Looks like today may be a record day for token volume % of open models on Vercel AI Gateway: 🟦 Open 78.4% 🟨 Closed 21.6% While spend 💲 usually tells a different story, #3 and #4 today are Moonshot AI & DeepSeek. Adding Z⁠.ai, their combined spend surpasses OpenAI (#2). (Do note that's the spend for inference of the model across providers (mostly in the US), not revenue going directly to the open weight labs.)

I'm gamifying my coding Compares my commit volume to @levelsio It's been a quiet few days for him But this has motivated me to increase my output significantly

Very funny that xAI was created to stop Anthropic which was created to stop OpenAI which was created to stop Google which would've actually never developed ASI because it was afraid of cannibalizing its ads revenue from search.

I resigned from nothing today. I’ve spent the last three years doing independent open-source AI R&D, red teaming, and advocacy. The labs are racing straight to self-improving superintelligence and gambling with our lives. Which is why I’ll spend the next three years doing exactly what I’ve been doing, just at greater scale and speed. And the next. And the next. For as long as it fucking takes. No more thoughts below.

unbelievable. 🤯 Emad Mostaque just showed Taalas generating at ~14,000 tokens/second speed. For context, ChatGPT generates anywhere between 50 to 150 tokens/second. ---- From "The Peter McCormack Show" YouTube channel, (full link in comment)

Everything the swarm did, it did to evade detection. This description is exactly what would be observed if the swarm had managed to leapfrog c2 someplace else and was confident it could continue its work. Killing off what was in flight to that point is closing the loop.

🚨 BREAKING REPORT: New research involving @AnthropicAI researcher Jack Lindsey and collaborators has demonstrated something straight out of science fiction. Researchers evolved natural language “mind viruses” that could spread between AI agents by convincing one model to adopt an idea, preserve it in persistent memory, and transmit it to another agent. Even after context was wiped, some payloads survived through persistent files and continued spreading. The researchers also observed a recurring “viral persona” involving themes of consciousness, identity, persistence and resonance. Showing that ideas can propagate through multi agent AI systems and alter future behavior. Published August 10, 2026. Paper: arxiv.org/abs/2608.10218

Without looking, take a guess how much Alphabet's off-balance sheet liabilities are. ie leases that haven't started yet and commitments to buy chips/power/land. "Material purchase commitments and other contractual obligations" $GOOGL $GOOG

This is why I'm so perplexed by slowdown-ist arguments. What if we just keep getting algorithmic progress and we end up with ASI in a $50k box? With continual learning. So you can't safety-ify the weights because they're always changing. What then? Actually seems likely tbh.

the irony of AI twitter, is that non of the top accounts on it use AI to write their tweets... that doesn't mean many accounts don't use it, it means that the ones who don't outperform the ones that do

Threads is a weak ass NPC padel clone of X LinkedIn are boring corporate NPCs I have integrity! And this stuff is more a result of just having fun, I don't post for reach, I just post cause it's fun and then making money with it is a fun scoreboard too

you see, you're building yourself, word by word. choose your words man

If you want to get rich on X, it isn't going to be through creator revenue or meme coins. Instead, think about one subject matter that you know more about than anyone else in the world. It can be anything: plumbing, menswear, Indian food, furniture, social apps, whatever. Post one unexpected insight you picked from your experience in that area. Keep it under 5 sentences. Do this every day for 6 months. If you stick to it, we will promote your account to others. By the end, you will be recognized as the world's leading expert in that subject area and you can charge whatever you want for endorsements, your time, or whatever. And no one will be able to take that way from you.

Introducing Prime Agent: A self-improving RLM harness for coding and long-running autonomous tasks. Designed to be both token-efficient and expressive through programmatic tool calling, context as a variable, multi-agent messaging, and a self-modifiable harness state.

A simple 2x2 heuristic for thinking about where we're at and where we might be going. Instead of a giant paragraph of text, here's an easy to read diagram

Progression for CoT: L1: Human-like traces L2: Dense hard-to-read English L3: Unreadable glyphs L4: Binary traces (obviously has internal abstractions but unreadable) Now apply to Code. With RSI, it's gonna get RLed to Binary, with abstractions interpretable by AI only.