TLDR Index

Search the archive

Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.

Results

16,678 matches

AMAZON UNVEILS NEW AI GADGETS

Amazon.com Inc. introduced an updated slate of Echo devices and pledged to bring ChatGPT-style artificial intelligence to Alexa-powered gadgets.

2023.05.18
archive.ph1 min readTLDR AI

TINY STORIES, 1 TRANSFORMER BLOCK CAN GENERATE STORIES

With models getting larger and more expensive we generally get better performance. But what happens if we go the other way? It turns out that tiny models can learn to generate coherent, novel, and grammatically correct stories. If you carefully construct the training data set and restrict the vocabulary to common words, you get some compelling completions from these miniscule models.

2023.05.17
arxiv.org25 min readTLDR AI

IMPROVING ACTION RECOGNITION: EXPLORING MULTIMODAL GENERALIZATION

This study introduces a new concept in egocentric action recognition called Multimodal Generalization (MMG), which focuses on how systems can adapt when data from certain sources is limited or missing. To investigate this, a new dataset called MMG-Ego4D was created, and new methods were developed to enhance the system's ability to generalize, potentially guiding future research in this area.

2023.05.17
arxiv.org9 min readTLDR AI

SUPERICL: MAKING BIG AI MODELS WORK BETTER WITH SMALLER ONES

SuperICL is a new technique that helps massive AI models, like GPT-3 and GPT-4, to learn better by working with smaller, locally fine-tuned models. This approach not only boosts the performance of these big AI models on tasks, but also improves their stability and the abilities of smaller models, such as understanding different languages and explaining their decisions.

2023.05.17
arxiv.org6 min readTLDR AI

REACT HOOK TO RUN LLM IN THE BROWSER

With the advent of WebGPU, you can now run hardware accelerated workflows in the browser. This repository exposes a React hook that downloads the model from the HuggingFace CDN, compiles to WASM, and runs inference.

2023.05.17
github.comGitHub repositoryTLDR AI

GUIDANCE

Guidance enables you to control modern language models more effectively and efficiently than traditional prompting or chaining.

2023.05.17
github.comGitHub repositoryTLDR AI

DIFY

One API for plugins and datasets and one interface for prompt engineering and visual operation, all for creating powerful AI applications.

2023.05.17
github.comGitHub repositoryTLDR AI

THE FIRST RNN ON THE TRANSFORMERS LIBRARY

The RWKV model is a linear attention model with built in recurrence. It is a powerful language model that runs fast with a theoretically long context window. The model development has an active discord community which now is more visible on the HuggingFace platform.

2023.05.17
huggingface.co15 min readTLDR AI

THE MOBILE AI RACE

Tech companies are racing to add AI to their mobile devices, however, high computing costs present a significant hurdle.

2023.05.17
archive.ph3 min readTLDR AI

CNET JOURNALISTS UNIONIZE

CNET journalists are pushing to unionize, seeking a formal say on issues including the use of artificial intelligence at the technology-news company.

2023.05.17
archive.ph2 min readTLDR AI

TOGETHER COMPUTE $20M SEED ROUND

Together compute is, among other things, building a cloud platform for efficient machine learning. With strong academic connections, and highly talented technologists, they have a good shot of building really useful tools for the ecosystem. One of their notable works to date is the release of the red pajama data set, which is an open replication of the closed data set used to train Llama.

2023.05.16
together.xyz2 min readTLDR AI

MEGABYTE - MILLION BYTE SEQUENCES

Transformers are not actually end to end. There is a separate training process for the tokenizer which is strange and often leads to odd performance in general. However, if we try to naively train on bytes we quickly run out of context length given the increased length of sequences. Additionally for truly multimodal problems training directly on bytes removes complex patching and tokenization schemes. This work allows models to train directly on bytes with sequence lengths up to a million bytes. Maybe this will even help nucleus sampling for rare words too!

2023.05.16
arxiv.org28 min readTLDR AI

VIDEO QUESTION ANSWERING

Visual question answering has seen massive adoption over the past several months, and rapid progress. Largely enabled by pre-trained models such as Blip and clip. Applying these models to video is challenging due to the dramatically increased computational requirements. This work introduces a cascade of different models that dramatically improves the robustness of these systems by operating at different timescales. They use Blip2 and find strong performance.

2023.05.16
arxiv.org45 min readTLDR AI

MICROSOFT GUIDANCE

Programs are how we talk to computers, prompts are how we talk to AI, and AI is now talking to computers. This means that we need a better way to talk with AI. Guidance is essentially a prompting language and toolkit developed by Microsoft that allows you to set guidelines and instructions in a way that encourages reproducible and robust prompts.

2023.05.16
github.comGitHub repositoryTLDR AI

ENHANCING 3D SCENE RECONSTRUCTION AND REALISM FROM LASER DATA

NFL is a new method that takes LiDAR data (a way of measuring distances using lasers) and creates realistic 3D scenes from new viewpoints. This technique works better than others and can help improve tasks like mapping and understanding the environment around us.

2023.05.16
research.nvidia.com5 min readTLDR AI

UNCENSORED MODELS

Eric Hartford's blog post explores the implications of uncensored AI models, suggesting that AI censorship might be leading us down a risky path. He argues for the development of systems to better manage the outputs rather than suppressing them. He mentions that more research and dialogue are needed around the ethics of AI use, and proposes the idea of a multi-stakeholder governance model for AI systems.

2023.05.16
erichartford.com10 min readTLDR AI

AMAZON PLANS TO ADD CHATGPT SEARCH

Amazon.com Inc. plans to bring ChatGPT-style product search to its web store, rivaling efforts by Microsoft Corp. and Google to weave generative artificial intelligence into their search engines.

2023.05.16
archive.ph2 min readTLDR AI

GPT4TOOLS

GPT4Tools is a centralized system that can control multiple visual foundation models.

2023.05.16
github.comGitHub repositoryTLDR AI

INSTRUCT BLIP

Blip is a model developed by Salesforce that can be used for vision and language tasks. Following the line of work in language models that tunes them to follow instructions, we can also tune these vision language models to follow instructions. This dramatically improves the performance, and in this case it even outperforms the published numbers of GPT-4.

2023.05.15
arxiv.org33 min readTLDR AI

ENHANCING IMAGE QUALITY WITH PRIOR KNOWLEDGE

A new technique has been created that enhances the quality of blurry images using pre-set text-to-image models. This method smartly uses a special encoder and doesn't need to change the existing image-making model, saving training time. Plus, users can control image quality with one simple adjustment. The strategy also handles larger images better than previous methods. Tests with both artificial and real-world images prove it's more effective than current solutions.

2023.05.15
arxiv.org16 min readTLDR AI

DETECTION THROUGH REASONING

When detecting objects in a scene, you usually pull from a set of predefined classes. Also, asking questions about the scene is challenging. In this case, we can use powerful language models (Vicuna) with instruction tuned detectors to reason about queries and detect objects as a result.

2023.05.15
github.comGitHub repositoryTLDR AI

OPEN-LLAMA

The complete training code of the open-source high-performance Llama model, including the full process from pre-training to RLHF.

2023.05.15
github.comGitHub repositoryTLDR AI

PALM

An open-source implementation of Google's PaLM models.

2023.05.15
github.comGitHub repositoryTLDR AI

COHERE LLM UNIVERSITY

Cohere is a startup that offers a language model API similar to other big players. They have a set of powerful models that you can use to build applications. This language model university is designed to bring you up to speed on modern language models and show how to build on them with cohere's tools.

2023.05.15
docs.cohere.com6 min readTLDR AI