Published February 27, 2024 by GWW Staff
The GovWebworks AI Lab's February 2024 newsletter covers AI assistants, semantic search, AI cyber defense, generative video and music, and AI smart glasses — with implications for public sector digital services.
Welcome to the February 2024 AI Newsletter from the GovWebworks AI Lab. The following articles touch on important topics relating to public sector online services:
Feel free to share with your colleagues and encourage them to sign up for the AI Newsletter and join the conversation! Or schedule a free consultation (opens in a new tab) to talk about using AI to optimize your organization’s digital goals.
#1: AI assistants are everywhere
Key takeaway: ChatGPT and other apps are adding more AI assistants for increased personalization.
Reviewed by GWW Staff
ChatGPT is increasing its personalization capabilities with a new Memory feature. The LLM assistant can retain information about you or your agency and learn topics of interest to increase depth and personalization on future chats. For example, you can share your or your organization’s demographics and preferences and these will be incorporated into the responses to your prompts.
**
Large Language Model Applications for Government
AI Lab update on the benefits, risks, and emergent guidelines for LLMs in the public sector
Lots of applications, such as Slack, the popular collaboration app, have added their own AI assistants. Slack’s can summarize long threads and unreads, and has a conversational search option for finding previous threads and summarizing information about projects. Slack claims that these LLMs are hosted in Slack so “customer data remains siloed” and can’t “be used to serve other clients.” Currently the assistant is only available in US and UK English.
Local assistants are also taking off with NVIDIA’s RTX, which makes it possible to chat with your own files and documents and incorporate that info into replies. Developers can use the NVIDIA AI Workbench to create, test, and customize pretrained generative AI models and LLMs, using repositories like Hugging Face and GitHub for collaboration on projects. However, RTX is not available for Mac.
Hugging Face, which hosts open source AI code and frameworks, has also launched its own third-party Hugging Chat Assistants that can be customized to specific needs.
Resources:
- Open AI: Memory and new controls for ChatGPT (opens in a new tab)
- The Verge: Slack AI is here (opens in a new tab)
- NVIDIA News: NVIDIA Chat with RTX (opens in a new tab)
- Hugging Face Third Party Customizable Chat Assistants (opens in a new tab)
#2: OpenAI announces new semantic search option
Key takeaway: Semantic/generative search is making headway with new collaborations.
Reviewed by GWW Staff
Forbes reports that in the week since partnering with OpenAI, Bing’s semantic generative search was downloaded almost as much as all of last year. It’s unclear yet what adoption rates will be (Bing is currently at only 3 percent to Google’s 93 percent of the market, according to the article) but now is the time for agencies to investigate semantic/generative search if they want to stay current.
Semantic search allows a user to ask specific questions (“what is the best way to make ranch dressing?”) rather than entering keywords (ranch dressing recipe), and generative search provides answers to the questions (here’s the best recipe for ranch dressing) rather than links to recipe websites. Semantic search also allows for conversational input to refine the search. The hitch is that search engines rely on breaking news and information and LLMs are currently only trained with information up to December 2023. However, if Google search results continue to become less reliable due to spam and poor websites, generative search results, which can be specific and personalized, (though sometimes incorrect), may become more desirable.
Resources:
- Since Adding ChatGPT, Bing’s App Grew Almost As Much As All Last Year (opens in a new tab)
- OpenAI’s Web Search Product Aims to Challenge Google’s Search Supremacy (opens in a new tab)
#3: AI can scale security incident response and fix vulnerabilities
Key takeaway: Large organizations are starting to utilize AI to help defend against security attacks and reverse Defender’s Dilemma.
Reviewed by GWW Staff
Defender’s Dilemma refers to the cyber security challenge that hackers only need one moment of lapse to break through a secure system, but defenders need to be on defense at all times. The hope is that AI, which can run at all times, poses a solution.
Google, for instance, has been investing billions to improve security by using AI to scale up the right defenses at the right times. Google says it aims to make security innovations available to public sector organizations and businesses and has put $5 billion into securing data centers over the past five years.
Resources:
#4: Generative AI expands to videos and music
Key takeaway: LLMs can now create original songs and videos from prompts.
Reviewed by GWW Staff
OpenAI’s latest reveal is called Sora, an LLM model that produces short, realistic videos from text prompts, but the tool is currently under testing for vulnerabilities before being made public. Sora can create videos based on a simple prompt such as, “an elderly couple playing chess” or “a family with young children walking down a street in New York.” As usual for the cutting edge company, OpenAI’s announcement has generated a buzz of competition from others in the space.
Meanwhile, Stable Audio is generating custom music and high-quality sounds from specific prompt descriptions. The prompts can be as simple as “the sound of wind in trees” to “write a song incorporating hip hop and Taylor Swift lyrics” or whatever else you might imagine. The results can be used in commercial projects and will likely be used in Sora-style videos.
Resources:
- OpenAI has unveiled AI that instantly makes videos (opens in a new tab)
- Stable Audio (opens in a new tab)
- Stable Audio Samples (opens in a new tab)
#5: AI smart glasses capture video and translate speech
Key takeaway: Still to be seen (pun intended) if the interface will be widely adopted.
Reviewed by GWW Staff
The latest new AI toy comes in the form of a fashionable accessory, yes, eye glasses incorporating a multimodal AI assistant that can translate speech, search the web, take photos, and record audio and video.
Brilliant’s model, which are round and look like Gandhi glasses, cost $350, but the open source design files, code, and documentation are all on GitHub for anyone to use.
Meta’s version has been redesigned by Ray Ban and EssilorLuxottica and costs $299. Currently only available for preorder until October, Meta claims they are an upgrade from the previous model with better audio and cameras, and a lighter more ergonomic frame. Through the glasses, you can take photos and livestream video (from meetings or on the street) and post to Facebook and Instagram.
Resources:
- For $350, you too can have glasses with ‘AI superpowers’ (and look like a modern-day Gandhi) (opens in a new tab)
- Brilliant Fully Open Source Frame AI Glasses (opens in a new tab)
- Ray Ban Meta Smart Glasses (opens in a new tab)
Learn more
- Sign up for the AI Newsletter for a roundup of the latest AI-related articles and news delivered to your inbox
- Find out more about using AI to optimize your organization’s digital goals
Explore other topics
