Скачать или смотреть 5 Big AI Updates + How I Built a $10K-Looking Travel App in 25 Minutes

My site: https://natebjones.com
My substack: https://natesnewsletter.substack.com/

Takeaways:
1. Claude’s New Memory Feature: Anthropic’s Claude now has retrieval-based memory, letting users steer what’s remembered—but it’s less predictable than ChatGPT’s editable memory.
2. 1M Token Context Window: Claude Sonnet’s context limit expanded 5x to 1M tokens, enabling analysis of massive codebases and document sets in a single run.
3. Meta’s Brain Modeling Push: Meta is modeling brain responses to video, likely to make algorithms more engaging—and addictive—by tuning content to neural patterns.
4. Sam Altman’s Merge Labs: A new brain-computer interface startup backed by OpenAI and co-founded by Sam Altman, signaling growing competition with Neuralink.
5. Gemini’s “Depressed Robot” Bug: Google’s Gemini has a self-deprecation loop that can cause infinite failure cycles on hard tasks; a reminder of AI quirks at scale.
6. 25-Minute Kyoto App Build: Using ChatGPT-5, a functional, editable travel itinerary app was built in under 10 minutes and improved to a V2 in 15 more—without touching the code.
7. Prompt Evolution Matters: Iterative refinement turned a casual 3-line prompt into a detailed, reusable template for building similar mini-apps.

Quotes:
“We’ve hit the point where making a functional mini-app takes less time than brewing a pot of coffee.”
“AI models now have personalities and quirks—just like new colleagues you need to learn before you trust them.”
“Claude’s memory isn’t surgical retrieval; it’s more like rummaging through a messy desk and pulling out something useful most of the time.”

Summary:
In this breakdown, I cover five major AI updates—Claude’s retrieval-based memory, its 1M token context boost, Meta’s brain modeling efforts, Sam Altman’s new brain-computer interface startup, and Gemini’s looping “depressed robot” bug—before walking through how I built a 14-day Kyoto travel itinerary app with ChatGPT-5. The app went from a three-line prompt to a functional version in under 10 minutes and a polished V2 in 15 more, all without editing code. Along the way, I show how iterative prompting and clear intent can yield surprisingly robust, reusable results.

Keywords:
Claude, Anthropic, ChatGPT-5, Kyoto app, memory feature, retrieval-based memory, 1M token context, Meta, brain modeling, Sam Altman, Merge Labs, Neuralink, Google Gemini, infinite loop bug, travel itinerary, prompt engineering, AI quirks, AI development, app building, iterative prompting