Posts

In his recent post on X, Andrej Karpathy uses the funnel as an analogy for the AI understanding gap — that gap being largely one of scale in terms of population exposure to the models. The same analogy also helps illustrate the cost and benefit gaps, the inverted nature of which are colliding with popular acceptance of AI. As shown in the exhibits below, this same model can be applied to the benefit and cost gaps experienced by the same populations. ...

I called this phenomenon out just last week, somewhat coincidentally, when I decided to generate an Adobe Acrobat replacement using Claude. It’s now taking on a life of its own. I did not reverse engineer Acrobat. I took the approach of describing it, then letting Claude Opus 5.5 “do its thing.” Of course, not to be left behind, my colleagues decided to go next level and just ask Claude to turn the original machine code for the entire Adobe product suite into open source projects. ...

I’m probably not the only one annoyed at Adobe Software’s licensing model, with its ever-increasing subscription pricing, combined with the incessant cross-selling ads in the products, and intrusive update reminders. So I decided to replace Adobe Acrobat with my own internal product written using Claude Code and Opus 5.5. It is called “MyPDF.” Figure 1. MyPDF Screenshot. The total cost of this effort was less than $20, and it took about six hours of my time. It is checked into a private repository on Github, and I can deploy new releases, as needed, in a matter of minutes. I’ve uninstalled Adobe Acrobat from all my internal systems and made “MyPDF” the default for *.pdf files. ...

A few months ago, I wrote, “The recent explosion of social media frenzy over agentic orchestration tools such as OpenClaw and GasTown, vibe-coded and quickly released into the wild, is yet another symptom of overhyped expectations related to generative AI models. If it takes three weeks to go viral after a few months of vibe coding, it’s a trivial solution, and it will be copied relentlessly.” See Agentic Orchestration is not a Moat. ...

Do LLMs dream of Pi? It’s a whimsical question in the vein of a Philip K. Dick story, exploring the meaning of artificial intelligence and reality itself. The reality we find ourselves in decades after Dick first wondered about electric dreams, a future containing large language models (LLMs) and chatbots, might offer new answers, or perhaps just better questions. I found a few of my own recently as I recalled a time when I was able to recite Pi out to many digits, basically on a dare, a few years ago. ...

I recently ran into an interesting thermal issue while running two NVIDIA DGX Sparks as a small distributed inference cluster using llama.cpp. The two Sparks are directly connected over QSFP and use RDMA as the transport. With llama-server running on Spark1, and ggml-rpc-server running on Spark2, I’m using llama.cpp’s layer-split configuration with the DeepSeek-V4-Flash-0731-UD-Q4_K_XL model. The model is too large to run on a single Spark, thus the need to layer it over two Sparks. ...