ahmad walidurrosyad

[MG193_7 @MGAldys4]

Guys, ASC is now on pip:
pip install droidasc

Optimize global xrefs with bytes.translate → ~30% faster.
Friends already using it to mass-scan big vendors and finding one-click bugs. Even my bug from yesterday got duped…
Not sure open-sourcing this was a good idea😂
#BugBounty

[ahmad walidurrosyad @rosyadrasta]

@ardisatriawan Parah sih

[Garry Tan @garrytan]

Downloading now... 1M token context window with supposedly usable coding agent capability all on a 128GB Macbook Pro is 🤯

[tylerni7 @tylerni7]

If people are really curious about https://t.co/zJqjRZkKcD, @5unKn0wn is the GOATed researcher who is responsible!

[Telegram Messenger @telegram]

On Telegram, anyone can utilize AI bots to easily develop, launch and manage their own bot – with no coding required. #TelegramTips

More information for developers is available here:
https://t.co/xBsBGCc664

[ahmad walidurrosyad @rosyadrasta]

@LayerZero_Core 👏

[The OSINT Newsletter @osintnewsletter]

Telegram for macOS can render spoiler text as pseudo braille.

Screenshots or screen shares can allow the hidden text to be recovered.

Spoilers ≠ security.

Treat them like blur, not encryption.

Try it here: https://t.co/mfqGTcGnJZ

[Bour Abdelhadi @BourAbdelhadi]

https://t.co/yQMOff2nhe

[Simon Le Marchant @marchantweb]

Big news! 🔥

Yesterday I left my role at Clerk to go all-in on Shaders, a brand new way to bring creative effects to frontend frameworks!

🚀 Early access it right around the corner. Join the waitlist today at https://t.co/Jq8h3jH3GR and let me know what you think!

Here’s the first look 👇

[Victoria Slocum @victorialslocum]

From context engineering to embeddings, chunking might be the most underrated part of the pipeline.

It has more of an impact on performance than most people think 👀

Chunking is the pre-processing step of splitting texts into smaller pieces - the "chunks" that become the actual units stored in your vector database. Each chunk gets vectorized and determines what information gets retrieved when you search.

Top 𝗰𝗵𝘂𝗻𝗸𝗶𝗻𝗴 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀 you should know:

📏 𝗙𝗶𝘅𝗲𝗱-𝘀𝗶𝘇𝗲 𝗰𝗵𝘂𝗻𝗸𝗶𝗻𝗴: The simplest approach - split text into chunks of consistent size (e.g., 100 words, 200 characters). Easy to implement but might break sentences awkwardly.

🔄 𝗥𝗲𝗰𝘂𝗿𝘀𝗶𝘃𝗲 𝗰𝗵𝘂𝗻𝗸𝗶𝗻𝗴: Hierarchically splits documents, preserving structure while creating manageable chunks. Great for maintaining context across different levels of detail.

📄 𝗗𝗼𝗰𝘂𝗺𝗲𝗻𝘁-𝗯𝗮𝘀𝗲𝗱 𝗰𝗵𝘂𝗻𝗸𝗶𝗻𝗴: Uses natural document markers like paragraphs, sections, or chapters as boundaries. Keeps related information together but can create wildly different chunk sizes.

🧠 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗰𝗵𝘂𝗻𝗸𝗶𝗻𝗴: Variable-size chunks based on meaning rather than arbitrary markers. More sophisticated but computationally intensive.

⏰ 𝗟𝗮𝘁𝗲 𝗰𝗵𝘂𝗻𝗸𝗶𝗻𝗴: Embeds first, then chunks - preserving more contextual information in the vectors themselves.

There's really no one-size-fits-all strategy, like most things, it depends on your data. Your chunking approach directly impacts both information retrieval AND the contextual information provided to your LLM, and is definitely part of this new world of context engineering.

My recommendation: start with fixed-size chunks of ~200-500 tokens with some overlap, then experiment based on your specific use case. Document structure matters - technical documentation might benefit from section-based chunking while narrative text might work better with semantic approaches.

Check out this page for a deep dive into setting up chunking: https://t.co/PFIp5hTlr9

or this ebook for more deep dives into advanced RAG strategies: https://t.co/r4liiopeGS

[Kristian @kristianfreeman]

Generating PDFs is hard. Including puppeteer in your app bundle is error-prone outside of Javascript apps.

Use Cloudflare's Browser Rendering API. You can add auth, deploy it, and use it to render PDFs across projects.

https://t.co/J5Qc0pgLGk

[ahmad walidurrosyad @rosyadrasta]

@LightningAI any chance I can get priority verification for Lightning Studios?

[GSAP @greensock]

🥳 It's an exciting day! 🥳

Thanks to @Webflow GSAP is now 100% FREE - including ALL of the bonus plugins like SplitText, MorphSVG, and all the others that were exclusively available to Club GSAP members.

We can't wait to see what you make!

[Rad @rad9800]

16,582 OpenAI keys - all active as of this post, with credit deposited on the account for this project.

Y'all have my authorization to go crazy in this environment. Good luck, and please be kind to my wallet 🙏

https://t.co/Hz62zcTfVi

[ginobefun @hongming731]

刚看到 Jina AI 发了个新东西，叫 jina-reranker-m0，感觉超有料，必须跟你唠唠！

这玩意儿是个 reranker，你知道吧？就是帮你把一堆搜索结果重新排个序，让最相关的排前面。但牛逼之处在于，它不光能处理文本，还能看懂图片、图表、表格这些视觉元素！而且支持 29 多种语言，简直是跨语言、跨模态的全能选手！

你想想，以前搜东西，如果文档里关键信息是在一张图或者表格里，普通 reranker 就抓瞎了。但 jina-reranker-m0 能直接理解这些视觉内容，然后告诉你这份带图的文档是不是和你搜的东西最相关。比如你搜 "small language model data extraction"，它能把包含相关论文截图、图表的文档排在前面，甚至比纯文本描述的文档得分还高！

跟它家以前的纯文本 reranker (v2) 比：

- 核心升级： 能处理图片了！这是质变。

- 文本能力也强了： 就算是纯文本，它在处理多语言、长文档、甚至是代码搜索方面，性能也比 v2 有提升。

- 架构大换血： 这次没用老套的 cross-encoder，而是基于 Qwen2-VL 搞了个 decoder-only 架构。参数量也大了不少 (24亿 vs 2.78亿)。

这个新架构有啥好处呢？

- 真正多模态： 能无缝处理文本+图片的混合输入，解决了以前文本向量和图像向量"玩不到一起去"的"模态鸿沟"问题。现在你可以放心地把纯文本文档和图片混在一起让它排。

- 上下文更长： 支持的输入 token 更多了（理论 32k，训练 10k）。

- 潜力巨大： 文章说这种 decoder 架构未来可能更容易实现更高级的功能，比如列表式重排（一次性看整个列表来排序，更智能）和可解释性（更容易知道模型为啥这么排）。

怎么用？

- API 调用： 最省事，注册就有免费额度，直接扔文本查询和文档（可以是文本、图片 URL 或 base64）就行。（不过 API 暂时不支持用图片做查询）

- Hugging Face： 开源大法好！可以直接下载模型本地跑，更灵活，支持图片查询、图文混合文档。他们还贴心地准备了 Colab 教程。

- 云平台市场： AWS、Azure、GCP 上架了，可以直接订阅部署。

感觉 RAG 这块儿，有个能理解图文的 reranker 会是个巨大的提升！以后喂给大模型的上下文质量能更高了。

[DeepSeek @deepseek_ai]

🚀 Day 4 of #OpenSourceWeek: Optimized Parallelism Strategies

✅ DualPipe - a bidirectional pipeline parallelism algorithm for computation-communication overlap in V3/R1 training.
🔗 https://t.co/NR3047Nnpo

✅ EPLB - an expert-parallel load balancer for V3/R1.
🔗 https://t.co/IilrQUfn49

📊 Analyze computation-communication overlap in V3/R1.
🔗 https://t.co/tY1WujA1Lw

[sudo rm -rf --no-preserve-root / @pcaversaccio]

TL;DR: A dev machine of Safe was compromised. This allowed access to AWS and their S3 bucket. A malicious JavaScript was pushed to the bucket and eventually distributed. The malicious JS code targeted specifically the Bybit contract address. The JS code changes the content of the transaction during the signing process.

Bybit shared to investigation reports here: https://t.co/2E9KCeOKPf

[s1r1us @S1r1u5_]

So, this is how lazarus drained 1.5 Billion

1) malicious JS injected into Safe{Wallet} at https://t.co/s8gAO3DM4d (because apparently, one of the nk devs just casually pushed it to production 🤡)

2) the JS modified executeTransaction() only if the signer was in a predefined list (Bybit’s multisig owners).

3) modified transaction now sets operation: 1 (delegatecall) to attacker address instead of a normal call.

4) delegatecall hits the attacker contract, which changed Safe contract's first storage slot which is masterCopy to a another attacker contract.

5) new masterCopy contract contained sweepETH() & sweepERC20(), draining $1.5B

[ahmad walidurrosyad @rosyadrasta]

@AgentOyen @MilkyOnSolana I hold the token, how can I access the beta tools

[H4x0r.DZ @h4x0r_dz]

Wait what !??