Usually, Data Points brings you the latest AI news, tools, models, and research in brief. But in today’s special winter holiday edition, you’ll find something different: a Data Points-sized summary of twelve of the AI tools that helped us find our way in 2024.
Want a sneak peek? We’ve got you covered:
- The Big Guys (Google, Anthropic, Perplexity)
- Making Media (Black Forest Labs, Udio)
- Building Software (OpenHands, Cursor, LangChain, Meta)
- Bringing It All Together (Replit, UX Pilot, and one more)
The Big Guys:
- OpenAI, Google, and other companies joined in, but Perplexity led the way in AI search. The startup added publisher partners and a shopping app, explored new ad-based revenue models, and generally refused to be outflanked by its even bigger competitors. It remains one of the best ways to get started on anything you want to know more about.
- Google’s NotebookLM went viral, helping introduce AI research assistants to the masses with one killer feature: auto-generated podcasts where multiple hosts discuss whatever document(s) you want to know more about. NotebookLM added even more features at year’s end and shows no sign of slowing down.
- Claude Artifacts showed that there was plenty of room to improve AI interfaces, offering a powerful way to preview, save, edit, and download multimedia documents created with an LLM. OpenAI’s Canvas is a worthy competitor.
Making Media:
- Black Forest Labs’s Flux models pushed everyone to improve their image generators, offering more beauty for fewer bucks. The startup added editing tools, partnerships with Grok, and more. (We still usually use DALL·E 3 when we generate images for Data Points, but we’re happy to have options!)
- Udio (and Suno) created better songs than ever, prompting music publishers to sue, claiming their copyrights had been violated. The music generators have responded, claiming that learning from public music is fair use. All we know for sure is that the music and audio generators we’ve seen in 2024 put those of yesteryear to shame, and we hope they get even better in 2025.
Building Software:
- OpenHands cracked the code for open AI software assistants. It started as OpenDevin, and was a pretty straightforward attempt to recreate the functionality seen in Devin’s influential demo. But OpenHands very quickly gained contributors, testers, and users, and soon became a very capable coding agent in its own right.
- Cursor (and Codeium’s Windsurf) changed the way we IDE. Don’t get us wrong: We still love VSCode, and love it with AI code completion at chatbot tools like Copilot even more. But Cursor’s multi-line edits and early support for multiple AI models helped push everyone to get better, and opened the door for tools like Windsurf that integrate coding agents to make getting started faster and easier. (Just keep a critical eye on those code suggestions and stay disciplined about accepting them too quickly!)
- LangChain’s LangGraph Studio was the first IDE designed to build agentic AI applications. Agents were perhaps the most significant mainstream development in AI in 2024, but LangGraph’s orchestration framework was there much earlier. LangGraph Studio coupled much of what we want from a traditional IDE with new visualizations and other tools that make prototyping an app much easier.
- Meta’s open-source LlamaStack API and distributions simplified customizing, building, and deploying agents using Meta’s open-weight multimodal models. The ability to build using a single endpoint and deploy everywhere from the cloud to an iOS device is a powerful one. We hope that the framework will support even more open models in the future!
Bringing It All Together:
- Replit Agent (and Bolt) didn’t just help developers make code, but build proper applications from scratch. The killer feature of Replit Agent isn’t its powerful agent-based code assistant, or even the ability to quickly prototype and visualize your webapps, but being able to seamlessly deploy to the cloud. Building apps (and for novices, learning how apps are built) with AI has never been easier than it became in 2024.
- But suppose you want to build something more complex than a simple webapp. UX Pilot helped developers and designers roll their own user interfaces, taking away a common barrier between back-end coding and front-end application development. It made it easier for developers to mock up wireframes and for designers to iterate and approve the look and feel of an application. The user experience world has not taken up AI as quickly as coders, for several reasons, but we hope tools like this will help them save time and create even better, more accessible, and more useful applications.
- Finally, for our twelfth tool, please forgive us if we briefly beat our own drums. DeepLearning’s own open-source AISuite, released just a few weeks ago, made it easier for developers to switch between AI models from a host of providers without needing to change their code. Most importantly, AISuite was (and continued to be) built with code and testing contributions from the entire DeepLearning.AI community. So really, this isn’t our twelfth gift to you, but a gift many of you have given to everyone who works in AI.
Looking Forward to 2025:
That’s it for our special winter holiday edition of Data Points. We’ll be back with news on Friday, December 27th. Be sure to check out this week’s equally special holiday issue of The Batch, which rounds up the most important ongoing news stories of the year. It also includes a special message from Andrew Ng. Also, be sure to check back in with us next week on December 30th, where we’ll have a SECOND special New Year’s edition of Data Points. Have a wonderful holidays and a happy new year!