Latest News

.: AI Updates & Developments

​

.: Teaching with AI ➜ OpenAI has published a guide for teachers using ChatGPT in the classroom. Coinciding with the return to school in much of the Northern Hemisphere, the guide includes educator prompts and information about bias, limitations and how it works.

.: Google’s Duet AI now available in Docs, Gmail, and other Workspace apps ➜ Google has launched its Duet AI assistant for its Workspace apps, such as Gmail, Drive, Docs, and more. Duet is a collection of features that can help users with various tasks, such as creating slides, charts, images, summaries, and more.

.: Baidu launches Ernie chatbot after Chinese government approval ➜ Baidu’s chatbot, Ernie Bot, is now available for download after getting government approval. Baidu and other AI companies had to comply with China’s generative AI guidelines, which require adherence to the core values of socialism and legitimate data sources.

.: Pass AI law soon or risk falling behind, UK MPs warn ➜ The UK Commons Technology Committee warns that the UK could fall behind the EU in regulating AI unless it introduces a new law in the King’s Speech on 7 November. Their report identifies 12 challenges that need to be addressed, such as bias, privacy, employment, copyright and misinformation.

.: Introducing ChatGPT Enterprise ➜ A new version of OpenAI’s ChatGPT tool offers enterprise-grade security and privacy, unlimited higher-speed GPT-4 access, longer context windows for processing longer inputs, advanced data analysis capabilities, customisation options, and much more.

.: Google launches watermarks for AI-generated images ➜ A new technology developed by Google’s DeepMind unit creates invisible, permanent watermarks on images that identify them as AI-generated. The technology, called SynthID, embeds the watermark directly into images created by Imagen, one of Google’s latest text-to-image generators. The watermark is imperceptible to the human eye but detectable by computers. The technology is meant to help prevent the spread of misinformation and fake images online.

Reflection

.: Why this news matters for education

​

One of the key announcements from Google is the broader integration of AI capability across their workspace tools. Their version of Microsoft’s Co-pilot idea is Duet.

I want to frame my concern about consent when using AI tools in online meetings. But first, a little bit more context for you.

Alongside Gmail, Docs and Sheets, there has been some news about the AI capabilities coming to Google Meet, the video conferencing tool.

Through other tools, such as Zoom’s IQ, Otter.ai and Fireflies, we have had access to an AI agent joining a call, transcribing the meeting and generating summary notes. You may have experienced these in one of your meetings.

Google’s announcement brings this toolset to a new mainstream user audience via Google Meet. Although Duet for Docs, Gmail and Sheets is available to Workspace users (I have access to a short trial), the notetaking agent in Meet is not widely available. But it looks like it will be rolling out soon.

But this commentary is not about feature announcements; I share this related to informed consent within meetings. Let me know if you have been in this situation already:

This leaves me in a quandary. Where does the data go? What analysis is being done? How does this change the dialogic dynamic? How can someone approve without more information? Doorstepping someone in a meeting and putting them on the spot is inappropriate.

A key issue is a lack of information and visibility to the data flow, storage and analytics that occur with these tools. We must be watchful that big tech like Google does not equal a default level of trust, even if every email we send is from their free tool.

I am still working through how to incorporate these AI agents into my meetings because they are beneficial. But we are running headlong into adopting great utilities without asking essential questions and demanding accountability and visibility.

Fireflies offer a set of conversational analytics for your calls, telling me the proportion of time spoken by participants, key topics and sentiment overview. Don’t be fooled; it is often more than just taking notes. What happens when the analysis is wrong? Or does it reveal something unforeseen?

If I want to invite others to embrace these tools, I need to offer more transparent information about how they work and what people agree to. I think this starts with us, the users, and a shared responsibility with the creators of these tools to facilitate understanding.

.:

​

~ Tom

Prompts

.: Refine your promptcraft

​

A recent video from Anthropic, the research lab behind the Claude large language model, illustrated a helpful little prompt you can add quickly to improve the richness of chatbot responses.

Their idea is based on the concept of Chain-of-thought (CoT) reasoning, which has been shown to deliver better responses.

Anthropic illustrated this by including <thinking> tags to encourage planning and intermediate reasoning.

Here is the prompt to try – replace the bracketed question with you own.

Of course, the Chatbot is not thinking, but it works well and increases visibility on the steps taken to achieve an output. Give it a try, and let me know what you think.

.:

Remember to make this your own, tinker and evaluate the completions.

Learning

.: Boost your AI Literacy

​

.: Introduction to Deep Learning | MIT​

“MIT’s introductory program on deep learning methods with applications to computer vision, natural language processing, biology, and more! Students will gain foundational knowledge of deep learning algorithms and get practical experience in building neural networks in TensorFlow.”

.: I used ChatGPT to rewrite my text in the style of Shakespeare, C3PO, and Harry Potter | ZDNET​

The author explores how ChatGPT can rewrite a paragraph in different styles, such as Shakespeare, Poe, C3PO, etc. The article suggests possible uses for AI’s rewriting capability, such as generating dialogue for characters, normalising writing styles for reports or marketing, and simplifying complex writing for a broader audience.

.: ChatGPT and LLMs: What’s the risk? – NCSC.GOV.UK​

This guide from the UK’s National Cyber Security Centre outlines key risks of ChatGPT and large language models that are important to consider as these AI systems advance. It notes they may inadvertently generate harmful, biased, or misleading content if not properly monitored. Additional concerns include the potential for criminal misuse, impacts on intellectual property, and the need for ethics and diversity in training data.

Ethics

.: Provocations for Balance

​

China’s AI developments highlight the geopolitics of AI. Chatbots, like Ernie from Baidu, are trained on verified data sources, controlled by Chinese companies, governed by Chinese laws and require adherence to the core values of socialism.

This scenario brings to the fore compelling questions about data sovereignty. How will this reshape the global AI landscape, especially considering the traditional dominance of tech giants from the United States?

What happens when we see a variety of nationally controlled AIs, each reflecting the values, norms, and laws of their respective countries? Will there be a time when we look for the “Made in…” label on our AI tools? What might be the global implications of such a trend?

~ Inspired by this week’s developments.