San Francisco has a unique feature that no other city in the world has. And if you want to build an outlier career, this matters a lot. And I'm not talking about the concentration of tech companies. San Francisco is the one place in the world that you'll be taken seriously no matter who you are. It has an unusually low barrier to being taken seriously. In most places, if you're 20 years old, fresh out of school, and building something, people are naturally skeptical. In SF, that person in flip-flops and a hoodie might be building the next billion-dollar AI company. And that changes the status quo of how people treat each other. There's a culture of radical open-mindedness toward ambition. People are more willing to take the meeting, hear the idea, make the introduction, or give someone young and unproven a shot. You see this everywhere. Most of the power players in SF are easily accessible – they do office hours, coffee shops, speak at events, etc. So your age, title, pedigree, and resume matter a little less. And what you're building matters a little more. I've traveled to many startup ecosystems, and I haven't seen this norm as strongly anywhere else. If you're young, ambitious, or trying to do something that sounds a little crazy, that's one of San Francisco's biggest advantages.

Your data belongs to you! We’ll deploy Replit to your cloud and data never leaves home. Designed for high stakes environments like governments, finance, and healthcare.

Coding agents were supposed to save us time. Instead we somehow became middle management for robots. Codex is doing one thing. Claude Code is doing another. Copilot has found something useful. And you’re the person carrying messages between them.

ChatGPT gave a man unlicensed medical advice, leading to mental health crisis, lawsuit alleges. cbsn.ws/4dFwATB

OpenCode is currently under a sustained cyberattack, and mid-incident they ran into the same problem Hugging Face did: the frontier models they were using for defense began refusing to help fight attacking agents because analyzing the attacks started triggering safety guardrails.

We just open-sourced another pair of state-of-the-art Decision Models: pplx-decider-flash-v1.2: 4B params, 1 cent per million tokens (top 4B model on HF Decision Index) pplx-decider-v1.2: 27B params, 2 cents per million tokens (#2 on HF Decision Index overall - for now, will improve rapidly) Both are available on the Perplexity Decisions API as well as in the form of open weights on @huggingface Enjoy!

🆕 How to Kill Your SaaS youtu.be/zlD-CkLVKtU A lot of people talk about the death of SaaS but mostly in theoretical terms. We took a real $43,000/year SaaS we use and launched a global hackathon to clone it in a weekend, with both AI and REAL HUMAN evals, using any models any harness. 72 submissions, a lot of manual evals and ~100B tokens later, here are the results In this special @latentspacepod edition we share the data collected, what's still hard about killing nontrivial, midtier SaaS, and reveal the winners of the $10,000 Kill My SaaS contest with qualitative interviews!

Anthropic just reported $16.8B of revenue in a single quarter. Two quarters ago it was $4.7B. That one number answers the question the whole market is panicking about this week. Here's the math, step by step. 1. The fear Big Tech is on track to spend ~$1.2T on AI infrastructure next year. More and more of it is borrowed. The bear case is simple: trillions go into chips, buildings and power, and the revenue never shows up. That's why AI infrastructure stocks just got crushed. 2. The only question that matters 
How much revenue does one unit of compute produce? The unit is the gigawatt. One GW of AI compute is a campus with hundreds of thousands of chips, drawing the power of roughly 800,000 homes. Goldman Sachs did the math on everything that goes into it: chips, buildings, networking, power. For the buildout to earn a solid 15% return, each GW needs about $11.6B of revenue a year. 3. What Anthropic actually earns 
Anthropic ran about 1.4 GW at the end of 2025 and plans ~5 GW by the end of 2026, so it averaged roughly 3.5 GW last quarter. $16.8B a quarter is $67B a year. Divide by 3.5 GW and you get ~$19B per GW per year. That's 1.6x Goldman's bar. Where the demand is real, the capex is already profitable today. Not in 2030. Today. 4. Proof the demand is real 
Anthropic spent this year rationing its own customers. It cut usage limits at peak hours because it ran out of compute. Now it's buying capacity from Amazon, Google, Microsoft, SpaceX, Nscale, Lambda and IREN all at once. ASML says the machines that make the chips are sold out through 2028. When the seller is rationing, the problem is supply. 5. The loop 
Revenue buys compute. 
Each new GW starts earning ~$19B a year the day it's plugged in. 
That revenue buys the next GW. That's why the debt doesn't scare me. Debt backed by $19B per GW of revenue is a business. Debt backed by hope is a bubble. 6. The line that ends the panic 
For the whole industry, the buildout breaks even at ~$300B a year of AI revenue. Anthropic alone: $67B.
OpenAI: ~$50B.
Microsoft's AI business: $37B+.
Then Google, Amazon, Meta and thousands of apps. At this pace the sector crosses $300B in 2027. Nobody panics about borrowing to build something that already pays for itself. The financing fear dies that day. 7. What's left after that
Once money stops being the constraint, the bottleneck is physical: chips, memory, and above all power, which takes years to add to the grid. Money is solving itself. Megawatts are not. Bubbles burst when the revenue never comes. This one arrived early, and it's being rationed. $BE

As I stated on the pod, it’s more critical for Ant to get public and get the $200B they need than to optimize for valuation. Find the market clearing price and go public asap. My guess, a few weeks ago, was that this price was +/- $1T. I’m still of this mindset and haven’t changed my mind.

Imagine an executive using the word investment grade in a conversation the same day his companies yields rise 225 bp in 2 days to 15.8% YTM Having $13.5 billion due in principle and interest by next year on $10 billion revenue and an operating loss while all your share sale money is tied up to construction obligations. Having no access to bond markets Having executives/founders who sold $11 billion of stock in a year (30% of the company) That’s $CRWV

So these guys were basically on the money again. Market now has the "all clear" signal to crash the AI trade. AI math went from virtually impossible to impossible.