AI wrap Friday July 7th

Kia ora koutou, welcome to this weeks AI wrap. 

The greatly anticipated first copyright lawsuit against OpenAI has been lodged. Two Authors, Mona Awad and Paul Themblay are filing a class action alleging their copyright protected books were used to train ChatGPT. They cite “very accurate summaries” of their work in the complaint which would only be possible if the model has been tr. riseained using their literature.

This isn’t the first lawsuit, there have been others. One I am really interested in is about the potentially existential threat of the AI. - “the of powerful AI will be either the best of worst thing ever to happen to humanity.” It warns of the potential for “civilizational collapse” and argues that OpenAI’s publicly available tools, including not just various iterations of ChatGPT but the image-creating product Dall-E, have stolen private and in some cases personally identifying information from “millions of internet users,” children among them, “without their informed consent or knowledge.”

Google sneaked a change into their terms and conditions this week - reserving the right to scrape everything you have ever posted online to build it’s AI tools. Wow. As this article says this means everything we have ever posted in the public domain is up for grabs. I’ve already clicked I accept to these new terms and conditions but does highlight some privacy and right to be forgotten issues. 

“The practice raises new and interesting privacy questions. People generally understand that public posts are public. But today, you need a new mental model of what it means to write something online. It’s no longer a question of who can see the information, but how it could be used. There’s a good chance that Bard and ChatGPT ingested your long forgotten blog posts or 15-year-old restaurant reviews.”

AI and Jobs

It seems we can’t escape the ongoing debate on AI and Jobs. 

This week on Stuff we had an article predicting the jobs that will survive, but first up talks about the jobs that will disappear eg: “Just last month, BT announced it would cut 55,000 jobs, and replace a fifth of its workers with AI.” To their predictions on the survivors. Medical specialists will be among those who survive and be the highest paid of us, oncologists, surgeons, paediatricians and the like. MP’s and CEO’s will continue to survive and be paid well too - which is kinda disappointing to hear. The other survivors according to the article are human labour intensive industries. 

This article from a UK based economist, asks the very important question - How can governments support GDP growth in a new era of persistent structural unemployment? With examples like “the global youth unemployment rate has been trending up, from 12.2% in 1995 to just under 13% percent after the 2008 global financial crisis to 15.6% percent in 2021…..AI will exacerbate these trends.”

Public servants are you using generative AI to write papers and policy advice etc? Well it seems Canadian public servants are dipping their toes in and admitting to it. I was told by a government CEO recently no-one in their agency would be stupid enough to, and have heard first hand from staff who work for other agencies who have banned ChatGPT at work that they still use it on their own computers. Don’t be naive government leaders, this technology can be a productivity enabler, you just need a set of guidelines in place so it is used for research but not for the writing of policy, regulations and legislation. 

ChatGPT for writing your copy

There are plenty of blogs and articles out there now suggesting ways we can use ChatGPT and the like to make our jobs easier. This one on using ChatGPT to write not-for-profit fundraising applications peaked my interest as this is something I do all the time. So using the free version I asked ChatGPT to write me an application for something quite specific, here is what I found. 

• Your results are only as good as the instructions you provide so add context, detail that must be included and you will get a better outcome - you can do this by saying "rewrite and add...."
• The standard output is very American, with simple language and sentence construction
• I asked ChatGPT to restructure the format quite a few times before I got the flow I was looking for
• In the end, I used it as a guide, cherry picking out the useful phrases and writing others - a starting point rather than the completed product

In brief

• New team at OpenAI - SuperIntelligence - to help solve many of the worlds most important problems
• Why transformative artificial intelligence is really hard to achieve - it’s a long read
• Paper on training the personalities of large language models - haven’t read this yet 
• AI’s trained on the output of other AI’s will eventually spiral into gibberish, according to experts. “We are about to fill the internet with blah.” - kinda logical really
• Info-Tech Research have getting started with AI resources, a getting started PDF and a Blueprint you can download. 
• The AI trained to recognise waste for recycling 
• Towards Measuring the Representation of Subjective Global Opinions in Language Models
• Can the USA restrict China’s access to cloud computing to run their AI? 
• Valve responds to claims it has banned AI-generated games from Steam

New products

I haven’t tried any of these this week. 

ChatCSV - up load your CSV file and start asking questions

Magical - the calendar that takes meeting notes for you

Sanebox - to manage your emails

MyReader - let AI read your books for you and then you can query it for insights

Image

This weeks image was generated using FreePik. I asked for an office with orange walls, people at desks and a lovely outlook out the windows. 

Enjoy, Vic