In March, the large tongue models seemed to suck all the air out of the room. There were so many announcements and requests to join, new queues, that it was hard to find news about other important technologies. Those technologies still exist and are still developing. There is a world beyond AI.
One important change in the past month. The new US cybersecurity strategy shifts responsibility from customers to software and service providers. If something bad happens, it’s no longer (entirely) your fault. vendors must create more secure software and services. Using memory-safe languages, particularly Rust, as well as older languages like Java and newer contenders like Zig, will help software become safer.
Learn faster. Dig deeper. See further.
- According to Simon Willison, gpt4All is the easiest way to get a large AI (small) model running on a laptop. It is the basic LLaMA model with further training on 800,000 questions and answers generated by GPT-3.5.
- Hugging Face has created a tool called Fair Diffusion to bias images produced by generative graphics tools. With minimal image changes, Fair Diffusion changes gender and ethnic characteristics to reflect the diversity of the population. A similar technique is suggested to work for language models.
- Databricks has released Dolly, a small big language model (6B parameters). Dolly is important as an exercise in democratization. it is based on an older model (EleutherAI’s GPT-J) and requires only half an hour of training per machine.
- ChatGPT has announced a plugin API. Plugins allow ChatGPT to call developer-defined APIs. These APIs can be used to retrieve data and perform actions for users. Unauthorized plugins became available almost immediately for hate speech and crypto price discovery.
- A quick and sobering guide to cloning yourself. Yes, you can. Start with ChatGPT, add a text-to-speech service that dubs your voice, and a service that creates a video from a still photo, and you’re there.
- Fast engineering, the technique of crafting hints that make a language model produce exactly the output you want, is a new subdiscipline in computer science. Here is a good summary of rapid engineering techniques.
- Modeling poor drivers significantly reduces training time for AI systems for autonomous vehicles. Simulation can quickly produce dangerous scenarios that rarely occur in real life.
- Google has opened a waiting list for its Bard chat app based on Google’s LaMDA language model. Unlike ChatGPT and GPT-4, Bard has access to web information. It doesn’t replace a search, although it will generate links to Google searches along with its answer.
- The Stanford Alpaca 7B model, a clone of LLaMA 7B, was partially trained on the output from ChatGPT, significantly reducing the training cost. The total cost of the training was less than $600.
- Glaze is a free tool for “masking” digital artwork. It changes images in a way that humans cannot detect, but it makes it difficult for the generative model to copy the work.
- Baidu introduced Ernie Bot, a multi-modal broad-language model and chat that should be similar to GPT-4. So far, reviewers aren’t impressed.
- Microsoft has announced that it will build ChatGPT-like features into its Office365 products (Word, PowerPoint, Excel and Outlook).
- Google has announced that it is building generative AI into every product. It also makes the API for its PaLM model publicly available.
- GPT-4 was released on Pi-Day with limited public availability; Chat access for ChatGPT + subscribers, API access waiting list. The most notable change is that it will be able to work with images, although this is not initially supported. Errors are still a problem, although they are less common.
- A Stanford research group has released Alpaca, a version of Facebook/Meta’s LLaMA 7B model, tuned to run on smaller systems. They will release the weights when they get permission from Meta.
- llama.cpp is a port of Facebook’s LLaMA 7B model to C++. It runs OS X (maybe just Apple Silicon). The author is working on larger models. Dalai is an NPM-based tool that automates downloading, building, and running llama.cpp. There is information about llama.cpp running on Windows, Android phones, and even Raspberry Pi.
- Writeout is a free audio transcription and translation service powered by the Whisper language model. Whisper was developed by OpenAI and is closely related to the GPT series of major language models.
- How can we design programming languages that can be easily generated by automated tools? An important question in the age of AI.
- The Romanian government has deployed an artificial intelligence “consultant” in the cabinet, which summarizes the comments of citizens. Romanians can submit comments through the website or social media using a special tag.
- Andrew Ng writes that economic incentives will prevent “watermarking,” in which generative AI systems add data to their product to prove it was created by AI, from being effective.
- Google has released an update to their Universal Speech Model as part of their 1000 Languages project. Their goal is to create a single model for the world’s 1,000 most common languages, many of which have a limited number of speakers.
- Someone has developed a StableDiffusion plugin for Photoshop. It is open source and available on GitHub.
- To keep up with Microsoft’s Kosmos, Google has announced Palm-E, an “embodied” language model that incorporates visual and other sensory inputs and is built into robots.
- Microsoft is incorporating conversational AI into its productivity tools, including its PowerPlatform and Dynamics 365, where it can perform tasks such as summarizing a website and composing responses to customer inquiries.
- Microsoft has created a Multimodal Large Language Model called Kosmos-1. Kosmos-1 is a language model also trained on images. It is able to solve visual puzzles and analyze the content of images while using human language; you can ask it about visual objects.
- Microsoft has created an experimental framework for controlling bots with ChatGPT. ChatGPT converts natural language commands into code, which is then reviewed by a human and uploaded to a computer. Robotics aside, this could be a preview of the future of programming.
- A judge in Cartagena, Colombia used ChatGPT as an aid in deciding a court case, including GPT’s full responses in the decision.
- The US FTC says companies selling artificial intelligence products must be careful that the claims they make about those products are accurate.
- The Zig programming language is worth a look. It’s a simple imperative, memory-safe language designed to compete with C, C++, and Rust. It has a long way to go before catching up to Rust (let alone C++), but it’s starting to gain traction.
- GitHub has announced Copilot X, its vision for the next generation of Copilot. Copilot will include a voice interface, the ability to explain code (based on GPT-4), add comments, answer questions about the documentation, and even explain Git pull requests.
- Slim.ai has a service that optimizes containers by throwing out everything that is not needed by the application. As Kelsey Hightower said, the best software is the software you don’t ship.
- Will WebAssembly become a general purpose programming tool? One area where it can fit is serverless. Minimal startup time, secure sandboxing, and cross-platform support are all desirable for serverless applications.
- Miller is a tool conceptually similar to sed, awk, and other Unix command-line utilities, except that it was designed to work with CSV, TSV, and JSON files.
- GitHub now requires the use of 2-factor authentication (2FA).
- The PostgreSQL database has long been recognized as the best of the open source databases, but its popularity has always lagged behind MySQL. According to a StackOverflow poll, it’s finally getting the attention it deserves.
- Rust was designed as a “memory safe” language and has probably the strongest memory safety guarantee of any widely used language. Here’s a post that shows what “memory safety” means.
- 8th Light has published a short series (and video) discussing what developers need to know about data regulation.
- Developed for Oak Ridge National Laboratory, the Evasive.AI platform generates malware samples along with training data that security systems will need to detect and quarantine the malware.
- Microsoft Exchange Online will begin delaying and blocking email messages from Exchange servers that are no longer supported or patched.
- VEX (Vulnerability Report Data Exchange) is a new machine-readable standard for reporting software vulnerabilities. It is intended for use with software bills.
- The US has published its national cyber security strategy. Its main points are that it shifts responsibility from end users to software and service providers and emphasizes the importance of long-term investment. The Lawfare blog provides an excellent summary.
- Phishing remains an important attack vector, with a voice call being used as a follow-up to a fake email about a service or fee.
The Web and the Metaverse
- Beauty filters are nothing new on social media. But the latest hyper-realistic beauty filters are almost unnoticeable, even on video (like TikTok). Regardless of the consequences, they will inevitably become part of the AR-enhanced metaversion.
- Lidar has become much cheaper and is now cheap enough to be integrated into consumer devices (including the iPhone 12). It enables many interesting projects, from building 3D worlds to supporting cities in Ukraine that may be destroyed by bombing.
- Web Fingerprinting is a technique to identify and track users based solely on the characteristics of the browser and computer they are using. It doesn’t require cookies, it doesn’t affect VPNs or even Tor. And it’s available “as a service.”
- Google has started limited implementation of client-side encryption for Gmail and Calendar.
- A more sophisticated version of LIDAR can better understand pedestrian behavior and its relationship with motor traffic.
- An autonomous robot was created to measure the angles of the leaves of corn plants. Measuring leaf angles is important because it indicates how efficient plants are at photosynthesis.
- More than 200 people have been treated with experimental gene therapy using CRISPR. While these treatments are effective in treating incurable diseases, they raise questions about the cost, which can easily run into the millions of dollars.