The Great Debate
The phrase "Open Source AI" has sparked passionate reactions, creating a divide among the tech community. In tandem, the Open Source Initiative (OSI) is laying the groundwork by drafting a standard definition for Open Source AI, but community consensus remains elusive.
People have different perspectives on whether or not Artificial Intelligence can truly be open source.
There are technologists who believe Open Source AI is a misnomer because:
- Due to privacy concerns, AI models rely on large data sets that aren't available to the public.
- AI models are not just large blocks of code. They involve massive amounts of data, specialized hardware, complex workflows, and deep expertise. Releasing only the source code for an AI model doesn't necessarily enable others to replicate or contribute to it unless they have access to the same data, infrastructure, and expertise.
Others are concerned about the potential misuse of Open Source AI:
- They worry that Open Source AI can put power in the wrong hands, enabling everyday people to create tools potentially lethal to society.
Then, there are those who support Open Source AI. For example:
- Some technologists embrace platforms like Hugging Face, which host open source AI models. These platforms address concerns about large datasets and complex models by hosting smaller, less resource-intensive models trained on publicly available datasets.
- Still, platforms like Hugging Face face a challenge: infrastructure, fine-tuning strategies, preprocessing pipelines, and scalability limit full replication through open source.
While I respect the open source community for pushing the OSI to reach an accurate definition, I also understand how difficult it must be to define such a complex concept.
In this blog post, I'll share:
- My opinions on Open Source AI
- Insights from my experience with Open Source AI tools
- Examples of tools developed by my company
- Why I believe the Open Source AI seems promising, despite its current limitations
My Stance on Open Source AI
Without open source, AI wouldn’t exist. My friend Dawn Wages, a Microsoft Developer Advocate and Python community expert, recently reminded me of this fact.
I was excited when I heard the phrase Open Source AI because it’s a combination of my two favorite ecosystems. I credit both Open Source and AI for accelerating my career. I also recognize that the concept is a bit idealistic and not straightforward to achieve.
While Big Tech has a monopoly on Artificial Intelligence because large companies have more resources, I believe it’s time for AI to get back to its open source roots. Moving in the direction of open sourcing AI presents an opportunity for transparency, democratization, and community-driven innovation. Yes, the concept is imperfect, but it’s a worthwhile effort to push AI systems towards a future where communities and individuals have more control over the impact AI will have on society. After all, AI is going to continue to expand in influence, and it's we, the people, who feel AI's impact.
So, I’m excited to see how our industry approaches this problem.
Open Source Tools that use AI
Since we haven’t yet perfected the ideal Open Source AI model, I’ve been exploring open source tools that use AI. I believe keeping these tools open source is a step in the right direction, as it answers key questions like:
- How does this software use AI, and how is it handling my data?
- Who controls this tooling?
- Can I customize it to fit my specific use case?
As with many open source projects, these tools empower individuals worldwide to solve problems beyond the original creators’ vision.
Recently, I hosted a Twitter Space about open source tools using AI, such as:
- Open Sauced’s Star Search - Copilot for your git history
- Cerebras Systems - AI hardware optimized for deep learning
- Goose - A developer agent that automates coding tasks
The panelists and company representatives on the Twitter Space resoundingly agreed that open sourcing these products enables global innovation.
Exploring Goose
What is it?
An open source AI tool I’ve recently been experimenting with is called Goose. It’s formally described as an interactive developer agent that can flexibly assist both programming (e.g. writing code) and operational tasks (e.g. fix my development environment).
To get a better understanding of what Goose is, it’s a semi-autonomous CLI tool (built with Python) with full context of your repository. Because of its nature, you can give it one instruction and it uses its “intelligence” to execute that plan from beginning to end without further human intervention. This contrasts with many other AI programming tools that require you to prompt them at each step.
How my skip manager describes Goose. Watch the full video: https://www.youtube.com/watch?v=lppDgrHY88Y
Why my skip manager uses Goose. Watch the full video: https://www.youtube.com/watch?v=lppDgrHY88Y
The History of Goose
A few months ago, engineers at my company created Goose to help streamline migrations for internal engineers. Simultaneously, as a company, we decided to embrace open source more intentionally. As Goose started to gain momentum internally, it made sense for us to share it with the broader developer community.
How Is Open Sourcing Goose Going So Far?
From my perspective, it’s an optimal move that opened the door to community-driven innovation and customizability.
Customizability
Goose shines in two key areas of customizability:
-
You can use it with any Large Language Model (LLM) - Goose is not an LLM. Instead, Goose is an agent that leverages LLMs. At the moment, you can Goose with any of the available providers such as:
- Azure
- Anthorpic
- Bedrock
- Databricks
- Ollama
- OpenAI
- If your preferred LLM isn’t on this list, don’t worry! Because Goose is open source, you can write your own code to add support for a new LLM provider.
Goose comes with built-in helpers called toolkits - While most AI pair programming assistants focus solely on writing code, Goose can do other things like screenshot a code snippet from your IDE or your application’s UI. It can even browse the web for you. Because it’s open source, you can always include a toolkit that would suit your use case.
You can refine Goose’s output - You can provide more context to Goose through a file called .goosehints leading Goose to provide you with more relevant results.
Community
One of the most rewarding aspects of working on Goose has been the camaraderie it’s fostered across my company's various teams. As a company that includes multiple brands it’s easy to feel siloed from others. However, working on Goose has brought together people from different parts of the company, providing an opportunity to collaborate with colleagues I might not have otherwise worked with.
How to explore Goose
The community is continuing to grow. If you find the idea of Goose interesting, I invite you to join the fun as a user or contributor. I think it’s a pretty cool opportunity to influence the direction of AI tools at early stages of development. Hang out with us in Discord
As a user
You can try it out as a user by installing Goose.
As a contributor
And if you’re itching to contribute ideas, code, or content, check out:
Should we Open Source AI?
In my opinion – yes. However, we must do it carefully and thoughtfully with the mindset that we want users to have data privacy and ownership over their internet usage. Yes, in tech we want innovation, but we often want it in exchange for people’s privacy. Instead, let’s work together to create a transparent internet where people know where their data is going and they feel empowered to continue to shape the future of AI.
For now, while fully open source AI models remain a challenge, I’m excited about—and supportive of—open source tools that leverage AI. I believe they represent an important step forward. Tools like Goose are examples of how we can start building toward a future where AI is truly democratized. I encourage you to explore and contribute to tools like Goose to help shape the future of AI.
Share your thoughts!
Where do you stand on the topic of open source AI?
Top comments (1)
Thank you for these valuable insights and the "cookies & milk" that you shared, Rizèl. I spotted BLOCK during the GraphQL Summit (virtual in my case) and tried to find out more about the company. By fluke, I bumped into your post about OSAI here and there you go, found you folks!