AI wrote this code and introduced a vulnerability. Can you spot it? . I've seen Claude, Gemini and GPT write similar code and probably exists in the "3M" sites build on Bolt and Loveable too. We need a Data Layer that creates a "pit of security".

Using strongly typed programming languages like TS for vibe coding seems to be no-brainer because of lint’s ability to guide AI towards the root cause. Anything that is pro-feedback is good. Anything that stops AI from getting too far in the wrong path is good.

Building my own personal finance app (inspired by Midday) to finally get a grip on my finances and freelance expenses. Goal: have something ready by Monday that spits out a tax report for my German Tax Advisor. I've a year's worth of receipts to reconcile. Wish me luck.

Found a VS Code Extension that makes TypeScript Previews on Hover look so much better!.

Just started using TS on a side-project, but this is just too ugly and hard to read. Is there anyway to show a more cleaner IntelliSense?

When you fire Soham, he gets a job at two new companies.

RT @tanayvk: I'm launching Nuxflare Chat - blazing-fast, open-source AI chat that stores everything locally in the browser. 🏗️ Built with….

RIP Pair Programming.Long live Pair Reviewing.

Blessed with 20 min of morning rays, all thanks to @hubermanlab’s holy wisdom! . #SunlightSavior #HubermanHalo

Is it just me, or does everyone end-up skipping over a button while nappy changing a squirmy baby? . Seriously, why don't they just alternate the male and female sides on these buttons? Feels like a design flaw only sleep-deprived parents will understand. #parenting

RT @TansuYegen: Dyson is now growing strawberries with robots, spinning trays and UV lights, making farming super smart and fun 🍓 https://t….

RT @caleb_friesen2: ASMR Hinglish TTS AI models weren't on my 2025 bingo card

It was able to fill in the city, state & zip that was missing. But the zipcode was incorrect - although pretty close to the area.

Digging into Mistral OCR. It doesn't just OCR, it also fills in the blanks. Just thinking it "OCR" can create very misleading. If LLMs' world knowledge hints at model size, it's probably using a medium- large model, similar to 4.1 or 4.1-mini. Bonus: it's good at math (see

Tu Bier hai

TIL Mistral OCR is really good at OCR, and can do LLM stuff. but doesn't accept a prompts - just a schema. As a workaround, to 1-shot extraction, reasoning & enriching, use ZOD's `describe`.

CleanShot's Designers 🧑‍🍳🤌. Highlighter automatically changes size based on the font-size in your image. #LittleBigDetails

What's interesting is that it was able to:. • get the right currency.• correctly determine credit or debit.• determine individual transactions with date, even though the date only exits for the first row.

Schema: