HackerNews

Enterprise AI Agents Are Leaving the Server

Article URL: https://focused.io/lab/enterprise-ai-agents-are-leaving-the-server Comments URL: https://news.ycombinator.com/item?id=48536050 Points: 1 # Comments: 0
HackerNews

Nvidia Is the Federal Reserve of AI

Article URL: https://ramansharma.substack.com/p/nvidia-is-the-federal-reserve-of Comments URL: https://news.ycombinator.com/item?id=48536189 Points: 2 # Comments: 0
Is Fable 5 Back? — Live Tracker
Reddit

Is Fable 5 Back? — Live Tracker

Got tired of refreshing the anthropic news page. Made this app with Opus that pings Claude's api with the fable model id until it works, then updates the page. People seem to be liking it, thought i'd share it here as well
Reddit

Idea: a local AI that watches your smart home devices and tells you in plain English if something's off — does this solve a real problem or does it already exist?

Been thinking about building this and want a sanity check before I commit time to it. The problem: Most homes now have a bunch of cheap IoT devices (smart plugs, cameras, TVs, thermostats) that constantly talk to the internet, and basically nobody — including most security tools — knows what's "normal" for any given device. Existing network monitors (Fing, GlassWire, router apps) show you raw traffic, but they don't tell you whether a behavior is expected for that specific device ...
Reddit

How should people share agent-security tests without making it vendor spam?

I’m asking because this topic gets messy fast. Prompt injection is more interesting once the model can use tools, but most posts end up as either scary headlines or someone sneaking in a product pitch. What would be a useful format here? My gut says small reproducible examples, clear limits, no “we solved it” claims, and enough detail that people can argue with the result.
Reddit

I gave Google AI a simple test and it gave me the wrong answer 3 times in a row in different browsers even though it said it would record the correct answer and remember it for future results.

I asked it something very simple: slimmest laptop ever Answer it gave: HP Spectre 13 at 10.4mm Correct answer: Acer Swift 7 at 8.98mm it's not a trick question, both are traditional clamshell laptops with keyboards. It just kept failing to learn from it's wrong answers. That's very concerning, because even when it admits when it is wrong, it still doubles down and continues to give the wrong answer to future questions. [link] [c...
HackerNews

The University in the AI Era

Article URL: https://htmx.org/essays/universities-and-ai/ Comments URL: https://news.ycombinator.com/item?id=48532306 Points: 1 # Comments: 0
HackerNews

Show HN: AI Cell Enrichment Workflow API

Hi HackerNews! I've recently added an API + Docs to ampledata - my AI enrichment platform. Idea is simple. Given a set of rows in a table you can hit ampledata via an api and a workflow will run to enrich the table with columns of your choosing by searching the web and giving you citations + exact sources. You can think of it as a cited bulk enrichment platform. Ampledata is also OSS so feel free to self host it and use your own api keys. Or if you are just interested in the tech. Comments U...
HackerNews

AI and the Red Queen

Article URL: https://huntersoftwareconsulting.com/posts/2026-06-14-ai-red-queen/ Comments URL: https://news.ycombinator.com/item?id=48532967 Points: 2 # Comments: 0
HackerNews

AI slowly sucking the joy out of work

Article URL: https://www.reddit.com/r/swift/s/YA5ELxU3av Comments URL: https://news.ycombinator.com/item?id=48533359 Points: 6 # Comments: 0
HackerNews

Terminal UIs Are an Abomination. AI Needs Better UX

Article URL: https://medium.com/@balajibal/terminal-uis-are-an-abomination-so-are-chatbots-ai-needs-better-ux-c7ee7b5a0018 Comments URL: https://news.ycombinator.com/item?id=48534128 Points: 2 # Comments: 0
Reddit

We’re building an AI factory

I’m one of the people building Since AI( https://sinceai.ai/ ). The idea is simple: bring together serious AI builders, give them real industry problems, compute and 72 hours — then help the strongest projects continue after the event. Less networking theatre. More working software. What you think would make this genuinely valuable rather than just another hackathon?