🤝 The AI Coder Has Arrived: Meet Devin
An autonomous AI software engineer capable of complex tasks.
Dear curious minds,
AI is transforming every industry it touches, and coding is next in line. Devin, the new AI from Cognition Labs, is a glimpse into this future. It's not just a tool, but a potential collaborator for software engineers, opening the door to amazing possibilities.
This week’s issue brings you the following topics:
Free Access to All Claude 3 Versions, Including the Super-Fast Haiku
Devin: The AI That Codes Like a Pro
AI-Powered Websites: Tailored Experiences for Maximum Conversion (podcast)
If nothing sparks your interest, feel free to move on, otherwise, let us dive in!
🆓🏃 Free Access to All Claude 3 Versions, Including the Super-Fast Haiku
The last issue highlighted the release of the Claude 3 LLM family.
Claude 3 Haiku, the fastest and most affordable model of Claude 3, was only announced but not accessible so far. As stated in a new blog article, Haiku is now available.
The model is three times faster than it direct competitors and processes 21K tokens (approximately 30 pages) per second at a competitive price.
So far, only Claude 3 Sonnet is accessible for free at claude.ai. The other two models, Haiku (fastest) and also Opus (best performing), are according to Anthropic only accessible via the paid API or a Pro subscription. However, you can access all three variants of Claude 3 for free at the Chatbot Arena.
My take: While the most complex AI tasks might crave the full power of models like Claude 3 Opus, Haiku's lightning-fast performance and budget-friendly pricing open up a world of possibilities. For many applications, the difference in output quality may be minor, while the gains in speed and cost-saving will be major. Imagine real-time, AI-powered customer support chats and efficient analysis of large documents - all made more accessible by Haiku. This isn't about replacing top-tier models, it's about unlocking AI's potential for a wider range of scenarios.
🤖👨💻 Devin: The AI That Codes Like a Pro
Cognition Labs has introduced Devin, an autonomous AI software engineer capable of planning and executing complex engineering tasks.
Devin is equipped with developer tools, can learn new technologies, build and deploy apps, find and fix bugs, train AI models, and contribute to open-source projects.
Videos shared in the introduction blog article demonstrate Devin's capabilities in action, including planting secret messages in images, making a Game of Life app, debugging code, fine-tuning language models, and completing freelance work on Upwork.
Devin generated this image along a tutorial in a blog post, which was used as an input query. [source] On the SWE-bench, which tests if LLMs can resolve real-world GitHub issues, Devin achieved a new state-of-the-art by correctly resolving 13.86%.
Devin outperforms other LLMs at solving real-world GitHub issues. Devin and GPT-4 were evaluated on a random 25% subset of the 2294 instances.
Devin is currently in early access. If you are interested, you can request access by filling a Google Forms. To get off the waitlist faster, you are asked to describe a detailed task which can directly be sent to Devin.
Cognition Labs, the company behind Devin, is an applied AI lab focused on reasoning, aiming to build advanced AI teammates. They recently raised a $21 million Series A funding led by Founders Fund.
My take: Devin's ability to perform complex programming tasks like building apps, debugging, and training models is highly impressive. The videos demonstrating these skills are eye-catching. However, it would be helpful to get more context on Devin's success rate and understand the scope of tasks it handles reliably. The SWE-Bench results suggest Devin possesses a stronger problem-solving capacity than other LLMs, but there is still a lot of room for improvements. The fact that Devin can handle freelance work raises interesting questions about self-funding. It could lead to a scenario where the system uses its capabilities to generate more computation resources, taking steps towards artificial superintelligence that solves tasks to become more powerful and autonomous.
🌐🧠 AI-Powered Websites: Tailored Experiences for Maximum Conversion (podcast)
In Episode 208 of the Marketing Against The Grain podcast, Kipp Bodnar (CMO at HubSpot) and Emmy Jonassen (VP of Marketing at HubSpot) explore the transformative potential of AI for websites.
They discuss the groundbreaking impact of AI-driven chats and how Hubspot performed experiments to revolutionize user engagement on their website.
Emmy envisioned a future in which AI will take over the web search and way fewer people are accessing websites directly. In this scenario, it will be more important than today to convert visitors to customers. To achieve this, it is essential to understand the intention of visitors as early as possible and create a tailored experience instead of a static page.
My take: The idea of adapting a webpage according to the user intention reminds me of the feature “Browse For Me” in the iOS app Arc Search from the Browser Company: For any search, an LLM creates a tailored website. This podcast episode carries this thought a bit further and envisions that also other websites evolve in a way that they adapt their content to their visitors.
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.