The AI Deck

What actually happened in AI today

Read at the source, always. We point, we do not replace: every story here carries a link to the people who did the work.

21 September 2026 56 sources 6 leads 48 also today window 72h sources checked 21 Sep, 2:42 PM CDT

01Today's Leads

The newest stories the deck has, in the order they were published. Where we have something of our own to say about one it is written underneath; where we do not, the card is the headline and the source, and that is the whole of it.

02The Watch

The second layer: what the creators covered, cross-referenced against the stories above. A story carried by both the outlets and the channels is the day's real lead.

GPT4 outlets · 1 creators

Both sides covered this

ReportedMeta’s Muse is outpacing ChatGPT’s early mobile launchTechCrunch1h 2m ago21 Sep, 2:19 PM CDT
ReportedHiggsfield AI ships new video features in a day with GPT-6 AstraOpenAI8h 21m ago21 Sep, 7:00 AM CDT
CLAUDE4 outlets · 1 creators

Both sides covered this

ReportedBetter Call Sol Or Better Yet Claude or AstraZvi Mowshowitz1d ago20 Sep, 2:47 PM CDT

The Creators

24 channels, each with something written about it. A channel with nothing of ours to say does not appear.

The Neuron

NEWEST · 1h 21m ago / 21 Sep, 2:00 PM CDT
AI Safety Will Never Be Solved 🤷🏻‍♂️

MEASURED FROM 15 TITLES: posts multiple a day; most on copilot (2)

Short developer-facing explainers at a high publishing cadence, on the tooling layer rather than the model layer: GitHub as a workspace, MCP, what happens when an agent deletes your code. The…
Developer tooling, explained plainly.Channel

Matthew Berman

NEWEST · 3h 35m ago / 21 Sep, 11:46 AM CDT
What People Are Building With Jev Is Incredible

MEASURED FROM 15 TITLES: posts multiple a day; most on gpt (4)

The most established channel here and the most willing to be unimpressed. Mixes model news with hardware and with commentary that sometimes argues against the prevailing read, which is rarer in this…
Broad coverage with a working scepticism.Channel

Sanji Nai-Chien

NEWEST · 3h 40m ago / 21 Sep, 11:41 AM CDT
Automate your entire creative pipeline in minutes #ai #efficiency #marketing

MEASURED FROM 15 TITLES: posts every few days; most on claude (5), gpt (4), blender (2)

AI as it meets production craft: model releases read through what they change in an editing and animation pipeline, plus robotics and spatial work. A channel that asks what a tool does to the actual…
Where model releases meet the edit bay. Young channel, short record.Channel

bycloud

NEWEST · 3h 58m ago / 21 Sep, 11:23 AM CDT
NVFP4: The Precision That NVIDIA Is Betting Everything On

MEASURED FROM 15 TITLES: posts weekly; most on deepseek (3), nvidia (2)

Research digest for people who want the mechanism. Covers papers the news cycle never reaches, with enough architecture detail to follow what changed and why it mattered. Roughly weekly.
Papers the news never reaches, explained properly.Channel

LangChain

NEWEST · 4h 21m ago / 21 Sep, 11:00 AM CDT
Building a Harness with Jev

MEASURED FROM 15 TITLES: posts about daily; most on agents (2)

The framework's own channel. PRIMARY on its own releases and nothing else: when LangChain announces LangChain, this is the source rather than a report of the source, which is the one case where a…
Tier 1 on its own releases. Marketing on everything else.Channel

Theo - t3.gg

NEWEST · 4h 21m ago / 21 Sep, 11:00 AM CDT
AI Model "Black Boxes" Are Dangerous

MEASURED FROM 15 TITLES: posts about daily; most on gpt (2)

A working developer's read on model releases, judged by what they do to real codebases rather than by benchmark tables. Valuable precisely because it tests claims against practice. The register is…
Developer practice over benchmark tables. Opinion, stated hard.Channel

All 24 creator pages

LINK-ONLY. Nothing is embedded and no player exists on this site, so the view, the session and the subscribe all land with the creator. Discovery runs entirely on public channel feeds: no API, no key, no quota.

03Also Today

The same 72 hours, ordered by how many outlets ran the same story. Cluster size is the closest thing this deck has to a measure of what the industry thought mattered, and it is a measure of circulation rather than of truth.

AMAZON

Amazon blocks Meta's AI agent Muse from online shopping

ReportedThe Decoder3h 21m ago21 Sep, 12:00 PM CDT3 sources
AI

Apple Mac Studio (M5 Ultra) review: Huge AI and graphics power at a huge premium

ReportedEngadget6h 27m ago21 Sep, 8:53 AM CDT3 sources
OPENAI

B.C. government to sue OpenAI after Tumbler Ridge mass shooting

CommentaryHacker News1h 10m ago21 Sep, 2:10 PM CDT
LLM

You can use any LLM just like JEV

CommentaryHacker News1h 17m ago21 Sep, 2:03 PM CDT
AI

Frontier AI on Your Own Hardware

CommentaryHacker News1h 28m ago21 Sep, 1:53 PM CDT
LLM

Benchmarking LLM Inference at Scale with AIPerf

PrimaryNVIDIA1h 36m ago21 Sep, 1:45 PM CDT
CHIPS
ANTHROPIC
GROK

xAI’s Grok 4.6 is now available in Amazon Bedrock

PrimaryAWS Machine Learning1h 50m ago21 Sep, 1:30 PM CDT
AI

Here's why Apple's Mac Studio has become so expensive

ReportedEngadget1h 51m ago21 Sep, 1:30 PM CDT
AI AGENTS

Meta’s AI agent has been blocked from using Amazon.com

ReportedTechCrunch2h 26m ago21 Sep, 12:55 PM CDT
AI

Big AI to humanity: drop dead

CommentaryHacker News2h 58m ago21 Sep, 12:23 PM CDT
AI

Tesla will soon roll out FSD Supervised in the Czech Republic

ReportedEngadget3h 4m ago21 Sep, 12:17 PM CDT
AI

With Tabby, a former accountant is using AI to make accountants obsolete

ReportedTechCrunch3h 42m ago21 Sep, 11:38 AM CDT
AMAZON

Run Positron on Amazon SageMaker AI for data science workflows

PrimaryAWS Machine Learning3h 47m ago21 Sep, 11:34 AM CDT
AMAZON

How Benchling secured multi-tenant AI agents with Amazon Bedrock AgentCore

PrimaryAWS Machine Learning3h 53m ago21 Sep, 11:27 AM CDT
AMAZON

Reducing medical claims review time with AI on AWS: The EXL Medical IDP solution

PrimaryAWS Machine Learning3h 56m ago21 Sep, 11:24 AM CDT
AI AGENTS
AI

Why is there a waitlist for Siri AI in iOS 27?

ReportedEngadget4h 21m ago21 Sep, 11:00 AM CDT
AI
GEMINI
SOFTBANK

SoftBank to borrow over $11 billion in risky bonds for OpenAI stake

ReportedThe Decoder5h 46m ago21 Sep, 9:35 AM CDT
AI

macOS 27: Workaround to avoid downloading AI models and save storage

CommentaryHacker News6h 10m ago21 Sep, 9:10 AM CDT
POLITICS
CLAUDE
AI

The End Of Upward Mobility – AI is coming for the meritocracy

CommentaryHacker News8h 14m ago21 Sep, 7:07 AM CDT
GPT

04How This Page Is Made

The part most aggregators leave out. This is the short version; the full method, every rule and both refusals, lives on the methodology page, and that page governs.

What the tiers mean

The colour is an argument about evidence, not decoration, and it never carries the meaning alone: every chip is also a word.

  • Primary (gold): the organisation said it itself, or this is the paper.
  • Reported (cyan): an established outlet covering it.
  • Commentary (violet): analysis, opinion, or community attention.
  • Unverified (stone): unattributed, rumoured, or contradicted.

What we read, and what we do not

We read published feeds only. No article page is ever fetched, nothing is crawled, and no publisher's text is passed through a model to make a summary. Where a publisher's robots file asks AI tools to stay out, we carry the headline and the link and nothing else, whatever their feed happens to hand us.

Who is not here, and why

MIT Technology Review is excluded by policy. Blocks even the bots whose only job is to cite and send traffic. A publisher refusing the referral is asking to be left out, and honouring that is the difference between a curator and a scraper. Kept in this file, with its reason, so nobody re-adds it in six months without reading this line.

What the notes are

The writing under each lead is editorial framing, written from the headline, the source pattern and the evidence tier. It is not a summary of the article and should not be read as one. If you want to know what a piece says, the link is right there, which is the point of the page.

Disclosure

The notes on this page were drafted by an AI collaborator and are published under human editorial responsibility. Headlines, links, dates and source names come untouched from each publisher's own feed. Sources today: AI Snake Oil, AWS Machine Learning, Alignment Forum, Allen Institute for AI, Apple Machine Learning, Ars Technica, BBC Technology, Berkeley BAIR, Chips and Cheese, EleutherAI, Engadget, Gary Marcus, Google AI, Google DeepMind, Google Research, Hacker News, Hugging Face, IBM Research, IEEE Spectrum, Import AI, Interconnects, Last Week in AI, Latent Space, Lil'Log, MIT News, Microsoft Research, Mistral AI, NIST, NVIDIA, One Useful Thing, OpenAI, Platformer, Qwen (Alibaba), ScienceDaily, Sebastian Raschka, SemiAnalysis, SiliconANGLE, Simon Willison, Stability AI, Synced Review, TechCrunch, The Algorithmic Bridge, The Decoder, The Gradient, The Register, The Verge, Together AI, Tom's Hardware, Transformer, Understanding AI, VentureBeat, Wired, Zvi Mowshowitz, arXiv, arXiv cs.CL, arXiv cs.LG.

Removal

Any publisher or creator who would rather not appear here is removed within 24 hours, no argument and no negotiation. That is a standing policy, not a courtesy.