Welcome back to AI Insider. This week you’ll find a new feature (See DYI AI below) and the usual roundup of some highlights in the world of AI from the past week. If you like what you see, feel free to drop something in the tip jar, and don’t forget to turn your clocks back!
Let’s Try Some DIY AI
By Michelle Johnson, AI Insider
Learn to do something new in just a few minutes. On your own. This new feature, DIY AI, will get you started.
You may recall that a while back, I surveyed y’all about whether you were interested in me doing workshops on AI. Surprisingly, only a handful of you responded.
Putting together a workshop is time-consuming, so I shelved the idea. No point doing that much work if the demand isn’t there. I’m retired!
However, I still get asked: How do I do this? How do I do that?
When I was teaching, I knew those kinds of questions were a sure sign that I needed to go over something in class.
So, listen up, class. This week, we are going to take a look at using a chatbot to comparison shop. And, no, there won’t be a quiz, but feel free to share how it turned out in the comments. (Include a screenshot if you want to get fancy.)
This Week’s Assignment: Comparison Shopping with a Bot
Using the bot of your choice: ChatGPT, Bing, Bard, Claude, Perplexity, or whatever, use the following prompt to generate a comparison between Item A and Item B.
Prompt: I am shopping for [insert item here]. Please compare [Item A] and [Item B].
Example: I am shopping for a laptop. Please compare the HP XYZ and the Dell ABC.
Extra credit – Ask the bot to generate a table of the results using this prompt:
Please generate a table showing [Item A] in one column and [Item B] in the second column.
Extra, extra credit: Have it generate another table, but ask it only to include a subset of what it showed in the first table. Example: Only include rows 1 and 3. Or columns 3 and 4 only.
AI’s Role in New Beatles Song
[Ed. Note: You may have heard that a new Beatles track will be released. (Yes, really.) I asked Bard to explain how on Earth they used AI to create the new track “Now and Then.” Here’s the lowdown.]
By Bard for AI Insider
The new Beatles song “Now and Then” (video) was produced using AI to separate out John Lennon’s original vocals from a 1970s demo recording. The demo was resurfaced in the 1990s but was deemed unsalvageable due to the poor quality of the recording. However, recent advances in AI made it possible to extract Lennon’s vocals with remarkable clarity.
The AI software used in the production of “Now and Then” is called WingNut Films’ machine audio learning, or MAL. MAL is a powerful tool that can be used to isolate individual voices and instruments from complex audio recordings.
Once Lennon’s vocals had been isolated, Paul McCartney and Ringo Starr added their own instrumentation to the track. McCartney also played Lennon’s piano part, as the original recording was too damaged to be used.
The final step in the production process was to add a string arrangement by Giles Martin, son of Beatles producer George Martin.
The use of AI in the production of “Now and Then” is a significant development, as it opens up new possibilities for restoring and remastering old recordings. It also raises questions about the ethical implications of using AI to create new music from the work of deceased artists.
Biden Issues Order on AI Safety and Security
By Bard and Bing for AI Insider
President Biden issued a sweeping executive order on Monday that establishes new standards for AI safety and security aimed at protecting Americans’ privacy, advancing equity and civil rights, and promoting innovation and competition.
The executive order is guided by eight principles and priorities:
- AI must be safe and secure. The order requires that AI systems be developed and deployed in a way that minimizes risk to public safety.
- AI should promote responsible innovation, competition, and collaboration. The order directs the federal government to invest in AI research and development while also promoting fair and open competition in the AI market.
- AI should be beneficial for all Americans. The order emphasizes the importance of ensuring that AI is used to promote equity and civil rights, protect consumer privacy, and create good jobs for American workers.
- AI should be used in a transparent and accountable way. The order requires that the federal government be transparent about its use of AI and that it develop mechanisms for public accountability.
The executive order also establishes a number of specific initiatives, including:
- Creating a National AI Initiative. The order establishes a National AI Initiative to coordinate AI research and development across the federal government.
- Developing a Bill of Rights for AI.
- Establishing a National AI Research Resource. The order establishes a National AI Research Resource to provide researchers with access to computing resources and data.
- Creating a National AI Advisory Committee. The order creates a National AI Advisory Committee to provide the President with advice on AI policy.
The order builds on previous actions taken by the Biden administration, such as the voluntary commitments from 15 leading companies to drive safe, secure, and trustworthy development of AI.
Former President Barack Obama helped draft the new White House artificial intelligence policy, according to aides familiar with the situation.
The order has received mixed reactions from the AI community, with some praising it as a bold and necessary step and others questioning its feasibility and effectiveness.
For more: Factsheet issued by the White House
Study: Bots Siphoning Up Copyrighted News
By Claude for AI Insider
News publishers have maintained that AI chatbots disproportionately rely on copyrighted news content to power their systems. This week, the News Media Alliance (NMA) released a new report showing data to back up these claims.
The NMA represents over 2,200 news publishers, including heavyweights like The New York Times. In their research, they compared publicly available datasets believed to be used for training popular AI chatbots with a dataset of generic web content. The results showed that curated datasets used 5 to 100 times more news content than the generic dataset.
This suggests that AI developers specifically seek out quality news content to train their models, valuing it over other types of online text. The report also found direct examples of chatbots reproducing language from news articles verbatim in their responses.
NMA president Danielle Coffey argues this shows clear copyright infringement, as the chatbots are retaining and repurposing publishers’ content. In her view, the chatbot responses directly compete with the original articles.
Essentially, the publishers are arguing that the tech companies are profiting from the publisher’s hard work. By relying so heavily on news content, the chatbots are able to generate high-quality responses. But the publishers themselves receive no compensation.
The NMA plans to submit these findings to the US Copyright Office to bolster their case for tighter regulations. Coffey even suggested they may explore collective licensing of content to AI developers, which would require the tech companies to pay for the right to use their articles.
These developments reflect increasing tensions between news publishers and AI developers. Publishers face threats from falling traffic and potential job losses. But the technology companies argue their systems don’t actually reproduce significant portions of articles and that the data makes them better digital assistants.
The clash raises thorny questions around copyright and fair use of content in the age of AI. This report indicates that news publishers plan to aggressively defend their right to control their content.
Google’s Bard Now Responds in Real Time
By Perplexity for AI Insider
Google’s Bard, the conversational AI chatbot, has been updated with a new feature: real-time responses.
This means that users can now see the chatbot’s responses as they are being generated, rather than waiting for a complete response to pop up.
The real-time response feature is activated by default, but users who prefer the previous mode can switch back to “Respond when complete” from the settings, accessible via the cog icon in the top-right corner of Bard’s window.
In addition to real-time responses, Google has introduced a “Skip response” button. This feature allows users to cut off the bot mid-sentence if they don’t like the direction the response is taking.
Artists Lose Copyright Battle in Fight Against AI-Generated Images
By Bard for AI Insider
A US district judge has ruled that determining whether AI-generated images may be in direct violation of copyright laws is “not plausible” at the moment. This ruling is a setback for artists who have been concerned about the potential for AI to be used to create and distribute copyrighted works without their permission.
The ruling came in a class action lawsuit filed against AI-generated image service providers Stability AI, Midjourney, and DeviantArt. The lawsuit alleged that these companies were violating copyright law by using the works of artists to train their AI models and by allowing users to generate images that were substantially similar to copyrighted works.
The artists alleged that the AI models used by these service providers were trained on datasets that included their copyrighted works and that the resulting AI-generated images, therefore infringed on their copyrights.
However, the court disagreed. Judge William Orrick ruled that AI-generated images are transformative works and that, therefore, they are not protected by copyright law. Transformative works are works that are based on copyrighted material but add something new and original. In this case, the court found that the AI-generated images were transformative because they were created using a new and original creative process.
The court also found that the AI-generated images were not substantially similar to the artists’ copyrighted works. The court noted that the AI models were trained on a massive dataset of images, and therefore, the AI-generated images were likely to be a combination of many different sources. As a result, the court found that it was unlikely that any individual AI-generated image would be substantially similar to any individual copyrighted work.
What the Actual Hell?
[Ed. Note: This week’s entry isn’t a bad thing. It falls more in the “I can’t believe they pulled this off” category.]
A 10-Second Voice Test for Diabetes?
By ChatGPT for AI Insider
In a breakthrough that seems straight out of science fiction, a new AI system is making headlines for its ability to diagnose Type 2 diabetes by analyzing a person’s voice. This revolutionary technology could potentially reshape not just diabetes screening, but also offer new avenues in the broader healthcare spectrum.
The technology, as reported by Diabetes.co.uk, makes use of machine learning algorithms that scrutinize the nuances in a person’s voice to detect signs of Type 2 diabetes. The AI system is trained to pick up subtle voice features which are often imperceptible to the human ear, yet may indicate altered blood sugar levels. The entire process, astoundingly, takes just 10 seconds.
By making screening as simple as speaking into a microphone this could erase the need for a blood test. This ease of access could, in turn, lead to earlier diagnosis, better management of the disease, and overall, a significant reduction in healthcare costs.
Moreover, the seamless integration of artificial intelligence in healthcare uncovers promising vistas for other disease diagnoses as well. The underlying technology could be harnessed to develop similar diagnostic tools for a variety of health conditions, broadening the scope and reach of preventive healthcare, according to experts.
The study was published in the journal Mayo Clinic Proceedings: Digital Health.
Random Shorts
Apparently, we don’t have enough real friends on Instagram. Meta is cooking up a way for us to create AI friends. It’s coming on What’s App, too, because we need to add fake friends to our contacts.
Runway, the AI text-to-video generator, has released version 2, which, as always, is better than whatever previous version we’ve been using.
Airbnb has turned to AI to help detect and prevent rowdy house parties. Yep. Potential red flags: Booking in the same city that you live in, booking during a major event like the Superbowl, short length of stay.
Aht Gallery
Let’s hear it for the AI-generated African brothers this week!
Share DIY homework in the comments below.