NOSIBLE AI

Startups are hard, I need your help! 🙏🏻 This year we found our niche: large firms who want to monitor the world at unprecedented scale. E.g. One customer does 2Mn searches a day looking for cyber-attacks, lawsuits, product recalls, and other material events over thousands of entities. We can charge ~$0.20 CPM because NOSIBLE was built for this. We’re 100% B2B and on a path to millions in ARR ... but our website looks B2C. It's bad! So, I am looking for an individual or agency who can design a world-class site that tells our story perfectly. We have the budget and want to go all in. For this project we can provide: ✏️ Site copy directions 📕 Brand bible (visuals + tone) 🎨 Access to our illustrator 🏗️ Customer case studies 💻 Code examples & tutorials 👋🏻 Lots of time 1:1 with me Ideally, we want to build in @framer. If you know someone great, please connect us. If not, a retweet would help. We want to find the best of the best.

Web search is holding back AI 👇🏻 Web search costs two orders of magnitude too much and is holding back large-scale AI adoption in enterprises. Here's a story that illustrates this. An agentic risk analyst agent does 100 searches a day to find lawsuits, cyber-attacks, recalls, strikes, and other material events for 25,000 companies. A dataset of 10 million results with 500 tokens each would cost around $5,000 to process with grok-4-fast. It would also cost <$5 a day to maintain. BUT ... how much would those web searches cost? Well at $5 CPM it would be $12,500/day ($5 ÷ 1000 x 100 x 25,000). Yes. $12,500/DAY! And that's only 25,000 companies. Imagine you're a big US bank tracking ~3 million business customers. You'd be spending a million dollars a day. It's obvious that web search is the optimal form factor for AI tooling. It's also obvious that $5CPM for search looks utterly ridiculous next to token costs. This story is not made up. This is exactly what I tried to build years ago. I honestly didn't think I would need to build a search engine first ... but here we are. On Oct 29 at @SHACK15sf I'll present a talk "The End of Blind Spots: Agentic Search as an Enterprise Risk Model" at the BrightData Web Discovery Summit. So, if you're in SF, and you feel like geeking out over web search you should definitely try attend. @RichardSocher and @paraga are also speaking!

Thought I'd share this.

Cybernaut-1 is here 🥁 Cybernaut-1 combines our hybrid-3 search algorithm with LLM-guided Monte Carlo Tree Search to deliver world class search results on difficult queries!

Finished an internal paper about compression, random walks, and predictive models. It shows, formally, how we used them to find a novel solution to ANN search. The paper reminds me of my Turing Finance blogs. I feel like I bonded with my past self --> turingfinance.com/stock-mark…

The vast majority of searches in the future will be conducted by AI, not people. My contention is that a search engine built for people and optimized using ad loops cannot be ideal for AI. So, we are scaling up a new search engine built exclusively for AI. 🧵

We landed in the best city in the world today - @NosibleAI 💚 NY

I have wanted to share this story for a very, very long time. Here is the story of how two quants - in South Africa of all places - found a solution to vector search that scales at 1/100th the cost of HNSW with pareto-optimal trade-offs and unique capabilities 🧵👇🏻