Anthropic urges coordinated pause in AI development
Anthropic published a blog post urging major AI labs to consider a coordinated and verifiable pause in model development, warning that rapid progress could allow AI systems to improve themselves without human intervention. The company’s leaders said the ability of AI to complete tasks on its own has been doubling every four months and could soon reach “recursive self‑improvement,” raising safety concerns.
Read more: https://www.anthropic.com/institute/recursive-self-improvement
Anthropic published a blog post urging major AI labs to consider a coordinated and verifiable pause in model development, warning that rapid progress could allow AI systems to improve themselves without human intervention. The company’s leaders said the ability of AI to complete tasks on its own has been doubling every four months and could soon reach “recursive self‑improvement,” raising safety concerns.
Read more: https://www.anthropic.com/institute/recursive-self-improvement
Linkedin
NVIDIA | LinkedIn
NVIDIA | 6,739,791 followers on LinkedIn. Since its founding in 1993, NVIDIA (NASDAQ: NVDA) has been a pioneer in accelerated computing. The company’s invention of the GPU in 1999 sparked the growth of the PC gaming market, redefined computer graphics, ignited…
Over the last 2 days, I have been volunteering at the AI Builder Lab in Bangkok as a technical builder.
The event was organized by DataKind as part of a collaboration with OpenAI Academy, with program partners Asian Disaster Preparedness Center (ADPC) and the Gates Foundation.
Together with disaster-management practitioners from over than 13 countries, OpenAI and DataKind staff, DataKind Singapore, and other volunteer builders, we worked on practical AI solutions for disaster-management agencies across APAC.
During the event, I built: Disaster Situation Analysis and Response Planning
An AI-powered web application that helps Civil Protection Authority officers in Timor-Leste quickly understand disaster situations and make faster decisions.
The app helps teams turn disaster reports into clearer situation briefs, response plans, resource gaps, and map-based insights.
What I liked most about this event was the focus on building something practical, not just discussing ideas.
In two days, we built, tested, learned, and improved together.
I'm very grateful to the organizers for the opportunity to contribute to digital public goods and to work on technology that can have a real impact on people's lives.
It was a pleasure to meet you Caitlin Augustin, Aaron Ngo, Gabriel Chua, and Sandy K. in person and work alongside so many amazing people Niki Prastomo, Srinath Sridharan, Anja Kimberley, Daryll Wong, Paisit Jarunnamsiri, Peerapat Sudthivivat, Nakarin Phoorahong, Emad Karim and many others from different countries and backgrounds.
Thank you to everyone involved. I'm happy I could be part of it.
In my next post, I'll share more details about the app we built, the team behind it, and the challenges we solved during the two-day workshop.
https://www.linkedin.com/feed/update/urn:li:activity:7471409170349481984/
The event was organized by DataKind as part of a collaboration with OpenAI Academy, with program partners Asian Disaster Preparedness Center (ADPC) and the Gates Foundation.
Together with disaster-management practitioners from over than 13 countries, OpenAI and DataKind staff, DataKind Singapore, and other volunteer builders, we worked on practical AI solutions for disaster-management agencies across APAC.
During the event, I built: Disaster Situation Analysis and Response Planning
An AI-powered web application that helps Civil Protection Authority officers in Timor-Leste quickly understand disaster situations and make faster decisions.
The app helps teams turn disaster reports into clearer situation briefs, response plans, resource gaps, and map-based insights.
What I liked most about this event was the focus on building something practical, not just discussing ideas.
In two days, we built, tested, learned, and improved together.
I'm very grateful to the organizers for the opportunity to contribute to digital public goods and to work on technology that can have a real impact on people's lives.
It was a pleasure to meet you Caitlin Augustin, Aaron Ngo, Gabriel Chua, and Sandy K. in person and work alongside so many amazing people Niki Prastomo, Srinath Sridharan, Anja Kimberley, Daryll Wong, Paisit Jarunnamsiri, Peerapat Sudthivivat, Nakarin Phoorahong, Emad Karim and many others from different countries and backgrounds.
Thank you to everyone involved. I'm happy I could be part of it.
In my next post, I'll share more details about the app we built, the team behind it, and the challenges we solved during the two-day workshop.
https://www.linkedin.com/feed/update/urn:li:activity:7471409170349481984/
Why do local AI agent scripts fail in production?
Because they lack a dedicated runtime. If a background worker dies mid-execution, your agent's state is wiped out.
In this video, I show you how to use Google ADK alongside AgentSpan (an open-source durable runtime) to make your agents crash-proof. I literally kill the Python process mid-run and reconnect to the exact same execution step using a Run ID.
If you're building agents for the real world, you'll want to check this out.
Watch the full video: https://youtu.be/eRKLD36WzQc
Because they lack a dedicated runtime. If a background worker dies mid-execution, your agent's state is wiped out.
In this video, I show you how to use Google ADK alongside AgentSpan (an open-source durable runtime) to make your agents crash-proof. I literally kill the Python process mid-run and reconnect to the exact same execution step using a Run ID.
If you're building agents for the real world, you'll want to check this out.
Watch the full video: https://youtu.be/eRKLD36WzQc
YouTube
Your AI Agent Doesn't Have a Runtime. That's the Problem
Most AI agents fail when things go wrong. Learn how to stop your AI agents from breaking during crashes or interrupted tool calling.
Most demos show the perfect path, but real-world AI agent development is messy. This video addresses the fragility inherent…
Most demos show the perfect path, but real-world AI agent development is messy. This video addresses the fragility inherent…
Hey devs! 😀🙏
Run Gemma 4 12B locally with Ollama in this step-by-step tutorial. I’ll show you how to install Ollama, download the model, test prompts, and use Gemma 4 12B as a private local AI assistant on your own computer.
Watch on YouTube: https://youtu.be/fppaj5s4CeM?si=kYLjO7bjtw74bUOU
Run Gemma 4 12B locally with Ollama in this step-by-step tutorial. I’ll show you how to install Ollama, download the model, test prompts, and use Gemma 4 12B as a private local AI assistant on your own computer.
Watch on YouTube: https://youtu.be/fppaj5s4CeM?si=kYLjO7bjtw74bUOU
YouTube
Ollama Makes Gemma 4 12B Easy to Run Locally
Run Gemma 4 12B locally with Ollama on your own computer, no API key, no cloud subscription, and no complicated setup.
In this video, I show you how to install Ollama, download and run the Gemma 4 12B model, and test it locally with real prompts for writing…
In this video, I show you how to install Ollama, download and run the Gemma 4 12B model, and test it locally with real prompts for writing…
Media is too big
VIEW IN TELEGRAM
I had the opportunity to volunteer at the DataKind AI Builder Lab in Bangkok, helping build AI solutions with disaster management professionals.
As a thank you to all volunteers, OpenAI gave us a gift. 🎁 Here’s a quick unboxing!
#OpenAI #DataKind #AI #Bangkok #Volunteer #Unboxing #DisasterManagement
As a thank you to all volunteers, OpenAI gave us a gift. 🎁 Here’s a quick unboxing!
#OpenAI #DataKind #AI #Bangkok #Volunteer #Unboxing #DisasterManagement
Hey everyone! I just created a video about Codex subagents that explains everything from scratch. If you want to know how to set up custom agents and run them in parallel, please watch it here: https://youtu.be/xItvCoWNPVo?si=xW6Af1ZaDP1wKxEo
YouTube
Codex Subagents for Beginners: Common Mistakes to Avoid
Are you using Codex but haven't tapped into the power of Subagents yet? In this video, Vlad (Codex Ambassador in Chiang Mai) breaks down everything you need to know about setting up and running custom sub-agents to supercharge your development workflow.
…
…
AI News for Developers #27 is out!
This week’s issue covers the AI updates developers should actually care about:
- Claude Fable 5 returns after its temporary restriction
- Cloudflare launches new AI traffic controls for websites
- Meta releases Brain2Qwerty v2, a brain-to-text AI research project
- Google shares new AI tools for local and multimodal development
- Apple previews the next generation of Apple Intelligence and Siri AI
- Codex Subagents for Beginners: Common Mistakes to Avoid
Read here: https://open.substack.com/pub/proflead/p/ai-news-for-developers-27-fable-5
Enjoy! 😉
This week’s issue covers the AI updates developers should actually care about:
- Claude Fable 5 returns after its temporary restriction
- Cloudflare launches new AI traffic controls for websites
- Meta releases Brain2Qwerty v2, a brain-to-text AI research project
- Google shares new AI tools for local and multimodal development
- Apple previews the next generation of Apple Intelligence and Siri AI
- Codex Subagents for Beginners: Common Mistakes to Avoid
Read here: https://open.substack.com/pub/proflead/p/ai-news-for-developers-27-fable-5
Enjoy! 😉
My first community meetup in Chiang Mai, and I'm happy that it was part of OpenAI Build Week!
We learn about the latest OpenAI updates, Codex, Sites, Ultra Mode, and the new developer workflow:
/plan → /goal A→ build → test → review → deploy
We also had a building challenge. The group completed 11 projects, with some deployed through Sites. Some people even built their projects on their phones!
hank you to everyone who attended, built something, shared ideas, and made the meetup so enjoyable.
It was great to connect with the developer community here in Chiang Mai!
Looking forward to the next one! ;)
#OpenAI #OpenAIBuildWeek #Codex #ChiangMai #DeveloperCommunity #openaidevs #openaidevelopers
We learn about the latest OpenAI updates, Codex, Sites, Ultra Mode, and the new developer workflow:
/plan → /goal A→ build → test → review → deploy
We also had a building challenge. The group completed 11 projects, with some deployed through Sites. Some people even built their projects on their phones!
hank you to everyone who attended, built something, shared ideas, and made the meetup so enjoyable.
It was great to connect with the developer community here in Chiang Mai!
Looking forward to the next one! ;)
#OpenAI #OpenAIBuildWeek #Codex #ChiangMai #DeveloperCommunity #openaidevs #openaidevelopers
This media is not supported in your browser
VIEW IN TELEGRAM
My setup at work 🧑💻
Which GPT-5.6 Sol reasoning level is best for coding?
I gave Medium, High, Max, and Ultra exactly the same task: build a complete React and TypeScript to-do application.
I compared:
• The final applications
• Build time
• Context usage
• Tests and verification
• Code quality
• Accessibility
Finally, I asked GPT-5.6 Sol to review all four codebases and choose the winner.
Does more reasoning always mean better code? Watch the full comparison to see the results:
https://youtu.be/wBkzXGNaZxY?si=k8GtGjuPnQ7LO34c
I gave Medium, High, Max, and Ultra exactly the same task: build a complete React and TypeScript to-do application.
I compared:
• The final applications
• Build time
• Context usage
• Tests and verification
• Code quality
• Accessibility
Finally, I asked GPT-5.6 Sol to review all four codebases and choose the winner.
Does more reasoning always mean better code? Watch the full comparison to see the results:
https://youtu.be/wBkzXGNaZxY?si=k8GtGjuPnQ7LO34c
YouTube
I Tested GPT-5.6 Sol Reasoning Levels So You Don't Have To
Does higher GPT-5.6 Sol reasoning actually improve your code? See how four different settings affect coding task results and speed.
This analysis tests the coding performance of GPT-5.6 Sol by running the exact same coding task across four distinct reasoning…
This analysis tests the coding performance of GPT-5.6 Sol by running the exact same coding task across four distinct reasoning…
New AI News for Developers
The main updates from July 19–26:
• OpenAI improves Codex multi-agent workflows
• Anthropic releases Claude Opus 5
• Google launches Gemini 3.6 Flash
• GitHub adds controls for AI agents
• Microsoft releases its Agent Framework Harness
• Cursor introduces automatic model routing
• JetBrains improves context for coding agents
Read the newsletter: https://open.substack.com/pub/proflead/p/ai-news-for-developers-28-codex-multi
The main updates from July 19–26:
• OpenAI improves Codex multi-agent workflows
• Anthropic releases Claude Opus 5
• Google launches Gemini 3.6 Flash
• GitHub adds controls for AI agents
• Microsoft releases its Agent Framework Harness
• Cursor introduces automatic model routing
• JetBrains improves context for coding agents
Read the newsletter: https://open.substack.com/pub/proflead/p/ai-news-for-developers-28-codex-multi
More cool stuff from OpenAI just arrived! 🫰
I received some Codex Ambassador swag and a postcard from Romain Huet and the OpenAI team.
Thank you for this. I’m proud and happy to be part of the Codex Ambassador journey and excited about what’s ahead!
For updates about upcoming Codex events, workshops, hackathons, and other activities, visit: https://proflead.dev/codex/
P.S. The unboxing video will be available soon on my YouTube channel. 😀
I received some Codex Ambassador swag and a postcard from Romain Huet and the OpenAI team.
Thank you for this. I’m proud and happy to be part of the Codex Ambassador journey and excited about what’s ahead!
For updates about upcoming Codex events, workshops, hackathons, and other activities, visit: https://proflead.dev/codex/
P.S. The unboxing video will be available soon on my YouTube channel. 😀
OpenAI Codex Ambassador Program: Complete Guide
What does a Codex Ambassador do? Who can apply? How much time does it require?
I created a simple guide with:
- Program details
- Current application status
- Preparation advice
- Official links and FAQ
- Photos and lessons from my Codex event in Chiang Mai
Read it here:
https://proflead.dev/posts/openai-codex-ambassador-program/
If you are in Thailand, follow my Codex page for upcoming events and workshops:
https://proflead.dev/codex/
What does a Codex Ambassador do? Who can apply? How much time does it require?
I created a simple guide with:
- Program details
- Current application status
- Preparation advice
- Official links and FAQ
- Photos and lessons from my Codex event in Chiang Mai
Read it here:
https://proflead.dev/posts/openai-codex-ambassador-program/
If you are in Thailand, follow my Codex page for upcoming events and workshops:
https://proflead.dev/codex/
New video: ChatGPT Work vs Chat vs Codex
These three modes can be confusing. In this video, I explain how they differ and show when you should use each one.
The simple version:
Chat = quick answers and ideas
Work = multi-step projects and deliverables
Codex = building and working with software
Watch here:
https://youtu.be/TDZ0FKHP5ts
Which one do you use most?
These three modes can be confusing. In this video, I explain how they differ and show when you should use each one.
The simple version:
Chat = quick answers and ideas
Work = multi-step projects and deliverables
Codex = building and working with software
Watch here:
https://youtu.be/TDZ0FKHP5ts
Which one do you use most?
YouTube
ChatGPT Work vs Codex: When to Use Chat, Work, or Codex
ChatGPT now has three different ways to work: Chat, Work, and Codex. But which one should you actually use?
In this video, I explain the differences between ChatGPT Work, regular Chat, and Codex. I also test Work and Codex on a real project to show what…
In this video, I explain the differences between ChatGPT Work, regular Chat, and Codex. I also test Work and Codex on a real project to show what…
Codex Community Meetup in Chiang Mai
Join me on August 28 for a free, in-person meetup focused on:
• ChatGPT Work vs. Codex
• Practical use cases
• Hands-on challenge
• Knowledge sharing and networking
• Prizes for three challenge winners 🏆
⚠️ Only 20 participants will be accepted.
Space is very limited, so register as soon as possible:
👉 https://luma.com/yvy17hzh
Registration is subject to approval.
Join me on August 28 for a free, in-person meetup focused on:
• ChatGPT Work vs. Codex
• Practical use cases
• Hands-on challenge
• Knowledge sharing and networking
• Prizes for three challenge winners 🏆
⚠️ Only 20 participants will be accepted.
Space is very limited, so register as soon as possible:
👉 https://luma.com/yvy17hzh
Registration is subject to approval.
Luma
Codex Community Meetup - Chiang Mai · Luma
Join the Codex Community Meetup in Chiang Mai for a morning of practical learning, knowledge sharing, and networking.
We will explore the differences between…
We will explore the differences between…
Official ChatGPT Desktop App for Linux is Available! 🐧
OpenAI has launched the official Linux preview with native Codex support.
What's covered in the new video:
- Checking Distro & Architecture (
- Installing Bubblewrap sandbox dependency
- Downloading
- Installing via App Center & Terminal
- Desktop app tour
Watch now: https://youtu.be/AfuHEdxktF8
OpenAI has launched the official Linux preview with native Codex support.
What's covered in the new video:
- Checking Distro & Architecture (
cat /etc/os-release, uname -m)- Installing Bubblewrap sandbox dependency
- Downloading
.deb package from - Installing via App Center & Terminal
- Desktop app tour
Watch now: https://youtu.be/AfuHEdxktF8
YouTube
ChatGPT Linux Is Finally Here: How to Install the Official App + Codex
The official OpenAI ChatGPT desktop app is finally available in preview for Linux, featuring Codex integration directly inside the app! 🚀
In this step-by-step guide, I'll show you how to check your system requirements, install dependencies like Bubblewrap…
In this step-by-step guide, I'll show you how to check your system requirements, install dependencies like Bubblewrap…
AI News for Devs #29
This week’s biggest updates:
⚡️ Gemini 3.7 Flash
⚡️ GPT-5.6 Sol Ultrafast
🐧 ChatGPT and Codex arrive on Linux
🔓 New open models from Qwen and Meta
🧩 Agent Plugins 1.0 for GitHub Copilot
🔀 New NVIDIA tools for AI agents
Read the full newsletter: https://proflead.substack.com/p/ai-news-for-developers-29-gemini
This week’s biggest updates:
⚡️ Gemini 3.7 Flash
⚡️ GPT-5.6 Sol Ultrafast
🐧 ChatGPT and Codex arrive on Linux
🔓 New open models from Qwen and Meta
🧩 Agent Plugins 1.0 for GitHub Copilot
🔀 New NVIDIA tools for AI agents
Read the full newsletter: https://proflead.substack.com/p/ai-news-for-developers-29-gemini
Substack
AI News for Developers #29: Gemini 3.7 Flash, GPT-5.6 Ultrafast & Codex on Linux
Hey Devs!
Researcher Library is now officially listed by Obsidian!
I developed it because Obsidian is my main note-taking app. When I conduct a literature review, it’s essential to keep my research papers and related notes organized properly.
Researcher Library helps you:
• Import and organize PDFs
• Extract available metadata
• Track reading progress
• Search and sort papers
• Create linked research notes
• Keep everything local and private
I’m genuinely happy to see it officially reviewed and listed in the Obsidian Community Plugins directory!
🔗 https://community.obsidian.md/plugins/researcher-library
#Obsidian #Research #LiteratureReview #NoteTaking #PKM #Productivity #OpenSource
I developed it because Obsidian is my main note-taking app. When I conduct a literature review, it’s essential to keep my research papers and related notes organized properly.
Researcher Library helps you:
• Import and organize PDFs
• Extract available metadata
• Track reading progress
• Search and sort papers
• Create linked research notes
• Keep everything local and private
I’m genuinely happy to see it officially reviewed and listed in the Obsidian Community Plugins directory!
🔗 https://community.obsidian.md/plugins/researcher-library
#Obsidian #Research #LiteratureReview #NoteTaking #PKM #Productivity #OpenSource
OpenAI Codex now has a new Agents Dashboard that makes it much easier to manage multiple coding tasks.
In this quick demo, I show:
•
• managing multiple tasks
• switching between agents
• renaming tasks
•
Watch here 👇
https://youtu.be/O_OCwzKPAHY
In this quick demo, I show:
•
codex agents• managing multiple tasks
• switching between agents
• renaming tasks
•
codex queue to send instructions from another terminalWatch here 👇
https://youtu.be/O_OCwzKPAHY
YouTube
How I Manage Multiple OpenAI Codex Tasks from One Terminal
OpenAI Codex just got a new Agents Dashboard for managing multiple coding tasks from one place.
In this quick demo, I show you how to:
- Open the new Agents Dashboard with codex agents
- Run and manage multiple Codex tasks
- Switch between active agents…
In this quick demo, I show you how to:
- Open the new Agents Dashboard with codex agents
- Run and manage multiple Codex tasks
- Switch between active agents…
A great morning at the Codex Community Meetup in Chiang Mai!
We started with the latest Codex updates, explored how ChatGPT Work and Codex can complement each other, and walked through practical use cases and simple demos.
Then we made it hands-on: everyone chose one annoying, repetitive task and tried to automate it using Codex, ChatGPT Work, or both.
The intimate format gave us an almost one-on-one experience. We could explore each idea together, answer questions, and help turn real problems into working solutions. It was great to see everyone actively experimenting, building, and sharing what they learned.
A big thank you to everyone who attended and brought such great energy and curiosity. I’m looking forward to seeing you all and hopefully some new faces at the next meetup!
We started with the latest Codex updates, explored how ChatGPT Work and Codex can complement each other, and walked through practical use cases and simple demos.
Then we made it hands-on: everyone chose one annoying, repetitive task and tried to automate it using Codex, ChatGPT Work, or both.
The intimate format gave us an almost one-on-one experience. We could explore each idea together, answer questions, and help turn real problems into working solutions. It was great to see everyone actively experimenting, building, and sharing what they learned.
A big thank you to everyone who attended and brought such great energy and curiosity. I’m looking forward to seeing you all and hopefully some new faces at the next meetup!