这个周末写了一篇略带中二的长文：《我在 AI 异世界重生为 #FDE ，来自北美 #AIAgent 公司的⼀线实录》。
FDE 到底干嘛、为什么越来越多 AI 公司在招？一句话：把 AI 产品量体裁衣，送到“最后一公里”，并对结果负责。欢迎拍砖👇 superlinear.academy/c/resour…
ALT 裁缝为客户量体的画面，标注 Customer/FDE/AI Product；配两张手绘饼图展示 FDE工作时间分布。
✈️ Air Canada is on strike. When will it end? Nobody knows.
In the meantime, check this out:
UI TARS Desktop = open-source version of OpenAI Operator / Manus. Automate your browser like an AI agent 🤖
GitHub: github.com/bytedance/UI-TARS…
Remember when every tech company was building a search engine? Now it’s LLMs, too many of them. Earlier this month, Kimi K2 impressed me. This week, GLM shows up and claims they are best in China, and the pricing is wild 🐗
🍜Ramen and 🤖LLMs
In Vancouver, a bowl of ramen = $20
In Shanghai, similar bowl = $5
Claude4, one coding convo = $5
Kimi K2 (Claude-compatible), same convo = $0.15
You can use K2 in Claude Code. When you get warning for $5, it's actually just $0.15, and 95% as good as Claude
💰Bitcoin hit $112,152.91 earlier today. Did you sell at the top? With @timeplusdata you can build real-time charts, alerts, even trading logic — all in SQL.
📈 Live demo: play.demo.timeplus.com/defau…
📚 Tutorial: demos.timeplus.com/#/id/mark…
Every week there’s a new tool claiming to support Apache Iceberg. But let’s be honest: demo is demo. Don’t fully trust the blog posts or shiny videos. Devil’s in the details. Iceberg is very friendly to Java and S3 users, not so much for GCP, not mention C++, but we're solving it
Google is very good to deprecate things. The 0605 model now is scheduled to be disabled in 2 weeks and we don’t know what will be the new name to set.
I used to pay hundreds a year for @docsbotai Today LLMs can handle millions of tokens, so I have a naive #n8n workflow: for each q, just to feed entire docs markdown to LLM and it just works. It got me thinking: are you solving a hard problem or just one LLM can brute-force?
Still broken after 12 hours 🤣 For infra, multi-cloud is a no-brainer. Same should apply to AI. But unlike infra, you can’t just run the same code. Each LLM has its own vibe. Swapping models changes the product feel
Inner Isshin might be the hardest boss in 'Sekiro Shadows Die Twice': fast, multiple weapons at once. Sounds familiar? That’s you, when working with a SQL engine like Timeplus, you can bring your own "weapons" -define your own functions with JavaScript, Python, or any HTTP Call
I just built a #UDF function to check if a number is prime (code: github.com/timeplus-io/examp…) And here’s the real boss move: SQL UDF: 158K events/sec JavaScript UDF: 660K eps Python UDF: 443K eps
On a 2-core VM. No tricks. Just fast.
Found a way to turn #excalidraw to animated arrows like this, inspired by the generated SVG from draw.io Code example github.com/excalidraw/excali…
Bitcoin hit an all-time high! 🚀 Build your own real-time trading strategies with #streaming #SQL with @timeplusdata cloud or self-hosting. Try it out at demo.timeplus.cloud! 📈😉
❤️ the blog ververica.com/blog/introduci… Challenges of Kafka: no natively support upserts; lacks direct query; have to replay entire log or send to OLAP for historical analysis; high network costs and bottleneck. That's why #Fluss #Paimon #Timeplus are future-proof, in open-source way
Just for fun, you may not know how to calculate PI with SQL. The fancy term is "Monte Carlo Simulation Approach". For more details, please check the blog from Gang Tao: timeplus.com/post/calculatin… live demo available
If you need to load data to Timeplus from #postgres, #mysql, #s3, #csv or even #Prometheus, you can just grab @SlingDataIO the 50MB binary for your Windows/Linux/Mac and run it with a single command 👉 docs.timeplus.com/sling – at Vancouver, British Columbia
