The AI Engineer World's Fair in San Francisco has just wrapped up, and it's clear that this was one of the most significant AI conferences of the year.
Some of the highlights from the conference included:
- New tools and frameworks designed to make AI more accessible to developers.
- Advancements in multimodal AI models that can process images, text, and voice simultaneously.
- Insights from major tech companies on large-scale AI deployment and infrastructure.
Below, you'll find a breakdown of the publicly available talks, along with more in-depth analyses of select presentations that stood out for their innovative ideas or potential impact on the field.
Whether you're an AI professional or simply interested in the future of technology, there are a lot of insights into the current state and future direction of artificial intelligence - and how we'll be building products with them in the future.
These videos are a must-watch.
Event Recaps
Summaries and YouTube links are available in each of the sections.
We've marked some of our favorite talks with a star (⭐).
If you'd like our deep dives on some of our favorites, you can find them below.
Day 1
Keynotes: Day 1
- ⭐ Llamafile: bringing AI to the masses with fast CPU inference
- Open Challenges for AI Engineering
- Convex Launch
- Hasura Launch: Realtime Data Connectivity for AI
- Hypermode Launch
- Hyperspace Launch
- BotDojo Launch: Enhancing AI Assistants with Evaluations and Synthetic Data
- Emergence Launch: AI Agents and the future enterprise
- Second Order Effects
- Spreadsheets-are-all-you-need: Decoding the Decoder LLM without de code
- The Future of Knowledge Assistants
- AI Engineering Without Borders
- Pinecone Launch: Pinecone Assistant
Click here for summaries of the talks and direct links to the recordings.
Codegen Track
- ⭐ The AI emperor has no DAUs: why most devs still don't use code AI
- GitHub Next Explorations
- Embeddings are Stunting Agents: How Codeium Breaks Through the Ceiling for Retrieval
- Cursor: Building the Human-AI Hybrid Engineer
- Code Generation and Maintenance at Scale
- Self-Evolving Code with AI: Enhancing Quality and Security in CI
Click here for summaries of the talks and direct links to the recordings.
Open Models Track
- ⭐ Everything you need to know about Finetuning and Merging LLMs
- ⭐ Building SOTA Open Weights Tool Use: The Command R Family
- Decoding Mistral AI's Large Language Models
- No more bad outputs with structured generation
- Training Albatross: An Expert Finance LLM
- Fixing bugs in Gemma, Llama & Phi-3
Click here for summaries of the talks and direct links to the recordings.
Day 2
Keynotes: Day 2
- ⭐ What We Learned From A Year of Building With LLMs
- Unlocking Developer Productivity across CPU and GPU with MAX
- From Software Developer to AI Engineer
- What's new from Anthropic and what's next
- LangChain Launch: Infrastructure for building reliable agents
- From Text to Vision to Voice: Exploring Multimodality with OpenAI
- Copilots Everywhere
Click here for summaries of the talks and direct links to the recordings.
Multimodality Track
- ⭐ Moondream: how does a tiny vision model slap so hard?
- Substrate Launch: the API for modular AI
- The era of unbounded products: Designing for Multimodal I/O
- State Space Models for Realtime Multimodal Intelligence
- The Hierarchy of Needs for Training Dataset Development
- The Multimodal Future of Education
- How to build the world's fastest voice bot
Click here for summaries of the talks and direct links to the recordings.
GPUs & Inference Track
- ⭐ Unveiling the latest Gemma model advancements
- ⭐ Making Open Models 10x faster and better for Modern Application Innovation
- Covalent Launch: The GPU Cheatcode: Fine-tune 20 Llama Models in 5 Minutes
- Compute & System Design for Next Generation Frontier Models
- Breaking AI’s 1 Gigahertz Barrier
- Accelerating Mixture of Experts Training With Rail-Optimized InfiniBand Networking in Crusoe Cloud
- Scott Wu and the Making of Devin by Cognition AI
Click here for summaries of the talks and direct links to the recordings.
Deep Dives
Our in-depth runthrough of different talks (and of course links to each of them to watch for yourself).
Technical Deep Dives
We liked these talks because of how deep the speakers went on the technical aspects of their talks.
- Everything you need to know about Finetuning and Merging LLMs
- Moondream: how does a tiny vision model slap so hard?
- Llamafile: bringing AI to the masses with fast CPU inference
On Product & Implementation
Honorable mention for these as we're a Product-focused company.
They covered interesting perspectives on product, operations, and building with AI in general.
- The AI emperor has no DAUs: why most devs still don't use code AI
- What We Learned From A Year of Building With LLMs (the "Avengers" talk)
Novel-Product Talks
Interesting talks which were more product-featuring. We felt like all of these had some elements of unique features, or novel concepts which haven't been brought up much previously.
Anthropic's talk was also quite good, but has been covered a lot recently.