RAPBN 2027 sebagai bahan untuk menetapkan angka usulan awal rancangan Undang-Undang APBN beserta nota keuangannya pada 16 Agustus 2026. Kemudian akan dibahas lebih detail lagi bersama dengan Komisi XI DPR, sebelum akhirnya ditetapkan sebagai UU pada akhir Oktober 2026. Pemerintah dan DPR sepakat pertumbuhan ekonomi pada tahun 2027 ditargetkan bisa tercapai dalam rentang 5,8 - 6,5%. Laju inflasi di rentang 1,5 -3,5%. Nilai Tukar Rupiah 16.800 - 17. per dolar Amerika Serikat. Sementara itu, pendapatan negara tahun depan ditargetkan di kisaran 12,01 - 12,40% terhadap Produk Domestik Bruto. Angka tersebut lebih tinggi dibandingkan usulan awal Pemerintah yaitu 11,82 - 12,40% PDB. Saksikan pembahasan selengkapnya dalam program Market Review bersama Prasetyo Wibowo @prast_ulrich, Senin 6 Juli 2026, pukul 21.00 - 21.30 WIB. LIVE di IDX Channel (Indihome Ch 119, MNC VISION Ch 100, MNC PLAY Ch 128 HD, First Media Ch 389, K-VISION Ch 129, OXYGEN Ch 137 dan MY REPUBLIK Ch 576). Saksikan juga LIVE STREAMING di idxchannel.com dan aplikasi IDX Channel TV di APPS store. #idxchannel #idxchannelcommunity #marketreview #ekonomimakro #rapbn2027

Most devs debug by spraying `print()` statements everywhere. Better move: use a debugger (`pdb` in Python, breakpoints in your IDE). Set a breakpoint, step through code line-by-line, inspect variables live. Takes 30 seconds to learn, saves hours of guesswork. Your future self will thank you.

CryoEM structure of the MLLT3_YEATS domain with SL4 of 7SK ncRNA (PDB code: 9AA8) #scivis #molecularart @dzine_ai @proteinimaging behance.net/gallery/25223489…

Claude Science的开源平替来了：openscience，给个目标就自己读文献→写代码→跑实验→出报告，整个科研循环在一个浏览器工作台里连着跑完。 跟 Claude Science 的区别是"不受制于模型"——任意模型provider 的 key均可、不用注册、BYOK 永久免费。其拥有250 skill、约 30 个科学数据库当工具，分 biology/physics/ml 角色 agent。 它是个正经的面向科学家的科研方向开源research harness，不是套壳： 1、分角色 agent：默认 research，另有 biology / physics / ml 专家，外加 critique 和文献综述子 agent，还有一个只读的 plan 模式先给你看计划再动手。 2、30个科学数据库当工具直接查：UniProt、PDB、Ensembl、ChEMBL、PubChem、arXiv、OpenAlex、Semantic Scholar，加起来约 30 个。 3、250 科研skill：从训练（DeepSpeed、PEFT、TRL）、评测、数据集，到化学信息学、LaTeX 出图、云算力（Modal、Tinker）。 4、浏览器就是工作台：文件树、编辑器、终端、会话历史，分子/结构/基因组/图表内联渲染；再往外接 LSP、MCP、插件、TypeScript SDK。

## dante's AI News 1. Hugging Face revamped 🤗 Kernels into a first-class Hub repo type for custom GPU kernels. The useful part for builders is security: trusted kernel publishers, signing with Sigstore/cosign, source SHA provenance, Nix-based reproducible builds, new CLIs, and compatibility metadata for accelerators, OSes, and backend versions. huggingface.co/blog/revamped… 2. A new arXiv study tested whether cleaner code changes coding-agent performance. In 660 Claude Code trials, cleanliness did not change pass rate, but cleaner repos used 7-8% fewer tokens and cut file revisits by 34%. Maintainability now shows up as agent operating cost, not just human taste. arxiv.org/abs/2605.20049 3. @github says git worktrees are having a real AI-era comeback. The reason is practical: parallel Codex/Copilot/Claude Code sessions need isolated working directories so agents can explore fixes without dirtying the same branch or stash. x.com/github/status/20738380… 4. @OpenRouter shared a case where its MCP cut inference cost by 24x at comparable quality. The MCP lets agents use OpenRouter's benchmarks, aggregate usage data, live performance stats, and codebase context to pick and test models instead of guessing from old training data. x.com/OpenRouter/status/2073… 5. Tomasz Tunguz modeled when AI spend can overtake engineering payroll. The current spread is huge: frontier AI-native firms can spend more on compute than people, top software firms are already near $89k per engineer per year, and the median company is still close to zero. Agentic workflows are the swing factor. tomtunguz.com/ai-spend-break… 6. Synthetic Sciences released OpenScience, an Apache-2.0, model-agnostic AI workbench for scientific research. It runs locally, supports Claude/GPT/Gemini/GLM/Kimi/DeepSeek/local models, includes 250 editable research skills, and exposes tools for sources like UniProt, PDB, ChEMBL, arXiv, OpenAlex, and Semantic Scholar. marktechpost.com/2026/07/05/… 7. Magic Patterns introduced a Design System Agent that can import real components, tokens, styles, assets, and rules from a codebase, then generate production-ready UI against that system. The direction matters: AI design tools are moving from generic mockups toward code-backed product systems. magicpatterns.com/blog/intro… 8. Runway opened its first France office, a Paris research hub focused on world models and physical AI. The initial team is 10 people, with a stated $30M regional investment and more Paris hiring planned. runwayml.com/news/announcing… 9. @rohanpaul_ai clipped Mark Zuckerberg describing Meta's Prometheus as a first gigawatt-plus single AI cluster, with hundreds of billions of dollars in capital behind the buildout. The strategic frame is blunt: concentrate elite people, capital, and infrastructure in one place. x.com/rohanpaul_ai/status/20… 10. @SemiAnalysis_ reports NVIDIA's Kyber NVL144 rack architecture has slipped to 2028, with NVL72x2 cancelled and CPO NVSwitch not expected until Feynman. Treat it as a sourced market report, but the implication is important: AI scale-up architecture is now a supply-chain and manufacturability story, not just a GPU roadmap story. x.com/SemiAnalysis_/status/2… Follow me for more daily AI, dev tools, and product design signals. Tap my profile for the next digest.

久しぶりにMolecule of the Month見に行ったらハンタウイルスのエンベロープだった PDB-101: Molecule of the Month: Hantavirus

New PDB Deposition from I04 pdbe.org/8PWW

New PDB Deposition from I04 pdbe.org/8PO3