Amazon is shredding rare books to train AI

AInegative -0.8#training-data #amazon #copyright #ethics #investigation

A reporter hid an AirTag in a shipment of rare books and watched it end inside an Amazon AI training facility, where the books get cut apart to be scanned.

What they showed / shipped

404 Media tracked a shipment of rare books with a hidden AirTag and found it terminated at an Amazon AI training facility.
The books are destructively scanned - spines cut off so pages feed through a scanner - meaning the physical copies do not survive the process (Ars Technica).
The irony is doing the heavy lifting in every writeup: Amazon started as a bookseller and is now destroying out-of-print texts for training corpus.

Why it matters

Builder lens: the scramble for high-quality non-web text is now physical and destructive, which tells you clean training data is the real bottleneck, not compute.
Creator lens: this is the rare AI story with a visible villain and a physical artifact - a cut-up book photographs better than a benchmark chart.

Sources

Qwen3.8-27B put a near-frontier model on a single 3090

Techpositive +0.8#open-source #qwen #local-llm #benchmarks #model-release

Artificial Analysis benchmarks land Qwen3.8-27B next to DeepSeek V4 and GPT-5.6 Luna Max, and people are running it on consumer GPUs today.

What they showed / shipped

Artificial Analysis put Qwen3.8-27B neck and neck with DeepSeek V4 and GPT-5.6 Luna Max, which means near-frontier on a single RTX 3090.
Someone measured 50 tokens/sec at 256K context on a 24GB GPU, and another posted a tuned llama.cpp config for 16GB VRAM at 73k context after 1M+ tokens of agentic coding.
The floor keeps dropping: $100 of GPU runs it at 7.39 t/s.
Same day, llama.cpp shipped v0.1.0 - its first real version number - plus Ling 3.0 support.

Why it matters

Builder lens: if a 27B open model trades blows with frontier APIs, the calculus for anything privacy-bound or high-volume flips to local.
Creator lens: 'near-frontier on a card you can buy' is the most repeatable video format in this space, and the config posts give you a real demo.

Sources

Anthropic finished training Mythos 2 and is not releasing it

AImixed -0.1#anthropic #mythos #model-release #strategy #safety

Anthropic says Mythos 2 is trained but staying internal, with the focus shifting to using it to improve their own systems rather than shipping it.

What they showed / shipped

Anthropic finished training Mythos 2 but does not currently plan to release it, redirecting toward internal improvements.
It lands the same week Anthropic's annualized revenue hit $65B and it shared a run-rate update ahead of an expected IPO.
TechRadar is calling it the 'Apple of AI' - most revenue despite being most expensive.

Why it matters

Builder lens: a trained-but-withheld frontier model is the clearest signal yet that the best capability is being spent on internal compounding, not the API price list.
Creator lens: 'they built it and locked it in a drawer' is a genuinely strange story that does not need hype framing to land.

Sources

A judge who leaned entirely on AI is immune from being sued

Policynegative -0.5#policy #legal #accountability #ai-in-government

A court ruled that a judge allegedly relying wholly on AI to write an order is still covered by judicial immunity, so there is no one to sue.

What they showed / shipped

The Volokh Conspiracy reports a court held that a judge allegedly relying wholly on AI in an order is covered by judicial immunity.
Separately, Israel reportedly created a fake think tank in a likely attempt to shape what AI chatbots say - training-data influence as a live tactic, not a hypothetical.

Why it matters

Builder lens: liability for AI-assisted decisions is landing on 'nobody', which is a much bigger deal for adoption than any model card.
Creator lens: this is the accountability gap in one sentence, and it is far more concrete than abstract alignment talk.

Sources

Copilot's autofix opened a path into Snowflake's Jira

AInegative -0.7#security #agents #ci-cd #copilot #vulnerability

Wiz found that an AI-generated GitHub Copilot autofix could be used to compromise Snowflake's Jira through the CI/CD pipeline.

What they showed / shipped

Wiz published a writeup showing an AI-generated Copilot 'autofix' allowed compromise of Snowflake's Jira via the CI/CD path.
It fits a pattern from today's other agent stories: frontier models are now being pointed at hard exploitation tasks directly.

Why it matters

Builder lens: an agent with write access to your pipeline is a supply-chain surface, and 'it only suggests fixes' stops being true the moment the fix auto-merges.
Creator lens: concrete named-company breach beats generic 'AI security risk' framing every time.

Sources

Cursor launched its own GitHub

AIpositive +0.5#cursor #developer-tools #agents #product-launch

Cursor shipped Origin - repo hosting, PR review, agent runs and Vercel deploys in one place - moving from editor to the whole loop.

What they showed / shipped

Cursor launched Origin: host repos, review PRs, run agents, deploy with Vercel.
Nous Research also shipped Bot Mode for Hermes desktop AI agents, and GE Aerospace says Devin nearly doubled engineering output.

Why it matters

Builder lens: the coding-agent companies are all converging on owning the repo, because whoever holds the code holds the context.
Creator lens: 'Cursor built a GitHub' is a clean one-line hook that needs no setup.

Sources

The money moved to inference and power, not training

Businessneutral +0.2#funding #infrastructure #nvidia #groq #datacenter

Groq raised $350M to pivot from chips to neocloud, Nvidia is putting $1.5B into a SoftBank data center developer, and Wispr raised $280M - the capital is chasing serving capacity.

What they showed / shipped

Groq raised $350M to fund a pivot from AI chips to neocloud.
Nvidia is investing $1.5B in the SoftBank data center developer behind the OpenAI project, and is the exclusive AI partner at SB Energy's PORTS-Pike campus.
Wispr raised $280M at a $2B valuation as it looks past dictation.
Meanwhile AI automation startup Relay shut down, with staff joining Google's Chrome team, and the WSJ argues big tech AI spending is $3T higher than it looks.

Why it matters

Builder lens: money flowing to serving capacity over training runs means inference price pressure keeps going your way.
Creator lens: the Relay shutdown next to the raises is the honest version of this market - both things are true on the same day.

Sources

DeepMind says LLMs cannot invent a new explanation

AImixed -0.2#research #deepmind #reasoning #limitations

A DeepMind paper argues LLMs cannot generate genuinely novel explanatory hypotheses - they interpolate rather than jump.

What they showed / shipped

The paper making the rounds: LLMs can't 'jump' - DeepMind showing models do not produce novel explanatory hypotheses.
Jack Clark's Import AI 469 covers science AI and an RSI simulator in the same week.
A separate study looks at LLM agent time awareness - whether agents know how long they have been running.

Why it matters

Builder lens: if the ceiling is interpolation, then the leverage is in what you feed the model and how you verify it, not in waiting for the next checkpoint.
Creator lens: this is the counterweight to every AGI-is-here take, and it comes from DeepMind rather than a skeptic blog.

Sources

Claude's text watermarks explained

AIneutral +0.1#anthropic #watermarking #provenance #synthid

Anthropic detailed how invisible SynthID text watermarks will work in Claude output, which matters for anyone whose work passes through a model.

What they showed / shipped

Anthropic explained how Claude's invisible text watermarks will work using the SynthID text system.
Context from yesterday's brief: translation via Claude already counts as AI-generated in some venues, so provenance marking has direct downstream consequences.

Why it matters

Builder lens: if output carries a detectable mark, 'did a model touch this' becomes a checkable property rather than an accusation.
Creator lens: platforms will eventually read these marks, so knowing what survives a copy-paste is practical, not theoretical.

Sources

theverge/Claude watermarks theverge.com

Unitree's Superman jumps higher than any human

Sciencepositive +0.6#robotics #unitree #humanoid #hardware

Unitree previewed a humanoid it says jumps higher than any human and tops Usain Bolt's speed, days before a 2000-robot games event.

What they showed / shipped

Unitree previewed Superman, a humanoid that jumps higher than any human with a faster top speed than Usain Bolt.
The Worldwide Humanoid Robot Games return Saturday with over 2000 robots and 666 teams from 16 countries.
Elsewhere in embodied: former SpaceX engineers are building a robotic steel-parts factory, Zipline hit 1M daily autonomous drone deliveries for Uber Eats, and researchers built a floating robot that communicates through movement.

Why it matters

Builder lens: the embodied layer is compounding on a separate curve from the model layer, and the demos are getting harder to dismiss as staged.
Creator lens: this is the most visual category in the brief - a jumping humanoid needs no explanation to land.

Sources

GPT-5.6 Sol got 50% cheaper and is OpenAI's best vision model

AIpositive +0.6#openai #gpt-5-6-sol #vision #pricing #model-release

Roboflow calls GPT-5.6 Sol the best vision model OpenAI has released, and its price just dropped by half on OpenRouter.

What they showed / shipped

Roboflow's evaluation says GPT-5.6 Sol is the best 'vision' model OpenAI ever released.
Pricing was cut 50% on OpenRouter.
On the image-editing side, Grok Imagine is doing Photoshop-style precise edits.

Why it matters

Builder lens: a best-in-class vision model at half price changes what is worth screenshotting and feeding to a model in an automated loop.
Creator lens: vision quality is what makes agent-watches-your-screen demos work, and it just got cheap.

Sources

The anti-AI backlash is organizing

Technegative -0.4#backlash #culture #adoption #open-source

A librarian's guide to avoiding intrusive AI, an 'AI;DR' manifesto and a fast.ai post about joining an AI startup while all your friends hate AI - the resistance is now a genre.

What they showed / shipped

How to disable or avoid intrusive AI - a practical opt-out guide, 243 points on HN.
AI;DR (AI; Didn't Read) hit 535 points - a label for content people refuse to read because a model wrote it.
fast.ai's my friends all hate AI; I just joined an AI startup names the social cost directly, and Tim O'Reilly argues open source is the answer.

Why it matters

Builder lens: 'AI-powered' is now a negative signal for a real segment of users, so shipping quiet AI beats badge-ing it.
Creator lens: this audience is large, articulate and currently underserved by AI channels that only cheerlead.