🤝 OpenAI's path to AGI
Dear curious mind,
Today, we're diving deep into OpenAI's strategic moves in the coding space. These moves reveal their fascinating plan toward Artificial General Intelligence.
Beyond OpenAI's master plan, I've gathered updates on impressive new AI tools transforming podcasts, voice generation, and coding assistance. Plus, a media recommendation that might change how you think about AI coding editors beyond their traditional use cases.
In this issue:
💡 Shared Insight
OpenAI’s Secret Master Plan: AI Codes Its Way to AGI
📰 AI Update
Snipd: AI-powered Podcast Player Adds Sharing Video Highlights
Dia: A Small but Impressive Open Text-to-Voice Model
GLM-4-0414 Model Family: Strong 32B Models Which Excel at Coding
AI Coding Editor Windsurf Expands Free Tier Capabilities
🌟 Media Recommendation
Article: AI Code Editor Are Not Only For Coders
💡 Shared Insight
OpenAI’s Secret Master Plan: AI Codes Its Way to AGI
OpenAI communicated mission is “to ensure that artificial general intelligence benefits all of humanity.” Artificial General Intelligence (AGI) is defined as an AI system that can match or surpass humans at any cognitive task and then quickly adapt to new ones.
If you want to glimpse where OpenAI thinks the road to AGI runs, follow the breadcrumbs of its coding tools. The company’s strategy looks a lot like solve software development → build a self-improving AI → unlock AGI. So far, the steps from OpenAI are as follows.
1. Canvas: The Code Review and Editing Interface
In October 2024, OpenAI released a new feature called Canvas in ChatGPT. At first glance, it was “just” a side-by-side editor for text and code. It was the company’s first big step from “chat about code” to iteratively working on the same source file with you.
2. Codex CLI: The Terminal Programming Agent
Last week, OpenAI shipped Codex CLI, an open-source agent that runs right in your terminal. It can read local code, execute commands, and commit fixes—all while chatting in plain English. It is a direct competitor of Anthropic’s Claude Code, which was released a few days earlier.
3. Take-over of an AI coding editor
Behind the scenes, OpenAI has been prowling for a full-blown, developer-loved IDE:
Cursor: OpenAI approached the company and discussed the acquisition, but Cursor declined as they want to stay independent.
Windsurf: currently in late-stage talks for a $3 billion acquisition. If this becomes reality, it would be the largest acquisition from OpenAI so far.
Why does OpenAI want to add an AI coding editor to their portfolio? Owning the workspace of developers gives OpenAI direct access to the actual codebase, not just snippets pasted into chat. That’s a data goldmine for training models that learn to reason about entire projects—and then rewrite themselves. So far, you can opt out of the companies behind the AI coding editors using your data in the settings. But this option might vanish, especially for free users.
Put the pieces together and a pattern emerges:
Insert AI into every layer of the dev stack: Chat about Code → Edit code in Canvas → CLI → full IDE.
Harvest data as the models read, write, compile, and fix real code.
Fine-tune on that data iteratively to make each model become a better coder than the last.
In other words, turn programming itself into a reinforcement cycle where the AI writes code that upgrades the AI that writes code.
Conclusion
OpenAI’s master plan isn’t locked in a vault; it’s hiding in plain sight in the actions OpenAI takes. By conquering the tools developers work with, OpenAI is building a self-improving coding engine. And if AI perfects that craft, it will code its own way to AGI.
📰 AI Update
Snipd: AI-powered Podcast Player Adds Sharing Video Highlights [LinkedIn post]
Snipd has launched Video Snips, which allows users to share highlights as video clips across platforms. The initial rollout includes support for the prominent podcasts Huberman Lab, Modern Wisdom, The Tim Ferriss Show, Diary of a CEO, Founders, Lenny's Podcast, Latent Space, and Diggnation, with plans to expand to more shows in the future. I really like this new feature, but would prefer even more to also have the option to consume the complete episode as a video stream.
Dia: A Small but Impressive Open Text-to-Voice Model [GitHub page]
Input script:
[S1] Oh fire! Oh my goodness! What's the procedure? What to we do people? The smoke could be coming through an air duct!
[S2] Oh my god! Okay.. it's happening. Everybody stay calm!
[S1] What's the procedure...
[S2] Everybody stay fucking calm!!!... Everybody fucking calm down!!!!!
[S1] No! No! If you touch the handle, if its hot there might be a fire down the hallway!
Dia is an impressive 1.6B parameter text-to-speech model which was released with open-weights. The model supports speaker differentiation through tags, maintains voice consistency, and runs at over twice real-time speed on modern consumer GPUs. While the examples on their demo page that compare Dia to competitors are impressive, they're likely cherry-picked to showcase optimal performance rather than everyday results.
GLM-4-0414 Model Family: Strong 32B Models Which Excel at Coding [GitHub page]
The GLM-4-0414 series introduces powerful open-weights LLMs which deliver performance comparable to OpenAI's GPT series and DeepSeek's models. This series includes specialized versions like GLM-Z1-32B-0414 for deep reasoning and GLM-Z1-Rumination-32B-0414 for extended complex problem-solving. The models excel in engineering code, artifact generation, function calling, and search-based Q&A tasks, with impressive benchmark results that rival larger closed models. Particularly noteworthy is the model's exceptional coding capabilities, generating complete, ready-to-run code that can be directly copied and pasted for immediate execution.
AI Coding Editor Windsurf Expands Free Tier Capabilities [Windsurf blog]
Windsurf is significantly enhancing its free tier by offering 25 premium credits monthly (up from 5), unlimited access to Cascade Base model with full agent capabilities, unlimited Fast Tab completions, and expanded Preview/Deploy functionality. Currently, free users can opt out of having their code used for other purposes. But especially with the discussions about an acquisition by OpenAI, users should monitor the privacy policies of Windsurf.
🌟 Media Recommendation
Article: AI Code Editor Are Not Only For Coders
Last month, I had a talk titled “Supercharge Markdown Notes with AI Code Editors” at the PKM Summit in the Netherlands. In the talk, I showed that using an AI coding editor, like Cursor and Windsurf, is an alternative to working with your markdown files natively in the notes applications like Logseq or Obsidian. Besides AI powered semantic search, also advanced file edits or even a combination of these is possible. Adding MCP servers to this setup adds even more tools without the need for a plugin for your notes application.
The article The modern AI workspace by Jake Handy extends my thoughts to many more domains. There are also various usage scenarios described in the article which might inspire you as much as it inspired me.
Furthermore, the two state-of-the-art AI coding editors Cursor and Windsurf have built in support for models from various companies like OpenAI, Google and Anthropic.
This effectively eliminates the need for separate subscriptions to ChatGPT Plus, Claude, or other AI services. You're getting enterprise-grade AI capabilities built directly into your workspace environment.
The potential of using AI coding editors for essentially all projects you are working on is huge. But you have to explore it yourself and judge if it is a path you want to follow.
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.