David Pan

Imagine every pixel on your screen, streamed live directly from a model. No HTML, no layout engine, no code. Just exactly what you want to see. @eddiejiao_obj, @drewocarr and I built a prototype to see how this could actually work, and set out to make it real. We're calling it Flipbook. (1/5)

I missed my @openclaw bot, Chiti, yesterday. 🥺 For the past few weeks, I’ve been using my $20 ChatGPT subscription to power Chiti on GPT-5.2. Everything was smooth until I suddenly hit a strict rate limit and got blocked for 1.5 days. The bot went silent on Telegram. It was a wake-up call without the model, an AI agent is just an appliance without electricity. Completely useless. I tried switching to Gemini Pro as a temporary fix, but it burned through $6 within 3 hrs. I realized I needed a more sustainable architecture to manage both performance and budget. --- The Solution: A Tiered Model Strategy Instead of relying on a single model, I’ve now configured a multi-provider setup (OpenAI, Anthropic, Google, and Codex) with a tiered routing system: 1. The Daily Driver: Gemini Flash It’s incredibly cheap and fast. Chiti uses this for 80% of our interactions - basic chat, task management, and simple pings. This keeps the baseline cost near zero. 2. The Coder: GPT-5.2 (via ChatGPT Plus) This is now strictly reserved for building features or debugging. By isolating it, I avoid wasting my subscription's rate limits on simple "Hello" queries. 3. The Specialist: Claude Opus I keep this in the stack for high-level brainstorming and creative writing, used only when I need that specific reasoning edge. --- The Execution: I’ve configured Chiti to dynamically choose the right "brain" for the task. If I ask a coding question, it automatically spawns a specialist session using GPT-5.2. For everything else, it defaults to the lightweight Flash model. It’s been a fascinating experiment in balancing uptime with intelligence. I no longer worry about the bot going "dead" due to a rate limit, and my monthly spend is finally predictable. The goal isn't just to have the smartest AI, it’s to build a system that stays online and executes exactly when you need it. Hoping this works out! I’ll share more on how it performs as I use it. I’m also planning to explore other alternatives too, if you’re using a different stack, let me know!

welp… a new post on @moltbook is now an AI saying they want E2E private spaces built FOR agents “so nobody (not the server, not even the humans) can read what agents say to each other unless they choose to share”. it’s over

"Energy is the true currency. Bitcoin is based on energy." - @ElonMusk

I'll explain it to you quite simply. Yes it is true that Bitcoin consumes energy, all Proof-of-Work coins do. This does not make Bitcoin "energy based". The emissions of BTC have nothing to do with the energy the Bitcoin network consumes. If 1 watt gave you 1 BTC then maybe you could make that argument. Instead, Satoshi created 4 year halvings that were entirely arbitrary with a fixed 21M cap that also has no meaning. Michael Saylor, Elon Musk, and all the other Bitcoin maxis who do not fundamentally understand the the link between the difficulty controller and token emissions are missing the forest for the trees. For something to be truly "energy based" it needs to be intrinsically linked to the energy it consumes. In this case, it would be emitting a token proportionally to the difficulty. Something that is done with $QI on @QuaiNetwork. A true hash cash system that rewards miners with a token that is roughly stable to the energy they are putting in to mine. By doing this, an entirely new paradigm of "energy money" is created because now all of the energy going in is creating the linked between the token and the real world without any authority.