Six Nerdy AI Actions for the Lengthy W/E. I’ve simply learn that a lot of AI engineers in the US are running the rate race, feeling burnout. Right here within the European AI scene issues are innately a bit extra relaxed.
Aah… A protracted financial institution vacation in London; a lot stuff to do on this superb metropolis! However in case you are feeling the AI FOMO kick and might’t survive a protracted weekend IRL, listed below are six AI actions for you:
Generate comics with AI. I gave it a go, generated a couple of brief comics, and having enjoyable up to now. The AI workforce at Bytedance simply launched a formidable diffusion-based, zero-shot, text-to-image and image-to-video mannequin that generates superb movies and comics. Checkout the demo, paper and repo right here: StoryDiffusion: Consistent Self-Attention for Long-Range Image and Video Generation. Ensure you click on the Comedian Technology Demo hyperlink and be affected person.
Learn to construct and use a strong AI Brokers stack. I completely imagine that the long run goes to be about tens of millions of AI Brokers working and producing revenue for the folks. On this vid, Tony exhibits how one can create an AI agent that fetches all of the feedback of a YouTube video and generates insights to enhance video content material. Tony makes use of an AI Agent stack that appears very strong: 1) CrewAI agent framework, 2) the nifty Ollama 3) Groq the super fast AI inference engine, and 4) AgentOps the observability tool for AI agents
Play AI City sport in your laptop. I’ve performed this sport and it’s fairly addictive! AI City is a MIT-licensed sport -developed by a16z- during which AI characters stay, chat and socialise in a digital city. You can play AI Town online here. However in case you hate cloud signups like me, and need to create your personal customized AI City, verify this out: The best way to create your personal AI City with Llama-3 primarily based brokers in your native surroundings utilizing the nifty Ollama and the one-click deployment with the amazing Pinokio AI browser. The video beneath gives extra particulars on deploying AI City regionally with Pinokio.
Learn the most recent on In-Context Studying (ICL.) There’s a debate amongst AI researchers on whether or not In-Context Studying inside a long-context window measurement can absolutely beat fine-tuning completed with highly-curated information, when it comes to area information and accuracy of mannequin outputs. Let’s see…
This can be a nice submit on that: Is Fine-Tuning Still Valuable? A reaction to a recent trend of disillusionment with fine-tuning
Ethan, a well-known AI researcher is extra assertive: Fine-tuning is dead. Prompts have closed the gap.
DeepMind: Many-Shot In-Context Learning. Many-shot in-context studying works very nicely and will be utilized universally. “We discover that each Strengthened and Unsupervised ICL will be fairly efficient within the many-shot regime, notably on advanced reasoning duties.”
In-Context Learning with Long-Context Models: An In-Depth Exploration. “We conclude that though long-context ICL will be surprisingly efficient, most of this achieve comes from attending again to comparable examples quite than activity studying.”
MSR: Make Your LLM Fully Utilize the Context. On this paper, Microsoft proposes an answer to the “lost-in-the-middle” long context problem, during which LLMs battle utilizing info situated within the center part throughout the lengthy context.
Learn the 2024 State of AI Readiness Report. Good report with good insights and funky charts. The analysis workforce at Scale AI interviewed 1,800 AI/ ML practitioners on the most recent AI traits, utilized AI, and what it takes past “adopting AI.” Hyperlink to the report: Scale Zeitgeist 2024 AI Readiness Report, 3r ed (pdf, 47 pages)
Learn this free e-book and fall down the rabbit gap of designing Neural Nets “This primer is an introduction to this fascinating discipline [of differentiable programming applied to NNs] as imagined for somebody, like Alice, who has simply ventured into this unusual differentiable wonderland.” Hyperlink: Alice’s Adventures in a Differentiable Wonderland
Have a pleasant week.
KANs: A New [better?] Alternative to the Multi-Layer Perceptron
[deep dive] MOMENT: A Foundation Model for Time-series Tasks
Google TeraHAC: A New Algo for Clustering Trillion-Edge Graphs
How to Build Domain-specific Datasets for Training AI Models
LMSYS Kaggle Chatbot Competition – Predicting Human Preference
[notebooks] Examples of Automated Multi Agent Chat with Autogen
Powerful Automatic Speech Recognition + Diarisastion + Speculative Decoding
Meta AI: A Simple Recipe to Improve CoT Reasoning with DPO+NLL
Octopus v4: A Graph of LMs to Integrate Multiple Specialised Open Models
WildChat Dataset: 1 Million Real-world User-ChatGPT Interactions
OpenStreetView-5M: 5.1 Million Geo-referenced Street View Images
Ideas? Ideas? Suggestions? email Carlos
Curated by @ds_ldn in the midst of the evening.