TLDR Index

Search the archive

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

Results

14,500 matches

MODEL ARBITRAGE

Model arbitrage refers to the practice of leveraging pre-existing large language models to generate new synthetic datasets for training other models.

2023.03.23
matt-rickard.com1 min readTLDR

DATA DEDUPLICATION FOR THE REST OF US

There have been many papers arguing both sides of the debate around the usefulness of data deduplication. It seems, in any case, that training on the same data point 300,000+ times is wasteful. A simple deduplication technique here is to embed your data points, cluster into k (e.g., 10k clusters), then remove a subset that has really high cosine similarity with other data points. This technique seems robust to the choice of embedding and the number of clusters, as long as the clusters are high enough. It also works across modalities.

2023.03.23
arxiv.org30 min readTLDR AI

TEXT TO ROOM

There has never been a community more interested in interior design than the generative AI community currently is. This work fits in well with the current trend. There are a lot of novel tweaks they needed to do to make the entire 3D space align with the prompt. In all, very interesting work for 3D asset generation.

2023.03.23
lukashoel.github.io4 min readTLDR AI

FREEDOM: A TRAINING-FREE CONDITIONAL DIFFUSION MODEL

The article presents FreeDoM, a training-free conditional diffusion model that leverages off-the-shelf pretrained networks to construct time-independent energy functions, enabling effective generation for various conditions. The proposed method offers advantages in simplicity, effectiveness, and low cost and has a broader range of applications than existing training-free methods, as demonstrated by experiments on diverse data domains, including image and latent code domains.

2023.03.23
arxiv.org15 min readTLDR AI

CHATLLAMA

A nice wrapper interface around the instruction tuned Alpaca model. Only the 7B model is available, with others coming. While the quality doesn’t match ChatGPT it is a fun open alternative with linked code.

2023.03.23
chatllama.baseten.coToolTLDR AI

GENERATIVE MODELS FOR MEDICAL

A good step towards evaluation, creation, and curation of generative models for a wide variety of tasks with a specific focus on medical. This project aims to assist by building tools for generative medical tasks and comes with metrics and tutorials to help you get started.

2023.03.23
github.comGitHub repositoryTLDR AI

NEXT.JS CHATGPT

Next.js ChatGPT is a responsible chat application powered by GPT-4.

2023.03.23
github.comGitHub repositoryTLDR AI

THE QUEST FOR INJECTABLE BRAIN IMPLANTS HAS BEGUN

While brain implants have the potential to significantly improve patients' lives, the brain and metal do not work well together. Electrodes can cause injury and scar tissue, which in turn makes devices less effective. To avoid this problem, a team of scientists has designed a gel that can be injected into tissues and solidifies into an electrically conductive polymer. This polymer electrode is harmless and can be moved just as easily as the brain tissue around it. The gel could enable flexible electronic neural implants in the future.

2023.03.22
archive.ph5 min readTLDR

CFNCTL

Cfnctl is a tool that simplifies the process of managing stacks in AWS CloudFormation. It uses the Terraform CLI workflow to provide a familiar interface for creating CloudFormation templates. Cfnctl is still under development and some features are missing, such as support for canceling stacks and uploading large templates to S3.

2023.03.22
github.comGitHub repositoryTLDR

TYPST

Typst is a typesetting system designed to be as powerful as LaTeX and easier to use. It comes with built-in markup for common formatting tasks and flexible functions for everything else, as well as a tightly integrated scripting system. Typst supports math typesetting, bibliography management, and incremental compilation for fast compile times. It provides friendly error messages if something goes wrong. Examples are available in the repository.

2023.03.22
github.comGitHub repositoryTLDR

Sponsored

ARE YOU STUCK IN INCIDENT MANAGEMENT DEJA VU?

FireHydrant’s end-to-end incident management platform makes it easy for anyone in your organization to resolve incidents faster, communicate with stakeholders, and learn from incidents. Did we mention that FireHydrant is free? Join thousands of reliability-obsessed engineers - get started at firehydrant.com/tldr

2023.03.22
firehydrant.comSponsorTLDR

HOW TIKTOK BROKE SOCIAL MEDIA

While TikTok's fate is still being determined, the app has changed social media for good in a way that will make life much harder for incumbent social apps.

2023.03.22
archive.ph8 min readTLDR

GOOGLE LAUNCHES BARD

It's here! Google’s long anticipated entry into the language models as a service game. Bard is a lightweight version of Lambda, which was a 70B parameter conversational agent trained a few years ago. They have spent a lot of effort making things safe and grounded in truthfulness, we’ll see when people start getting added to the waitlist if it lives up to the promises!

2023.03.22
blog.google4 min readTLDR AI

GOOGLE PROPOSES SVDIFF, A PERSONALIZED TEXT-TO-IMAGE DIFFUSION MODEL

Researchers from Google propose a novel approach called SVDiff to address the limitations of existing text-to-image diffusion models for personalization, including overfitting and inefficient parameter storage. SVDiff involves fine-tuning the singular values of weight matrices, resulting in a compact and efficient parameter space that reduces the risk of overfitting, language-drifting, and has a significantly smaller model size, making it more practical for real-world applications.

2023.03.22
arxiv.org11 min readTLDR AI

ZERO-1-TO-3: ZERO-SHOT ONE IMAGE TO 3D OBJECT

Zero-1-to-3 is a framework for changing the camera viewpoint of an object from a single RGB image, using a conditional diffusion model that learns controls of relative camera viewpoint from a synthetic dataset. This approach shows strong zero-shot generalization abilities to out-of-distribution datasets and in-the-wild images, including impressionist paintings, and can be used for 3D reconstruction from a single image, outperforming state-of-the-art models by leveraging Internet-scale pre-training.

2023.03.22
arxiv.org9 min readTLDR AI

CLIP GOES 3D

CG3D proposes a framework for zero-shot 3D geometric feature extraction by training a 3D encoder using point clouds, rendered 2D images, and text. Contrastive loss aligns features in a multimodal embedding space, and prompt tuning with trainable input parameters overcomes distribution shift issues. CG3D demonstrates impressive zero-shot, open-scene understanding, and retrieval capabilities, and serves as a strong starting weight for downstream 3D recognition tasks.

2023.03.22
github.comGitHub repositoryTLDR AI

AI COMMITS

AI Commits is a CLI that writes your git commit messages for you with AI.

2023.03.22
github.comGitHub repositoryTLDR AI