X / Twitter
Nobody explains meditation like Seinfeld.
"It's like you have a cell phone and then somebody gives you the charger. Oh, I can get this thing up to a hundred anytime I want?!
"It doesn't feel like anything. Doesn't do anything. I don't get it. I don't understand it. But here's the difference: at 1pm that day, my head does not hit the desk like it used to. ... I sail through the day."
The Ultimate Prompts For Strategic Decision-Making by @johncutlefish @amplitude https://t.co/FqVXOrg0cn
Lotus
Diffusion-based Visual Foundation Model for High-quality Dense Prediction
Leveraging the visual priors of pre-trained text-to-image diffusion models offers a promising solution to enhance zero-shot generalization in dense prediction tasks. However, existing methods often uncritically use the original diffusion formulation, which may not be optimal due to the fundamental differences between dense prediction and image generation. In this paper, we provide a systemic analysis of the diffusion formulation for the dense prediction, focusing on both quality and efficiency. And we find that the original parameterization type for image generation, which learns to predict noise, is harmful for dense prediction; the multi-step noising/denoising diffusion process is also unnecessary and challenging to optimize. Based on these insights, we introduce Lotus, a diffusion-based visual foundation model with a simple yet effective adaptation protocol for dense prediction. Specifically, Lotus is trained to directly predict annotations instead of noise, thereby avoiding harmful variance. We also reformulate the diffusion process into a single-step procedure, simplifying optimization and significantly boosting inference speed. Additionally, we introduce a novel tuning strategy called detail preserver, which achieves more accurate and fine-grained predictions. Without scaling up the training data or model capacity, Lotus achieves SoTA performance in zero-shot depth and normal estimation across various datasets. It also significantly enhances efficiency, being hundreds of times faster than most existing diffusion-based methods.
@jxmnop we can take out the garbage for you https://t.co/zlgxT1seEv
If you're building an AI application with LLMs, do not start by writing code!
Instead, start by putting together an evaluation set, so that you can measure the improvement or decline of your RAG, Agent, or other LLM-based system as you make changes to your system (e.g trying different embedding models, switching your LLM from GPT4o to Claude Sonnet, adding filters to your semantic search etc.)
This the core part of Evaluation Driven Development, a very useful idea I learned from @jxnlco via his consulting work and his Systematically Improving Your RAG Application course (both of which I highly recommend).
Ai Automations
Sam Altman's spent years refining his note-taking process. Here's exactly what he does. https://t.co/vpIWkK7WYG