TrendPulse
🇬🇧 English 🇨🇳 中文
🔖 Saved RSS
HackerNews

Tyler Cowen: A Dangerous Turn in AI Regulation

Article URL: https://www.thefp.com/p/tyler-cowen-a-dangerous-turn-in-ai Comments URL: https://news.ycombinator.com/item?id=48560807 Points: 1 # Comments: 0
AI
HackerNews

We built an agent that runs our AI data platform

Article URL: https://encord.com/blog/merlin-encord-mcp-agentic-intelligence/ Comments URL: https://news.ycombinator.com/item?id=48560889 Points: 2 # Comments: 0
AI
Most attempts to reverse-engineer Fable 5 are missing the point
Reddit

Most attempts to reverse-engineer Fable 5 are missing the point

A lot of people are trying to reverse-engineer Fable 5 right now. Wrappers. Prompt packs. “Long-horizon agent” scaffolds. Tools that try to look like Fable from the outside. I think most of this is pointed in the wrong direction. If Fable 5 were just a prompt pattern or a wrapper, it would already be cloned. The real problem is not appearance. The real problem is robustness. Most coding agents look good at the start. Then the cracks show. - scope starts drifting - public tests b...
Artificial
Reddit

The beautiful ugly shape

The whole system has a clean architecture: X supplies the public behavioral graph. Premium/Premium+ supplies a paid, high-signal user cohort. Grok supplies the private conversational layer. Image/video generation supplies visual preference data. Voice/personality/companion features lower inhibition and increase intimacy. Ads/subscriptions/enterprise/government/API monetize the resulting platform. The S-1 (SpaceX's IPO filing) calls X a distribution and data engine and describes A...
Artificial
Reddit

board prep used to eat a full saturday for me, the gathering more than the writing

counted it once last quarter: roughly 3 hours just pulling inputs before i opened a single slide. Last month's metrics out of notion, the roadmap state from linear, founder check-in notes sitting in granola, the open investor threads in gmail. None of it is hard work, it's just scattered across five tabs and i'm the courier carrying it between them. The actual writing was never my bottleneck. a chat assistant drafts the narrative fine once everything is pasted in, but i'm still the one doing t...
Artificial
Reddit

I have 3,000 photos and videos in OneDrive. How can I organise them with AI?

Looking for a bit of advice because I feel like I’m missing something obvious. Over the last few weeks I’ve finally consolidated my photo library and got everything into OneDrive. I’ve now got two folders: Photos Videos Between them there’s around 3000 files in total. The files go back years and are a mix of family photos, holidays, screenshots, random phone pictures etc. I’ve been trying to use AI to help me organise everything properly. Things like: - Finding duplicates and near-du...
Artificial
Reddit

Would you pay for an independent alert service that tells you when an LLM's behaviour has drifted - before your users notice?

Following up on a thread I posted yesterday about how developers detect LLM API degradation. The responses were useful enough that I want to validate a specific idea. It is a 3 layer independent alert service: Layer 1: Transport health alerts: Independent probes checking TTFT, error rates, and latency across major models (Claude Sonnet, GPT-4o, Gemini, Grok) every 5 minutes. Alerts you before the provider's status page updates. This part already exists and is free at tickerr - the question...
Artificial
Reddit

What happens when frontier LLMs are deployed in rural Rwanda? Lessons on usefulness, language gaps, and incorrect answers [D]

At GiveDirectly, we recently ran a pilot in rural Rwanda that paired unconditional cash transfers with access to a general-purpose AI chatbot. One of the most interesting findings: people often used the chatbot as an always-available advisor—for business decisions, learning, and getting second opinions. But the pilot also exposed important limitations, including language gaps, locally irrelevant responses, and confidently incorrect answers. The writeup explores both sides: where participants...
Artificial
Reddit

Built a Paninian Retrieval-Augmented Generation (PRAG) framework for safer medical AI — seeking feedback

Hi everyone, ​ I'm an independent AI/ML researcher and I've been working on a project called PRAG (Paninian Retrieval-Augmented Generation) for safety-critical medical AI. ​ The idea is to combine traditional RAG with a Paninian rule engine inspired by concepts such as Utsarga-Apavada, Paribhasha, Nitya-Anitya, and Antaranga-Bahiranga. The goal is not just better retrieval, but safer medical reasoning with full auditable rule traces. ​ Current findings: • 71% r...
Artificial
HackerNews

South Korea Politicians Wants AI Profit Sharing

Article URL: https://www.gadgetreview.com/south-korea-floats-citizen-dividend-ai-profit-sharing-as-tech-stocks-tank Comments URL: https://news.ycombinator.com/item?id=48552626 Points: 2 # Comments: 1
AI
HackerNews

Show HN: A prompt generator to help AI agents implement my new email API

I love building apps with Firebase but felt that, other than basic auth, app emails (notifications, billing etc) were missing. Wanting to implement a solution to this, it got me thinking about how to tell a coding agent about a new api and how to implement it. I dynamically generate an openapi spec via a script and have been playing around with using this spec to then create a prompt generator. It's not something I've come across before and I can sort of understand why... Generating a prompt t...
AI