Hacker Newsroom AI for 24 May: AI Profitability, AI Cost Creep, Dont Paste AI, Deep Learning Bottlenecks

Hacker Newsroom AI for 24 May recaps 5 major AI Hacker News stories, moving through ai profitability, ai cost creep, dont paste ai, deep learning bottlenecks. 1. AI Profitability The next story is about a site called Is AI Profitable Yet? that tries to stack up estimated AI spending, revenue, and burn across companies like Google, Microsoft, OpenAI, Anthropic, and Nvidia, which matters because it turns the AI boom into a blunt scoreboard about whether all this capex is producing a real business. Hacker News immediately split between people who saw a clear gold-rush picture and people who argued the math is too rough, too blended, and too dependent on capital spending assumptions to settle the question. Story link Hacker News discussion 2. AI Cost Creep The next story is about a Fortune report arguing that heavy internal AI adoption can produce bigger bills than expected, citing Microsoft's reported pullback from direct Claude Code licenses, Uber burning through an AI coding budget early, and Gartner's warning that agentic workflows may drive token costs up even as per-token prices fall. Hacker News largely pushed back on the framing, with skepticism about the headline, doubts that Microsoft is cutting back for cost reasons alone, and a broader complaint that corporate AI mandates are turning token spend into a distorted management metric. Story link Hacker News discussion 3. Dont Paste AI The next story is about a tiny manifesto called Don't just paste the AI at me, where the author argues that if someone asks for your view, sending raw chatbot output misses the point because they wanted your judgment, context, and actual voice. Hacker News agreed with the basic complaint but turned the thread into a debate over tone, with some people cheering the backlash against lazy AI proxying and others saying the message becomes less useful if it is too angry to share with coworkers. Story link Hacker News discussion 4. Deep Learning Bottlenecks The next story is about Horace He's deep learning performance essay, which breaks optimization down into compute, memory bandwidth, and overhead, and argues that first-principles thinking can tell you whether to chase faster matmuls, fewer memory transfers, or less Python and framework overhead. Hacker News found the piece useful but got hung up on the examples, especially the dramatic comparison between Python throughput and an A100, which turned into a long argument about what exactly is being compared and where CPU, GPU, and framework bottlenecks really live. Story link Hacker News discussion 5. Models Dev Open Source Database The next story is about Models.dev, an open-source database of AI model specs, pricing, limits, and capabilities that is stored as community-contributed TOML, exposed as an API, and used by opencode, which matters because comparing models has become messy enough that people now want a shared source of truth. Hacker News liked the utility right away, but the enthusiasm came with a familiar warning that model catalogs get stale fast and need better filtering, benchmarking, and change tracking before they can become a dependable default. Story link Hacker News discussion That's it for today, I hope this is going to help you build some cool things.