@Alex_code_
Account based inThailand
About this account
- Account based in
- Thailand
- Connected via
- Thailand Android App
Account-level information from X, not a live location or the device used for a specific post.
Joined August 2025
- Tweets117
- Following172
- Followers4
- Likes174
Aleksei retweeted
谁把教材上传Arxiv了？
《Agentic Large Language Models》（智能体大语言模型）：来自荷兰莱顿大学的AI教材，内容涵盖大模型工作原理、Agent原理等相关AI内容，共472页。
地址：arxiv.org/abs/2610.13455
Aleksei retweeted
Introducing Indra, a 1M parameter OCR model:
- 49 pages/s on CPU, 11x faster than PaddleOCR and 17x Tesseract
- Beats them on OmniParseBench text (41.3 vs 35.8 and 19.3)
- Full layout and line bounding boxes
Aleksei retweeted
🎉白嫖预警！Cursor Pro 这次不用学生证
MLH 这周办 Global Hack Week，官方挑战原话：这周用 Grok Bot 搞开发？先送你个免费 Cursor Pro。13 岁以上、上班族都能报。
🔥步骤极简（懒得点，整条丢给 Grok Bot）
1. 打开 mlh.link/ghw，登录 MyMLH 报名 Hacktoberfest
2. 等 MLH 发来的兑换码邮件
3. Cursor 后台 Billing & Invoices 兑换，再手动升级 Pro
over！
懒人版：让 Grok Bot 替你报名，复制这句发给它👇🏻
"打开 mlh.link/ghw，帮我报名 MLH Global Hack Week: Hacktoberfest。遇到登录、邮箱验证码、勾选协议的地方停下来让我自己操作，别替我点同意，也别问我要密码。报完截图给我。"
登录那步点 Open computer 自己输，Bot 看不到密码。
顺便说一句，这期挑战 3 到 7 就是教你用 Grok Bot，报完能接着刷分。
10 月 16 日凌晨 1 点（北京时间）截止。
兄弟们，速度冲！
Aleksei retweeted
how to claim $4000 credits for $0
gmail gets rejected they want your own domain and matching work email
domain is ~$10. email is free. one site, three more applications
> buy the domain + get the email
- grab a .com on cloudflare domains (~$10)
- turn on whois privacy
- set up free email (zoho free works)
- add mx/spf/dkim
- create [email protected]
- make a simple 1 page site that loads and explains what you actually build
> google cloud start tier = $2,000
> go to: cloud.google.com/startup/pre…
- founded in last 24 months
- working mvp + clear business model
- planning to seek vc
- website domain, email domain, and billing account email must all match
- have your 18 character billing account id ready
- 12 months. a few business days (longer if manual)
> aws activate founders = $1,000 (select up to $5k)
> go to: aws.amazon.com/startups/cred…
- self funded, pre series b, founded last 10 years
- make a builder id with your professional email
- link an aws account on the paid plan (free tier doesn’t qualify)
- they ask about product, funding stage, company profile
- 5 - 10 business days
> deepgram = $1,000 starter (up to $100k possible)
> go to: deepgram.com/startup-program
- live or launching within 6 months
- make a deepgram account first
- keep your project id handy
- ~1 week review, then 24 - 48 hrs to land
- 12 months
> extra ones worth stacking
- azure self serve: $1k now + $4k after business verification (new azure account)
- cloudflare for startups: bootstrapped tier still exists (~$5–10k range)
- oracle for startups: $500 + 70% off for 2 years
- claude for startups: open to bootstrapped, but the free team + $1k credits are currently over capacity
> things you need to know
- 2k + 1k + 1k = $4k base. azure can push it higher
- google is discretionary. no agencies, consultancies, personal blogs, crypto mining, or token projects against local rules
- aws: count on the $1k. the extra up to $5k is select only
- credits expire. google + deepgram are 12 months. aws dates show in billing
> what not to do
- don’t apply from gmail or free email
- don’t fake a company or park a domain with no site
- don’t apply as an agency, dev shop, or consultancy
- apply != approved. these are for real products, build real products
one domain. matching work email. working site. stack the rest
bookmark this. terms change often. read each page before you apply. good luck.
Aleksei retweeted
passo a passo de como que eu consegui entrar no programa de startups do claude
1. cria uma conta nova e com dominio corpotativo (seu dominio proprio)
2. preencha os dados do onboarding
3. entre no link claude.com/programs/startups
4. preencha os dados, explique o proposito da sua empresa, resposta séria
5. você talvez não receberá email, mas o seu status vai mudar para "VERIFIED"
6. clique em "claude startups"
7. vai aparecer para pagar, mas tem um botão "SKIP FOR NOW"
8. agora clique no seu perfil no canto inferior esquerdo
9. clique em "CLAIM GIFT"
10. na primeira caixa, vai ter como dar CLAIM
Aleksei retweeted
hotpath-rs 是一个帮开发者找出 Rust 程序为什么慢、内存花在哪里的工具。它能检查哪些代码最耗时、数据库查询和网络请求是否拖慢了程序，以及多个任务有没有互相等待。结果会显示在实时面板和报告中，帮助开发者判断该优先优化哪里，也能交给 AI 助手辅助分析。
github.com/pawurb/hotpath-rs
Aleksei retweeted
I’m happy to share that our team's work on PiPNN—an ultra-scalable algorithm for high-dimensional Nearest Neighbor Search—has won three awards at KDD’26, VecDB’26, and SISAP’26 with up to 78x speedup on index build. Nearest-neighbor search is fundamental to deduping in data curation and embedding-driven retrieval for RAG for LLMs. It is great to see our work recognized across the community:
Best Paper at KDD’26: PiPNN: A Framework for Ultra-Scalable Graph-Based Nearest Neighbor Indexing. This paper introduces the PiPNN framework, which delivers an order-of-magnitude speedup for building navigable proximity graphs. (dl.acm.org/doi/10.1145/37708…)
Best Paper at VecDB’26: Turbocharging PiPNN for Proximity and 𝑘-NN Graph Building. The paper improved the original PIPNN kernels, resulting in an additional 8x speedup (up to a 78x total speedup!), and introduced a highly optimized GPU version. (openreview.net/forum?id=8Gzb…)
1st Place at the SISAP’26 Challenge: Our PiPNN-based solution ranked 1st for Task 1, which challenged teams to develop the fastest k-NN graph building algorithm. (sisap-challenges.github.io/2…)
Nearest-neighbor search is an important part of Data Curation for LLMs, and retrieval for RAG pipelines, and it only becomes more important as we move to more long-horizon agents and handling long-context inference-time scaling.
Congratulations to the team behind this work: Tobias Rubel, Richard Wen (UMD), Guy Blelloch, Laxman Dhulipala, Lars Gottesbüren(github.com/larsgottesbueren), Jakub Łącki, and Vahab Mirrokni.
Aleksei retweeted
we are offering bug bounties of up to $300k if you manage to exploit Muse. bug bounty program is here bugbounty.meta.com/ and more details about our approach to making Muse secure are here security.muse.ai
Aleksei retweeted
最近看了一篇很好玩的 paper：How to Write to SSDs
vldb.org/pvldb/vol19/p1469-l…
论文讲的是 write amplification 和 DB/SSD co-design，却让我重新想起了经典的 Law of Leaky Abstractions：
joelonsoftware.com/2002/11/1…
DBMS 以为自己优雅地写了一个 page，SSD 却在地下室忙着 FTL、GC 和搬家，最后一次 logical write 变成了好几次 NAND write。
Architecture must follow hardware.
抽象可以关上地下室的门，但不能让地下室消失。物理定律也不会因为 API 设计得好看就下班。
Aleksei retweeted
Every question on this benchmark ships with a list of the papers that count as correct, and those lists are incomplete.
We measured the gap on one question that asks for papers meeting a set of metadata conditions. Running those exact conditions over the corpus ourselves turned up 1,578 papers that satisfy all of them. The question's answer list holds 171, so a system that returns any of the other 1,407 qualifying papers is charged a mistake for each one.
The score also divides by a fixed target count rather than by the number of papers that actually qualify, so a question with many correct answers cannot be fully credited even by a perfect system. Every entry on the board pays these penalties, which is why published numbers run lower than they look and why the gaps between the leaders are small.
We report on the benchmark anyway, because a flawed measure that every system runs still lets you compare them, and a private one lets you compare nothing. IRIS is open source, and you can run it on Rasyn now.
IRIS is now available on Marigold for you to try: app.rasyn.ai
HuggingFace: huggingface.co/rasynai/rasyn…
Aleksei retweeted
🚨ANTHROPIC ESTÁ PAGANDO $3,850/SEMANA A PERSONAS SIN EXPERIENCIA EN ML.
Y más del 50% acaba trabajando en Anthropic.
Se llama Anthropic Fellows Program.
Dura 4 meses y te pagan semanalmente, además de darte financiación adicional para cómputo.
Lo interesante es que no necesitas un doctorado, experiencia en ML ni publicaciones previas.
Solo necesitas:
→ Saber Python
→ Resolver problemas ambiguos
→ Aprender rápido
Un investigador de Anthropic te propone varias ideas de proyecto.
Eliges una y la desarrolláis juntos durante 4 meses, haciendo investigación real sobre seguridad de IA.
Los resultados son bastante absurdos:
→ +80% de la primera convocatoria publicó un paper
→ +50% terminó trabajando en Anthropic
Hay 6 tracks abiertos ahora mismo, con opciones en San Francisco, Londres, Ontario y REMOTO.
Aplicar es gratis y puedes hacerlo en una tarde.
Dentro de 4 meses podrías estar investigando para Anthropic sin haber tenido experiencia previa en ML.
Aplica hoy mismo: job-boards.greenhouse.io/ant…
Aleksei retweeted
Causal AI and Bayesian Networks (680-page PDF eBook): web4.cs.ucl.ac.uk/staff/D.Ba…
+
Get a print copy here: amzn.to/4t1aLTC
———————
#DataScience #ML #MachineLearning #Causality #Inference #DataScientist
Aleksei retweeted
不想花钱，又想看书、追剧、听歌？这 15 个资源站可以先存着。
📚 电子书
1. Z-Library：zh.z-library.sk —— 综合电子书搜索
2. WeLib：zh.welib.org —— 图书、教材、学术资料
3. Anna’s Archive：annas-archive.gl —— 图书与论文搜索
4. 苦瓜书盘：kgbook.com —— PDF、MOBI 电子书
5. 电子书宝库：github.com/jbiaojerry/ebook-… —— GitHub 开源书单合集
🎬 影视
6. 在线之家：zxzjhd.com —— 电影、美剧在线观看
7. 泥视频：nivod.vip —— 电影、剧集、综艺、动漫
8. 发现 TV：faxiantv.cc —— 在线电影、电视剧
9. Libvio：libvio.lat —— 海外剧、电影资源
10. 不想看片：v.bxkp.org —— 影视搜索与多种资源入口
🎵 音乐
11. 泡椒音乐：pjmp3.de —— 在线音乐搜索与下载
12. 1Music：1music.cc/zh-CN —— 高品质音乐资源
13. music-dl：music.zkkp.nyc.mn —— 多平台音乐搜索
14. FLAC Downloader：flacdownloader.com —— FLAC 无损音乐
15. GD 音乐台：music.gdstudio.xyz —— 多平台音乐搜索与下载
Aleksei retweeted
🚀 x64dbg-MCP Server — fully automated, dynamic, agentic reverse engineering for x64dbg.
🧠 71 MCP tools
🤖 Fully automated & dynamic agentic RE
🔴 22 debugger event callbacks
⚡ Native Zig plugin — zero dependencies
🛠️ Breakpoints, memory, registers, tracing, xrefs, PE analysis & more
🌐 Streamable HTTP + SSE
📦 x32 + x64
Let your AI agent control, analyze, debug, and dynamically reverse engineer binaries directly through x64dbg.
🔗 github.com/duty1g/x64dbg-mcp…
#x64dbg #MCP #AgenticAI #ReverseEngineering #CyberSecurity #RedTeam #MalwareAnalysis #Zig
Aleksei retweeted
Google just quietly dropped something big for agent builders
It's called SAM: Sovereign Agent Mesh
A peer-to-peer network where AI agents auto-discover each other, authenticate every packet and call tools across the mesh without a central server
Think of it like BitTorrent, but for your agents
Right now every agent stack you build is a walled garden. Claude Code talks to its MCPs. Codex talks to its MCPs. If you want agents running on your laptop, your VPS, and your Mac Studio to share tools with each other, you're duct-taping SSH tunnels and REST endpoints together like it's 2015
SAM makes that whole layer disappear
You run a sam-node on each machine and the nodes find each other automatically. Every connection is cryptographically identified, so your agents plug in and start calling tools across the whole network
The identity is portable too. Same node, same keys, whether it's on GCP, your laptop, or a Raspberry Pi in your closet
If you're serious about running agents across multiple devices, get on the testnet this weekend:
1. Install sam-node (binary or Docker)
2. Point it at bananas.sam-mesh.dev
3. Connect Claude or Gemini to your local MCP endpoint
4. Watch your agent discover tools running on other people's nodes
Still early. Not officially supported by Google.
This is where the agent stack is heading. Get on it early
Repo: github.com/google/sam
Aleksei retweeted
谁下次再给你说Github不能搞钱
把这8个仓库甩他脸上！
先收藏再看，离搞钱更近一步！
1、github.com/XiaomingX/ai-mone…
这个网站主要是关于各种赚钱的信息差，涵盖AI副业赚钱、初创公司0-1避坑、赚钱认知等等干货内容。
2、github.com/Usagi-org/ai-goof…
这个牛的！做闲鱼的朋友必须要去看，闲鱼智能监控仓库，可以实时监控闲鱼商品、价格，还可以多账号运营，专业闲鱼玩家必看！
3、github.com/lukasz-madon/awes…
数字游民必备，这个仓库里面收集了上百个远程工作招聘网站，支持5种语言，还有面试、租房等等一系列干货内容。
4、github.com/AZeC4/TelegramGro…
不会用TG的看过来！仓库里面有10000+ 优质Telegram群组、Telegram频道推荐，让你轻松玩转Telegram。
5、github.com/sindresorhus/awes…
这个网站主要是关于收集创作内容，获取灵感，可以一键获取别人整理好的选题、内容、素材等等，帮助你快人一步做内容！
6、github.com/n8n-io/n8n
n8n自动化工作流平台，提供专业的AI自动化解决方案，涵盖9000+工作流模版，1500+工作集成，帮你提高工作效率，节省工作时间！
7 github.com/paperswithbacktes…
超厉害的量化交易仓库，40多种量化交易策略，97个有关交易的仓库和软件包，还有大量适合学习量化交易的数据，新手友好！
8、github.com/awesomedata/aweso…
超级优质的公共信息来源，项目由上海交大的陈夏明博士创建，致力于收集整理各个领域的高质量、主题明确的信息，更新频率高。
Aleksei retweeted
🚨LAUNCH🚨
We keep our promises. HelixDB is properly open-source at last. We are accepting contributions again and look forward to growing the community 😃💻
Github: github.com/HelixDB/helix-db
Aleksei retweeted
We partnered with @EngramLab to train a model that learns from a firm's accumulated knowledge.
We first built a 100M token synthetic law firm containing ~10k documents across 250+ matters and 46 clients.
The model then internalized the firm's knowledge in two ways:
1) parametric knowledge from training on the firm's documents
2) 1M tokens of structured notes on key matter-level features
These two forms of memory allow the model to understand and search the firm's work much more efficiently across 250 representative legal tasks.
It bested every model we tested, with a 67% mean criteria score at just $0.13 per query - 10-24x cheaper than Opus 4.8 on high/xhigh reasoning.
The model also exhibited interesting qualitative behavior after training: it began 99% of trajectories with a targeted search and reached the first relevant matter in 3.7 turns, versus 5.9 for the base model.
It also acquired substantial knowledge of the firm itself. When asked questions about individual matters without using any tools, its accuracy increased from 4.7% to 72.6% over the course of training.
More with @EngramLab:
