Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
RasaGPT is the first headless LLM chatbot platform built on top of Rasa and Langchain. It is boilerplate code and a reference implementation of Rasa and Telegram utilizing an LLM library like Langchain for indexing, retrieval and context injection.
MLC LLM is a universal solution that allows any language models to be deployed natively on a diverse set of hardware backends and native applications, plus a productive framework for everyone to further optimize model performance for their own use cases.
Pandas AI is a Python library that adds generative artificial intelligence capabilities to Pandas, the popular data analysis and manipulation tool.
This article is a fun dive back into the days when IBM’s Watson was the next big AI thing, and then dives into the question of whether ChatGPT will meet the same fate of irrelevance.
This article dives into how teachers believe ChatGPT will enhance education.
Language models take text queries as input and output text responses. The length of both query and output are referred to as the context. Language models have limited context length (ChatGPT is 8k 'words'). With some cool recent algorithmic changes, MosaicML has released a model that can operate in 65k 'words'. This is enough to fit the entire novel Great Gatsby in context and have the model write an epilogue.
We live in a somewhat divisive world and it's hard to have meaningful conversations where understanding is fostered. Interestingly, if you have an AI chatbot intermediate conversations about difficult topics - both parties report improved outcomes and understanding.
Let's say we have two different models of the same architecture trained on different tasks. Is it possible to merge them into a single model that works well on both tasks? Current methods like rebasin or model soup don't work because of the presence of different task-specific features. This new work proposes a way forward with the two step process which improves as much as 60% over previous algorithms.
OpenAI has quietly been working on text to 3D with a few releases. This new work is a fun addition to that line of research. It works fairly well in standard objects or objects with a few compositional ideas. This model can also create 3D objects from 2D images, although that doesn't seem to work quite as well.
Encoding long chunks of text is a tough task for natural language processing (NLP) models. This study improves the process by using graphs and a new attention mechanism, which helps the model understand relationships between sentences and perform better on longer sequences.
Today's language models can still make silly mistakes when it comes to commonsense knowledge. VERA is a new model that estimates the plausibility of statements based on commonsense, helping to filter out incorrect information and improve performance in real-world settings.
This study introduces a new method, called MIMIC-IT, for improving how models like Otter understand and follow instructions in different situations, including images and text. By making these models more accessible, researchers can use them more easily to create better AI systems.
Sponsored
While the content of this piece is a bit sensationalized, the fact that Dr. Hinton has left Google is notable. He claimed, on Twitter, that his goal was to distance himself from Google so he could speak more openly about the potential harms of AI. He claims his previous employer acted responsibly with AI product releases.
Llama is a language model from Meta that was released under a mostly open license. It was trained for much longer than GPT-3 and even longer than was optimally suggested by Deepmind’s Chinchilla. It does ok on medical tasks, but when tuned more scores improve by 30%+. Lots of folks have been talking about Parameter Efficient Fine-Tuning which only updates a fraction of the underlying model. This paper shows that while it is cheaper to use PEFT, performance is sometimes significantly worse.
Diffusion models are a class of machine learning models that learn to slowly remove noise from data to produce high quality output. The Stable Diffusion text-to-image model is an example of such a model. One challenge with these models is setting constraints on the output (imagine an application where you only want to use blue pixels). This paper introduces an inequality constraint on the output so that you can produce diffusion paths for anything within the approved region. They showcase examples of diffusion models for biomedical applications where constraints are extremely tight.
The paper explores the potential of using large-scale text-to-image diffusion models for generative data augmentation to improve challenging discriminative tasks. The study shows that fine-tuned class-conditional models generated by these models have state-of-the-art FID, Inception Score, and Classification Accuracy Scores. Augmenting the ImageNet training set with samples from the generated models results in significant improvements in ImageNet classification accuracy over strong ResNet and Vision Transformer baselines.
The Natural Speech 2 project aims to advance the state of the art in speech recognition by developing new algorithms and models that are better equipped to handle the inherent complexities and nuances of human speech, ultimately improving the performance and usability of speech-based AI systems.
The godot-dodo project presents a pipeline to finetune open source language models on human-created, language-specific code retrieved from GitHub.
Sponsored
An open-source version of GPT.
NewsBytes is an AI generated podcast that covers global headlines in roughly 5 minutes.
Sponsored
INSTRUCTCTG is a controlled text generation framework that incorporates various constraints by using natural language descriptions and demonstrations. This flexible approach maintains generation quality and speed, and allows the model to adapt to new constraints without retraining through few-shot task generalization and in-context learning.
This study presents a method for realistically inserting people into scenes while respecting scene affordances, using a self-supervised approach and a large-scale diffusion model. The model synthesizes more realistic human appearances and natural human-scene interactions compared to previous work, and enables interactive editing.
LLaMA-Adapter V2 is an improved, parameter-efficient model that enhances large language models' ability to follow instructions and incorporate visual knowledge. By optimizing learnable parameters and incorporating expert models, it performs better in multi-modal reasoning and chat interactions.
ChatVideo is a prototype system for versatile and multimodal video understanding, using a tracklet-centric paradigm and Video Foundation Models (ViFMs) to annotate properties. This approach demonstrates effectiveness in addressing diverse video-related problems in real-world scenarios.