Really painful experince using @Escrow_com 99% of the Escrow process = waiting & chasing unclear info. 1% = actually handing over and accepting the asset (1h). Feels like they’re optimizing for holding the funds for as long as possible, not for user experience.

Remember when a new JS framework dropped every week? Now it’s new AI models instead. Not gonna miss it, I prefer this kind of chaos.

Very useful tool — it should be part of everyone's workflow especially if they're worried about security.

now that O3 is so smart use it to fix your model‑name jungle

Vibe coding could be a significant shift It allows non-devs to easily build prototypes and showcase ideas Freeing developers to focus on architecture, performance, and maintenance

Source:

#OpenAI o3 hitting 100% accuracy at 120k context is insane

Context size is easy to measure. Context understanding isn’t — but that’s what actually matters, especially in code generation. Source: Fiction.LiveBench https://t.co/Y52ZQcBuYb

🧠 Top models for long-context understanding: o1: one of the most accurate models overall, especially in the mid-context range Claude 3.7 Thinking: strong up to 8K Gemini 2.5 Pro: the only model that holds up past 60K Everything else starts hallucinating.

Most models can handle large context windows. Few can reason over them accurately. Context size ≠ context understanding.

Large LLM context windows (100k+) aren’t always better. 📉 After ~16K tokens (~600 lines of code), most models start missing up to 20% of the code context. Even the “MAX” context models aren’t reliable for deep reasoning past that. 🧵

Gemini Deep Research is underrated and very cheap - crawled 200+ sites in my last query - clear citations - 1-click audio summary - copy over to NotebookLM to get better insights But please fix the iOS app @sundarpichai, still feels beta and dosen't support deep research