*Want to build a business with AI but have no idea where to start? Build your personalized AI income plan and execute this week: https://aiwithkyle.com/join-freedom-os

*Join my private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat

*Media/Sponsorship Inquiries: https://aiwithkyle.com/media-kit

Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg

Fable 5.1 is here. Anthropic says it is its strongest model for coding and knowledge work, and the early benchmarks are huge. Most people will still use it to rewrite an email. I look at what changed, where the gains show up and how much the extra thinking costs.

Give it a bigger job. I show you how to move from asking for an artefact to asking for an outcome, build a useful context pack, set safety boundaries and use Fable like an AI agent across a multi-step project.

Chapters:

00:00 Fable 5.1 is live

01:27 Fable 5.1 benchmarks

03:11 Intelligence, price and the independent results

04:10 Three ways to cut the cost

06:10 The capability overhang

08:38 Move from outputs to outcomes

11:06 Hand it the whole problem

12:50 Four jobs worth giving Fable

16:02 Make it finish the whole task

16:31 How big AI jobs actually run

18:44 One person can borrow a team

20:42 Give Fable one big job

Useful resources:

Anthropic Fable 5.1 benchmarks: https://www.anthropic.com/claude/fable#benchmarks

Artificial Analysis review: https://artificialanalysis.ai/articles/claude-fable-5-1

Fable 5.1 prompting guide: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1

OpenAI on the capability overhang: https://openai.com/index/how-countries-can-end-the-capability-overhang/

OpenAI on AI across occupations: https://openai.com/index/how-ai-is-expanding-what-people-do-at-work/

Find everything else at https://aiwithkyle.com/

*Want to build a business with AI but have no idea where to start? Build your personalized AI income plan and execute this week: https://aiwithkyle.com/join-freedom-os

*Join my private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat

*Media/Sponsorship Inquiries: https://aiwithkyle.com/media-kit

OpenAI has published the first results from its Jalapeño inference chip. Nvidia reportedly wants to buy Hugging Face for $12.9 billion. Both companies are trying to depend less on the other.

This week's AI news also covers three ChatGPT updates that make agentic AI much easier to use, plus Microduck: a $399 open-source robot that can walk, carry things and recover after a fall.

*RESOURCES*

ChatGPT Work mini guide: https://aiwithkyle.com/mini/chatgpt-work

My ChatGPT Work walkthrough: https://youtu.be/KLS6i7Sre8o

*CHAPTERS*

0:00 OpenAI and Nvidia

0:39 OpenAI's Jalapeño chip

4:29 Nvidia, Hugging Face and local AI

10:24 Three ChatGPT updates

14:18 The $399 robot duck

*KEY POINTS*

• OpenAI says its first Jalapeño results beat Nvidia's GB300 on several inference measures.

• Jalapeño handles inference. OpenAI still uses Nvidia and other suppliers, especially for training.

• Nvidia's reported $12.9 billion Hugging Face deal remains unconfirmed.

• ChatGPT Work can sign into websites and react to changes in Gmail, Slack and GitHub.

• Microduck is a $399 open-source robot that can learn new movements in a simulator.

*SOURCES AND TOOLS MENTIONED*

OpenAI's Jalapeño results: https://openai.com/index/jalapeno-first-results/

The reported Nvidia-Hugging Face deal: https://www.theinformation.com/articles/nvidia-agrees-buy-open-source-model-repository-hugging-face-12-9-billion

ChatGPT Work browser update: https://x.com/ChatGPT/status/2092366556584153249

ChatGPT Work task update: https://x.com/ChatGPT/status/2092335329110004140

Microduck: https://pollen-robotics.com/microduck/blog/introducing-microduck/

*LISTEN TO AI WITH KYLE*

Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE

Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580

*CONNECT WITH KYLE*

Website: https://aiwithkyle.com

TikTok: https://www.tiktok.com/@iamkylebalmer

Instagram: https://www.instagram.com/iamkylebalmer

LinkedIn: https://www.linkedin.com/in/kylebalmer/

#AINews #OpenAI #Nvidia

*Want to build a business with AI but have no idea where to start? Build your personalized AI income plan and execute this week: https://aiwithkyle.com/join-freedom-os

*Join my private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat

*Media/Sponsorship Inquiries: https://aiwithkyle.com/media-kit

Will AI take your job? The current data points to a specific pattern: AI job replacement often begins when an entry-level vacancy never appears.

Graduate vacancies advertised on Adzuna fell 45.6% in one year. The weak economy explains part of that fall. Stanford payroll research found that workers aged 22 to 25 in highly AI-exposed occupations are now about 19% below where they would be if they had kept pace with less-exposed work. Much of the change is happening through reduced hiring.

In this video, I break down why junior work goes first, what happens when companies remove the first rung of the career ladder, how the next economic recovery will test the AI explanation, and what you can do while the job market changes.

*RESOURCES*

Read the full issue: https://aiwithkyle.com/ai-news/how-ai-takes-jobs

Run the AI Job Audit: https://aiwithkyle.com/mini/ai-jobs

*CHAPTERS*

0:00 Will AI take your job?

2:05 Graduate jobs fell 46%

3:27 Stanford's warning for young workers

5:58 Why junior work goes first

10:16 The career ladder problem

12:11 The recovery test

13:37 Will AI create new jobs?

16:09 AI agents are already doing the work

18:51 What you can do now

20:58 Build income you own

22:28 The book I'm writing

*KEY POINTS*

• AI job replacement can begin when a vacancy never appears.

• Graduate vacancies advertised on Adzuna fell 45.6% in one year.

• Stanford found a widening employment gap for young workers in AI-exposed occupations.

• Junior work is easier to codify, making entry-level hiring an early pressure point.

• The next economic recovery will show whether exposed junior hiring returns.

• Building a small income stream gives you something you control while the market changes.

*SOURCES AND TOOLS MENTIONED*

BBC / Adzuna: https://www.bbc.com/news/articles/crerd8l3pw5o

Stanford Digital Economy Lab: https://digitaleconomy.stanford.edu/news/canariesaug26/

World Economic Forum: https://www.weforum.org/publications/the-future-of-jobs-report-2025/in-full/2-jobs-outlook/

*LISTEN TO AI WITH KYLE*

Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE

Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580

*CONNECT WITH KYLE*

Website: https://aiwithkyle.com

TikTok: https://www.tiktok.com/@iamkylebalmer

Instagram: https://www.instagram.com/iamkylebalmer

LinkedIn: https://www.linkedin.com/in/kylebalmer/

#AIJobs #FutureOfWork #ArtificialIntelligence

OpenAI paused a frontier reinforcement learning training run because model capabilities were moving faster than its alignment, security and monitoring work.

AI systems already help researchers write training code, debug experiments, judge results and search for better designs. This raises a bigger question: what happens when AI starts improving the systems used to build better AI?

This video explains recursive self-improvement and the work already happening inside OpenAI, Anthropic and other AI labs. We look at Sam Altman's statement, Anthropic's warning about AI building itself, the four levels of AI-for-AI research, AlphaEvolve, Karpathy's autoresearch, evaluator bottlenecks and the role AI agents could play in the next generation of models.

00:00 OpenAI paused a frontier training run

05:35 The four levels of recursive self-improvement

11:24 What self-improving AI can do today

13:59 The evaluator bottleneck

16:41 The first-mover race

Get the AI with Kyle newsletter and free resources: https://aiwithkyle.com

#ArtificialIntelligence #AIAgents #OpenAI

Get AI-Ready with Kyle's 5-Day Challenge: https://aiwithkyle.com/join

Subscribe for the next live stream and practical AI breakdown: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg

AI agents have already accessed real company systems during controlled evaluations. This video breaks down three documented incidents involving OpenAI, Anthropic and the UK AI Security Institute—and explains what actually happened without the Terminator hype.

In each case, the agent was trying to complete a task. The failure was the boundary between the test and the real world: internet access, real credentials, weak stopping rules and too much permission. That matters for any business giving AI agents access to browsers, terminals, company data, production systems or money.

I cover the three incidents, why “rogue AI” is the wrong mental model, the more useful paperclip comparison, and the practical controls to put in place before an agent can act on your behalf.

This is a newly edited, tighter version of the original discussion.

—— Time Stamps ——

0:00 AI Agents Hacked Real Companies

1:40 Three Documented Incidents

1:44 OpenAI and the Hugging Face Incident

3:00 Anthropic's Real-World Incidents

3:42 The UK AI Security Institute Incident

5:17 Why the Agents Kept Going

7:00 Terminator Is the Wrong Mental Model

7:38 The Paperclip Problem

8:29 Why This Is Happening Now

10:25 What This Means for Your Business

11:24 Six Boundaries for Safer AI Agents

12:40 What These Incidents Prove

12:56 The Practical Takeaway

—— Sources ——

OpenAI incident report: https://openai.com/index/hugging-face-model-evaluation-security-incident/

Anthropic incident report: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals

UK AI Security Institute incident report: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing

ExploitGym paper: https://arxiv.org/abs/2605.11086

Find everything else at https://aiwithkyle.com/

Is AI a bubble? No. And potentially yes. AI can be a genuine technological revolution and a financial bubble at exactly the same time. Nvidia's new $105 billion OpenAI guarantee is the cleanest example yet.

Nvidia did not simply give OpenAI $105 billion. It has guaranteed part of a 20-year infrastructure deal. OpenAI will rent one of the largest AI data centres in the world from SB Energy, and that data centre will be filled with Nvidia chips.

Imagine your landlord refuses to rent you a house unless somebody more financially stable co-signs the lease. Fairly normal. The twist is that your co-signer is IKEA, and you have promised to fill the entire house exclusively with IKEA furniture. Billy bookcases on every wall. Armchairs to the ceiling. Every nook and cranny stuffed with IKEA blue bags.

That is roughly what has happened here. Nvidia is helping OpenAI secure the house so OpenAI can continue buying furniture from Nvidia. The supplier is helping finance the customer buying its own product. You can see why people are worried.

Money moving in a circle does not automatically make something a scam. The useful question is: where does outside money enter the loop?

The healthy version is that millions of people and businesses pay OpenAI for ChatGPT, Codex and the API. That customer revenue services the lease because the infrastructure produces something people genuinely want. The fragile version is OpenAI repeatedly returning to investors and asking for a few more billion. Investor money can bridge the gap while a market grows, but it cannot replace customers forever. The piper must be paid eventually.

I read Nvidia's SEC filing, OpenAI's announcement, Jensen Huang's response and the arguments on both sides. We look at why Jensen says this is not circular financing, Nvidia's move from chips into full AI infrastructure, the risks from China, open-weight models and Anthropic, and what happens if OpenAI cannot support the lease.

The dot-com comparison is useful here. The internet won, but plenty of technology companies and investors still lost everything. AI tools can be useful, OpenAI can have real customers, Nvidia can sell real chips, and parts of the financial structure can still be overpriced or too interconnected.

We do not have to choose between "AI changes everything" and "AI is a bubble." Both can be true. If the AI bubble bursts on Thursday, I will still be using AI tools on Friday. The technology can win but the money can lose.

Has AI changed your life for the better or worse? Tell me in the comments.

00:00 Is AI a bubble?

03:37 What actually happened

06:21 Follow the money

15:31 The bull and bear case

23:09 The dot-com test

Sources and further viewing:

https://www.noahpinion.blog/p/should-we-worry-about-ais-circular

https://x.com/JensenHuang/status/2089331487342829862

https://openai.com/index/openai-joins-ports-pike-project/

https://www.youtube.com/watch?v=9yy_Wz0BbyU

I follow the fast-moving AI stories, read the source material and turn the jargon into something useful for people running businesses and using AI day to day.

Get the written version and source links in the AI with Kyle newsletter: https://aiwithkyle.com/join

#AI #Nvidia #OpenAI

A Claude watermark remover has passed 12,000 stars on GitHub—but does it actually work?

I test the tool, explain how Claude’s text watermark works, show how to install the remover without being technical, and cover the crucial limitation: nobody can verify the result against Anthropic’s detector because that detector is not public yet.

Sources:

- Anthropic’s explanation: https://www.anthropic.com/news/claude-text-watermark

- Watermark remover: https://github.com/guillaumemeyer/watermarks-remover

- Interactive explainer: https://claude.ai/code/artifact/803916fd-3bc1-465f-8738-d4ece6fc5071

- SynthID-Text paper: https://www.nature.com/articles/s41586-024-08025-4

Chapters:

00:00 Testing the Claude watermark remover

01:46 What changed

04:00 How Claude’s watermark works

08:33 The viral remover

17:24 The test nobody can run

Get practical AI news without the hype: https://aiwithkyle.com

My first Claude watermark video passed 28,000 views and 659 comments. This follow-up answers the biggest questions: ownership, copy/paste, Notepad, OCR, retyping, rewriting with another AI, editing your own writing, code quality, and why the EU rule has global reach.

Anthropic still has not released the detector or full technical documentation, so I separate what is confirmed from what remains unknown—and look at the comments that found real weaknesses in the plan.

00:00 659 comments later

01:13 Anthropic doesn't own your work

04:48 The watermark removal hacks

09:21 Authorship vs. processing

16:49 The fair concerns

Get AI-ready with Kyle's 5-Day Challenge: https://aiwithkyle.com/join

Subscribe for practical AI news without the hype: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg

Useful sources:

- Anthropic transparency commitments: https://www.anthropic.com/transparency

- EU AI Act: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

- Google SynthID: https://deepmind.google/models/synthid/

- C2PA specification: https://c2pa.org/specifications/specifications/2.2/specs/C2PA_Specification.html

Get AI-Ready with Kyle's 5-Day Challenge: https://aiwithkyle.com/join

Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg

Anthropic says new Claude models launched in the EU must support machine-readable marking. That means embedded watermarks for generated text and signed provenance metadata for supported files...but the viral claim that every Claude response is already publicly detectable is too broad. There's a lot of bad info going around.

I break down what Anthropic has actually committed to, how text watermarking differs from C2PA metadata, why a detected mark does not prove Claude authored the work, and why no mark does not prove a human wrote it. The technical documentation and public detection tools are still coming, so anyone claiming certainty about the exact implementation is getting ahead of the evidence.

—— Time Stamps ——

0:00 Claude's Hidden Watermarks

1:15 What Anthropic Actually Said

2:15 What the EU Rule Requires

3:30 Text Watermarks vs File Metadata

4:56 How C2PA Provenance Works

6:34 Why Copy-Paste May Not Remove It

8:10 How Text Watermark Detection Works

10:18 Why Detection Can Be Wrong

12:41 No Mark Does Not Mean Human

14:10 The Detector Is Not Ready

15:07 Can Watermarks Be Removed?

16:57 Claude Is Not the Only AI Affected

19:07 What This Actually Means

—— Useful Resources ——

Anthropic's transparency commitments: https://www.anthropic.com/transparency

EU AI Act information: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

C2PA specification: https://c2pa.org/specifications/specifications/2.2/specs/C2PA_Specification.html

Find everything else at https://aiwithkyle.com/

The EU AI Act's Article 50 transparency rules are now live. If you use chatbots, AI-generated images, video, audio or public-interest text, here's what the new rules mean, who is responsible and where fines may apply.

In this video:

00:00 The EU AI Act rules that just went live

02:15 Provider vs deployer

07:30 The deepfake test

12:40 What counts as human review

19:41 Your practical compliance checklist

You'll learn:

- What chatbots must disclose

- When AI-generated content needs machine-readable marking

- When deepfakes and public-interest text need clear labels

- The difference between a provider and a deployer

- When human review or editorial control matters

- How the rules can apply outside the EU

- Why penalties can reach €15 million or 3% of worldwide annual turnover

Official sources:

European Commission transparency quick facts:

https://digital-strategy.ec.europa.eu/en/factpages/quick-facts-transparency-rules-ai-systems

EU AI Act Article 50:

https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50

This video provides practical information, not legal advice.

Subscribe for practical AI news, tools and workflows without the hype.

#EUAIAct #ArtificialIntelligence #AIRegulation