Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
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.
While all the rage lately has been L2s, the team behind Lens, a decentralized social media protocol, has started beta-testing Momoka, a Layer 3 scaling solution. The Ethereum ecosystem is shifting towards Layer 3 solutions, otherwise known as AppChains, to further scale the base chain. Lens' Momoka is an Optimistic Layer 3 data availability layer, helping to reduce on-chain storage requirements and costs. It processes Polygon transactions off-chain and stores them on a data availability layer. Scaling solutions like this are going to be necessary to onboard the masses to crypto, especially when web3 games and social media platforms begin to gain traction.
Bhutan has apparently been utilizing its hydroelectricity reserves to power a state-owned Bitcoin mining operation. The project, which began a few years ago, remained undisclosed until recently. This comes to light as Bhutan has been in talks with Nasdaq-listed Bitdeer to build a 100 MW Bitcoin mining data center in the country and only a couple weeks after reports that the country’s alleged sovereign wealth fund had been investing in crypto. Bhutan's growing involvement in the crypto space has caused concern among international supporters, especially as it seemed to keep its involvement in the space under wraps.
GitHub.com has been a Ruby on Rails monolith since it’s creation. Today, the application is nearly two million lines of code and more than 1,000 engineers collaborate on it daily. A unique aspect of GitHub’s Ruby on Rails infrastructure is that they update their Rails systems every week and test Ruby continuously.
This is a tutorial for rebuilding the featured news section on Vox.com using modern CSS. It analyzes the news section, builds the main layout, and contains examples of creating a component library for cards. The tutorial uses advanced CSS concepts, such as container queries and text wrap balancing.
A URL is made of different parts: the scheme, host, path, port, query, and fragment. However, an issue with URLs is that it’s not clear what is and isn’t allowed in each part of the URL. The article continues to go over risks with URLs and how to protect software systems from outside URLs.
There is a structured approach that you can take to increase your chances of a promotion. Climbing up the career ladder as a software developer is often easier than other professions, especially at the lower levels. You don't have to wait for someone to leave to get promoted, it's based on a combination of skill and experience. This article goes over a variety of tips that can help you get promoted as a software engineer.
Software estimation is the bane of many developers’ existence and the scourge of project managers everywhere. But estimations serve a crucial purpose in the realm of software development. Without them, no software project would be green-lighted. There are some rules of thumb for software estimations, such as communicating often, factoring in uncertainty, and acknowledging that no two projects are the same. While estimates can never be perfect, the article explains ways to refine estimates to the benefit of the team and stakeholders.
There’s strong evidence that AI will surpass the ability of human coders. However, rather than eliminate coders, the emergence of this AI tech is more likely to turn any and all of us into coders. Technology has always allowed individuals to do more, faster. Researchers at Microsoft have found that software developers using AI assistants completed tasks 55% faster than those not using AI assistants. An MIT study showed that the top 5% of programmers performed orders of magnitude better while partnering with AI. Now and for the near future, the best coders are humans working with AIs.
Vercel now offers a suite of serverless storage solutions: Vercel KV, Vercel Postgres, and Vercel Blob. Vercel KV is a serverless Redis solution that's easy and durable. Vercel Postgres is a serverless SQL database built for the frontend. Vercel Blob is a solution to upload and serve files at the edge.
Whisper.cpp is a port of OpenAI's Whisper model (automatic speech recognition) in C/C++, which allows Whisper to be run locally and offline. The new 1.4 release adds integer quantization and partial GPU (NVIDIA) support, leading to performance increases.
Cube is the semantic layer for building data applications. It helps data engineers and application developers access data from modern data stores, organize it into consistent definitions, and deliver it to every application.
UC San Diego Health, UW Health, and Stanford Health Care are testing a tool that uses OpenAI's GPT to read patient messages and draft responses from their doctors. The pilot program aims to see if AI can cut the time medical staff spend responding to patients' online queries. Early responses required heavy editing. The team is working to improve replies and to add a kind bedside manner. Preliminary data suggests that AI could add value as it scored better than real doctors at responding to patient queries posted online.
Lamini is an LLM engine for rapidly customizing models. This repository contains a 71K dataset of instructions for finetuning instruction-following LLMs, Lamini's data generator, and fine-tuned LLMs. All data and LLMs are under a license that allows commercial use.
Pandas AI is a Python library that integrates generative artificial intelligence capabilities into Pandas. It allows users to ask questions about their data and get answers back in the form of Pandas DataFrames. A browser demo is available.
Salesforce's run of success took a turn after its co-CEO resigned and the company missed internal sales goals.