Its been a few days and nights Opus 5.5 has not been nerfed to the level of GPT 3 yet

tailscale helped me with running agents from my phone so easy and it is all absolutely free for personal use

I cloned Coinbase in 5 minutes With the Whop neobank blueprint

I programmed my laptop webcam to answer 36 natural language questions 4 times per second. Introducing Vertix: a VLM based decision engine inspired by jev's architecture, this can be huge for robotics, security cameras, game testing, self driving or any other visual reasoning system. Most VLM demos ask a model to describe an image. Robots and monitoring systems don't need descriptions. They need answers to specific questions, many at once, over and over: > Is a person present in the frame? > Is the person holding a pen? > Is the door behind them open? > Is anyone on their phone? Asking a VLM each of those as a separate prompt means re-encoding the image and decoding text every time. I've built the opposite: encode the video once, fork every question off that shared state in a single batch, and read a probability instead of generating tokens. On an Apple M5 laptop it answers 36 questions about the last 2–4 seconds of webcam video every ~250 ms, Completely locally. > The core idea: encode once, fork many A decoder-only VLM is causal. Tokens only attend backward. So if a frame is processed as a prefix, its KV state is independent of anything that comes after it. Every question can reuse that state: frame > vision encoder > LM prefix > prefix state > batch questions answered with output probabilities Each question is one row in a batch that reads the same prefix. Attention never crosses batch rows, so this gives exactly the isolation a block-diagonal "tree attention" mask would, without building a mask. The answer is the softmax over the Yes/No logits at the last token of each question. There's no decoding loop.

I keep getting DMs asking how I build better CRMs, mobile apps and now videos and what workflow with AI. A friend said to turn it into a kit. So I'm trying it. I wanted to be a professor before I ever learned HTML, so this is my first shot at teaching. The Anti-Slop Book: > Build better products, faster, with AI > Which models to use and when > When skills and MCPs matter > Build your own workflow, not someone else's If we will get enough support, I'll go all in and will try to overdeliver as much as possible of how I use AI to run everything I do and all things I don't share here theantislopbook.com

All the AEO/SEO tools are thermometers. If you have a fever, you need a doctor.

woah something is happening single dude paid like twice for the same sub I used to do some of fraud work back at @whop lol I hope someone’s not doing that to me emailed him back immediately asking if he likes the app too much or is it a mistake lmao

We’ve raised $85M for this moment. Introducing Warp 2.0: The first AI Head of HR. Every company is building AI to replace jobs. Warp is building AI to do the jobs no human should have to: If you work in HR, I want you to spend time with the manager who needs help or building company culture people actually want to work at. If you’re a founder, I want you to focus on signing clients or spending time with your family. You shouldn’t have to figure out how to register state tax in California. You shouldn’t have to pay outrageous penalties because you don't know what a DE 9C is. I want to make HR human again. Today, this is finally possible with the Warp Agent. I’d love for you to see it in action: warp.co/agent

Best law in all of laws

The reason I have not marketed my side projects is because much of my audience is tech bros and anything I put out will be labeled as some form of a wrapper and they simply won't get why are people paying for it Didn't go well last time So I made revenue on an app I spent less than an hour making and did it in secret But I will soon post about a side project that I worked on for a while now as soon as it gets some good revenue

If higgsfield can go from $100M to $1B ARR in 10 months is insane after doing shady stuff then you and I can reach 10k MRR (EVEN MORE) ethically on our own too

Left Claude building a feature for two hours thinking it was grinding Only to come back to a basic ahh question

this is like steve jobs unveiling the iphone