AI News HubLIVE

Today's must-reads

Policy

Two AI Futures to Choose From

The article contrasts two visions for a safe AI future: a Singleton with a single ultra-powerful AI, versus a Plural world with democratized access to many AIs. The author argues that history favors plural societies, which avoid the over-concentration of power while embracing the messy but resilient nature of distributed systems.

  • Singleton safety comes from perfect design; Plural safety from distributed checks and balances.
  • Historical evidence shows plural societies are more resilient and beneficial.
In-site article
Chips

Kimi K3 is not cheap

Contrary to popular claims, the open-weights LLM Kimi K3 from Moonshot AI is not cheap. While its performance is close to top US models, its cost per task is comparable to OpenAI's top model and significantly higher than other Chinese models like DeepSeek V4.

  • Kimi K3 is a new open-weights LLM from Chinese lab Moonshot AI, sparking debate over its cost.
  • Commentators mistakenly claim K3 is cheap; in reality, its cost per task is similar to top US models.
In-site article
Models

An Inside Look at the Relay Market Powering Token Resellers and Fraud

Matt Lenhard's investigation reveals a market where LLM tokens are resold at a discount by abusing free trials, unprotected support bots, stolen credit cards, and chargeback attacks. Primarily in China, resellers use open-source proxy software like one-api and new-api to pool API keys. Buyers seek cheap tokens, avoid geo-restrictions, or collect data for model distillation. The author calls for stricter API key caps from vendors.

  • Resellers offer discounted LLM API access by abusing free trials, unprotected support bots, or stolen credit cards and chargeback attacks.
  • Open-source proxy software one-api and its fork new-api are used to pool API keys.
In-site article

Black Forest Labs Releases FLUX 3: A Multimodal Flow Model for Image, Video, Audio and Robot Action Prediction

Black Forest Labs (BFL) releases FLUX 3, a multimodal foundation model that learns from images, videos, and audio within a single architecture. It is the first FLUX model to output video, audio, and action predictions from one set of weights. The model builds on the Self-Flow method and excels in video generation, producing clips up to 20 seconds with native audio. In human preference tests, FLUX 3 outperforms many competitors. The same backbone also drives the FLUX-mimic robot policy with sub-80 ms latency.

  • FLUX 3 is a multimodal foundation model unifying images, videos, and audio. It can generate up to 20-second video clips with native audio, leading in human preference evaluations.
  • Training uses the Self-Flow method; video consumes over 95% of compute, audio less than 0.5% of tokens.
In-site article
Agents

Please ship APIs, not AI

The author urges companies to prioritize exposing core functionality through well-designed APIs rather than building product-specific AI chat bars or assistants. APIs provide better tools for AI agents and enable more efficient automation, even if it means becoming infrastructure providers.

  • Companies should design easily accessible, fine-grained permission APIs.
  • APIs need token efficiency, human-in-the-loop hooks, and good documentation.
In-site article

Show HN: HART OS – an open-source AI OS built so frontier AI needs no datacenter

HART OS is an open-source AI-native operating system that runs locally on any device, from laptops to robots, providing AI inference as a system service. It features self-improvement, federation across peers, and an OpenAI-compatible API, eliminating the need for datacenters.

  • HART OS is an AI-native OS that runs on any device, local-first.
  • It provides AI inference as a system service via Model Bus, so apps don't need their own models or API keys.
In-site article

"Grill Me" Skill

The grill-me skill relentlessly interviews you about a plan or design, asking one question at a time with a recommended answer. Matt Pocock now recommends domain-model over grill-me for most planning workflows, but grill-me remains useful as a narrower pressure-test. Use it before writing a PRD, implementing a feature, or committing to a data model. The recommended workflow starts with domain-model, then moves to to-prd, to-issues, and tdd.

  • grill-me is a skill that conducts a relentless interview about a plan or design, resolving the decision tree.
  • Matt now recommends domain-model as the default planning skill, with grill-me as a lighter option for specific situations.
In-site article

Show HN: ModelFuzz – Open-source runtime guardrails for AI agents

ModelFuzz is an open-source tool that helps detect and prevent prompt injection attacks on AI agents, using a red-team scanner and a protective decorator to secure infrastructure and sensitive data.

  • Prompt injection can hijack LLM agents to execute arbitrary commands or exfiltrate data.
  • ModelFuzz provides a scanner to red-team any OpenAI-compatible endpoint and find vulnerabilities.
In-site article
Tools

Illume Labs – Your 24/7 personalized health companion

Illume Labs offers round-the-clock personalized health tracking to help users optimize their longevity strategies using AI.

  • Illume Labs launches a 24/7 health companion
  • Utilizes AI for personalized health monitoring
In-site article
Research

What Americans Think About the Global AI Race

A Pew Research Center survey finds that 36% of Americans believe China is more advanced in AI than the U.S., while only 12% think the U.S. leads. 43% say it is extremely or very important for the U.S. to be the world leader in AI, and 51% expect AI to increase the gap between rich and poor countries.

  • 36% of Americans think China leads in AI; 12% say U.S.; 33% are unsure
  • 43% say U.S. global AI leadership is extremely/very important; 22% say not too/not at all
In-site article
Other updates (5)
Agents

SP/1.0: Deterministic, Reproducible Verdicts for AI-Agent Decisions

SHACKLE is a runtime circuit breaker protocol for autonomous AI agents, featuring 9 mathematical invariants, Ed25519-signed audit logs, and a Redis-backed distributed state engine for deterministic and verifiable decisions.

  • SHACKLE is the first open-source runtime circuit breaker for AI agents with cryptographic audit chain-of-custody
  • The decision function is a pure function with 8 layers, auditable in under 10 minutes
In-site article

Show HN: Boffin – Staff-engineer layer for AI coding agents

Boffin provides AI coding agents with architectural constraints relevant to the current edit and forces verification, preventing unnecessary rewrites. It routes only the rules for the exact file being edited, supports multiple agent platforms, and has public case studies showing minimal changes with all tests passing.

  • Boffin routes architectural constraints relevant to the current edit.
  • Requires verification proportional to the change.
In-site article

MCP’s biggest update removes the machinery many servers were built around

The Model Context Protocol (MCP) receives its largest update since launch, removing session state and initialization handshake to simplify remote server operations. The release candidate is frozen, final spec due July 28. Deprecations include core features like Sampling, with migration directions provided.

  • MCP's update eliminates session affinity by making requests stateless, reducing operational complexity.
  • Capabilities and protocol version are now carried per-call via _meta, enabling caching and routing.
In-site article
Policy

AI Use Policy for My Essay Writing

A personal policy on using AI for essay writing, focusing on transparency and ethical guidelines.

  • Outlines a clear policy for personal AI use in writing
  • Emphasizes maintaining originality and academic integrity
In-site article

This July I Was Fired from Simple AI (A Deeply YC Company)

Andy Trattner recounts his abrupt firing from Simple AI after three weeks, yet maintains a positive view of the company and founders, and seeks new opportunities.

  • Andy Trattner was fired three weeks into his role at Simple AI, with Slack access deleted mid-work on a Tuesday at 6pm.
  • He was recruited by founders Cat and Zach without applying for any position.