Unleash the power of AI-driven agents in a user-friendly, low-code database environment. Leverage AI Agents to automate the workflows you once spent countless human hours on. Reframe - where operational efficiency meets the future of data management.

Get In Touch

This wiki is home to content for Machine Learning and AI-focused material, covering everything from the intricacies of LLMs to the process of creating vector embeddings.

Generative Image Dynamics

Animating still images to create realistic videos is a challenging problem in computer vision and graphics. Existing methods like video GANs can generate relatively short videos from static images, but often suffer from temporal incoherence over longer durations. In the paper "Generative Image Dynamics" published at CVPR 2023, researchers from Google propose a new approach to model scene dynamics that enables generating coherent, indefinitely long videos from single images.

Spellburst: A Node-based Interface for Exploratory Creative Coding with Natural Language Prompts

Creative coding often involves exploratory programming to produce digital artwork. Artists typically start with a high-level semantic goal like a "stained glass filter" and then vary code parameters like shape, color, lines, and opacity to achieve visually pleasing results. However, translating semantic constructs to code syntax can be effortful. Current programming tools also do not facilitate rapid creative exploration. To address these challenges, we present Spellburst, a large language model powered creative coding environment. Spellburst provides: (1) a node-based interface for creating generative art and exploring variations through branching/merging, (2) expressive prompt-based interactions for semantic programming, and (3) dynamic prompt-driven interfaces and direct code editing to seamlessly switch between semantic and syntactic exploration. Evaluations with artists show Spellburst's potential to enhance creative coding and inform the design of computational creativity tools that bridge semantic and syntactic spaces.

Docker Image OS names. What is the difference between Alpine, Jessie, Stretch, and Buster?

Effects of Chunk Sizes on Retrieval Augmented Generation (RAG) Applications

The retrieval module is designed to find the most relevant passages from a large collection of texts to augment the context for the language model. Operating on longer chunks of text allows the retriever to identify documents with broad topical relevance even when matches may be approximate. In contrast, the language model benefits from conditioning on shorter, more coherent spans during synthesis so it can maintain consistency in generated text.

Retrieval Augmented Generation (RAG) Architecture

RAG systems have a two-step pipeline to generate responses - retrieval and generation . First, in the retrieval phase, the model searches through databases or document collections to find the most relevant facts and passages for the given prompt or user question. For open domains like general web searches, this could leverage indexed webpages. In closed domains like customer support, it may retrieve from a controlled set of manuals and articles.

Retrieval Augmented Generation (RAG) for Large Language Models Explained

RAG is an AI Framework that integrates large language models (LLMs) with external knowledge retrieval to enhance accuracy and transparency. Pre-trained language models generate text based on patterns in their training data. RAG supplements their capabilities by retrieving relevant facts from constantly updated knowledge bases. This grounds the language model's predictions on factual information rather than just its encoded patterns.

Revisiting Langchain: Developers Share Experiences Building with This Popular AI Library

As one of the first libraries for streamlining AI development, Langchain gained quick popularity. But a recent discussions amongst developers in the AI community provided an opportunity for us to reflect on using Langchain versus custom solutions.

This critique of Langchain makes some reasonable points, but could be expanded on in a more constructive manner:

ControlNet Explained

ControlNet is a new neural network architecture built to control existing large image diffusion models - like Stable Diffusion - by enabling them to support additional input conditions. Users of vanilla Stable diffusion models will understand why this is immensely valuable. While SD models produce fantastical images , getting them to generate specific details that match your expectations such as a human in a specific pose is an extremely consuming process that takes dozens of seed/prompt-tweaking → generation cycles. These process essentially requires the interpretation of raw inputs into object-level or scene level understanding. With ControlNet, it is now possible to use a wide array of input conditions such as outlines, human poses, depth maps, outlines, edge maps, segmentation maps, key-points etc to control diffusion models making them highly malleable.

Can I get your attention? Attention Models for Generative Models Explained

Attention mechanisms are simply input processing techniques enable neural networks to process on specific details of the entire dataset at a time. In the case of image processing, the model checks that highly detailed features in distant portions of the image are consistent with one another. Similarly, when it comes to text processing, the model strives to make sure that distant portions of the document are thematically consistent.