📡🧠 This is how AI will overtake humanity. Drugs in space? It’s not what you think (maybe). What is AI model collapse and why should we care?
Happy July, Bizarro readers.
I hope this finds you well. This month we’ve got some more interesting AI news to share, including a piece from yours truly about how AI might eventually take over our societies. If this is something you’ve given any thought to, I’d love to hear your reaction and what your own opinion on the matter is. In addition to that, we’re going to take a look at how gravity effects drug development and the potential looming problem of AI model collapse.
Here’s the breakdown:
📡🧠 This is how AI will overtake humanity.
💊🧑🚀 Drugs in space? It’s not what you think (maybe).
🤖💣 What is AI model collapse and why should we care?
Let's get into it!
📰 From the Newsroom
In an original thought piece, I argue that an AI takeover of society is very possible in the future. While some might agree and others might think it’s farfetched, I think that even those in the latter camp would have to concede that it is at least plausible. Here is an abbreviated list of my reasons why:
Humans have been gradually giving away more and more of our thought processes and decision making over to computer systems and with the growth of LLM capability and its widespread adoption, that is only going to accelerate.
Biotech and augmented human-machine hybrids are eventually going to go from sci-fi to reality and it won’t be so strange to see humanoid hybrids walking the streets among 100% ordinary humans.
With each successive generation, humanity becomes desensitized to increasingly more sophisticated technology. What seems outside the realm of possibility now, won’t be in the future. It’s the same way that a modern-day mobile phone would have blown the mind of a person living in the 1700s. Future generations aren’t going to have as much innate resistance to AI being in charge as many of us have now because of this gradual desensitization.
When you combine these factors and others that I go into in the article, you wind up with a set of conditions that lend itself well to a complete AI takeover. Let me know what you think about it and if you’ve ever considered some of the things I mention. I’d love to hear your opinion.
SpaceX has successfully launched the world's first space factory, created by California startup Varda Space Industries, with the aim of manufacturing pharmaceuticals and other products in microgravity. It was inspired by experiments conducted by the pharmaceutical giant, Merck.
Merck conducted an experiment on the International Space Station (ISS) in 2019, revealing a method to enhance the stability of its cancer drug Keytruda, enabling administration through shots rather than IV infusions.
The behavior of particles in microgravity played a crucial role in this discovery, as observing the drug's crystallization in space helped Merck refine the manufacturing process on Earth.
Microgravity research has the potential to uncover more breakthroughs and even facilitate the development of new medications, but limited time on the ISS restricts commercial experiments. Access to microgravity outside of orbit has been currently limited to parabolic airplane flights and drop towers, which provide only brief moments of free fall.
The launch of Varda’s space factory marks a significant milestone in leveraging microgravity for manufacturing purposes. By exploring the behavior of particles in space, Varda aims to produce pharmaceuticals with improved stability and potentially discover new drugs. This development opens up possibilities for advancements in various industries and invites us to contemplate the potential of manufacturing beyond Earth's boundaries.
Two new papers highlight the consequences of training AI models, including Large Language Models (LLMs) and diffusion models, on data generated by the models themselves. This might not seem like a problem at the moment, but as the models run out of human data to train on, they could self-implode from self-reliance.
The research shows that errors such as “hallucinations” as they are commonly referred to nowadays, could accumulate over time. This could lead to the domination of errors rather than original data, ultimately causing model collapse.
Even early generations of models showed deteriorating quality, with responses becoming nonsensical and unusable within a few generations of training on AI-generated data.
The problem of model collapse poses a significant challenge for AI researchers and raises concerns about the integrity and reliability of AI-generated content on the Internet.
This issue highlights the need for careful curation of training data and the consideration of AI-generated data as a potential cybersecurity threat. The development of effective watermarking techniques and improved data collection methods may offer some hope in addressing this problem. However, without appropriate measures, early models may have an advantage over later models due to access to untainted datasets. What do you think? 🤔
⛓️ Ten Must See Links of the Month
Pope Francis has officially decided to get in on the AI craze - and no, we’re not talking about the viral AI generated photos of him wearing a puffy winter jacket with an iced out necklace.
A website called the Enlightened Mindset currently ranks in the first position for roughly 700 keywords as of the time of this writing (we manually checked in Ahrefs). Why is that so bizarro? Well, here’s the kicker: the entire site of more than 1,000 articles is 100% AI generated.
Related to the above, according to a report by NewsGuard, ads for over 140 major brands are running on low-quality websites that are filled with content produced by bots. The ads are being automatically placed via Google ads. In other words, people are making money off of garbage AI generated content.
More than 100 artists, including Tom Morello and Zack de la Rocha, co-founders of Rage Against the Machine, as well as Boots Riley and Speedy Ortiz, have declared their intention to boycott concert venues employing facial recognition technology.
Harvard University plans to use an AI teaching assistant to grade assignments in its popular introductory computer science course, CS50, starting this fall. The AI will provide feedback to students, freeing up teaching assistants to focus on more complex issues.
The first drug discovered and designed with generative AI has entered Phase II trials, with the first patients dosed. The drug, INS018_055, was designed using AI technology from Insilico Medicine.
Scientists have successfully cryopreserved rat organs, thawed them, and transplanted them, which could revolutionize transplant medicine and potentially save tens of thousands of lives each year.
📽️🎞️ Do you happen to be employed as an orchestra conductor in Seoul? If so, then AI is definitely coming for your job. According to a rep from the Korean Chamber Orchestra, their “new hire” demonstrates that humans and robots can co-exist and complement each other.
If you’ve been wondering how to get started with editing block themes using JSON, then check out this JSON beginner’s guide. It’s specifically geared toward WordPress users and block themes.
While we’re on the subject of coding and WordPress, if you are tired of your code editor but aren’t sure which other one to switch to, then you need to take a look at this roundup of the best code editors for WordPress.
🎤 It’s How They Said It
"This idea of surpassing human ability is silly because it’s made of human abilities. It’s like saying a car can go faster than a human runner. Of course it can, and yet we don’t say that the car has become a better runner."
🧮 The Numbers Game
$120 USD is how much a retired computer designer is selling 2,200 brand new 1980s NABU personal computers on eBay. He even has them in their original packaging. For historical tech aficionados, these vintage computers are equipped with the legendary Z80 processor found in iconic 80s computers like the TRS-80 and Sinclair ZX80.
$1,199 USD is the cost to buy a physical set of encyclopedias. It is the last remaining general reference encyclopedia set still published today (at least in English). In comparison, the online edition of the same encyclopedia is a mere $250.
$12,000,000 SEK (about $1,112,429 USD) is how much it will cost Sweden to build the world's largest wooden city, Stockholm Wood City. It will cover 250,000 square meters and have 7,000 new offices and 2,000 new housing units - all made from wood.
⚒️ Tools and Resources
TL;DV: An AI-powered meeting recorder that can summarize key moments during your meetings with a simple click or shortcut. It provides an automatically generated AI summary of the meeting, including an overview of everything discussed as well as key action points. The tool is available for Zoom and Google Meet and has both a forever-free plan and a (paid) pro version.
ChatGPT Prompt Cheatsheet: This free resource includes prompts and tips for using ChatGPT for various tasks, such as coding, email, spreadsheets, social media, reading, design, and analyzing. It also includes links to additional resources for learning how to write effective ChatGPT prompts and for using ChatGPT to summarize articles.
🖼️ Image of The Month
A Belgian construction company has made clever marketing use of the moment, by reminding job seekers that there are some blue collar jobs that have job security against any AI takeover.
📊 Results of Last Month’s Polls
Although last month’s polls showed a justifiable concerned sentiment about the direction we’re heading in, I do have a bit of good news to share. While 60% of you answered that you think AI could replace you at your job in the near future, the story that this poll was attached to, ended up doing a complete 180.
If you recall, it was about the largest nonprofit organization dedicated to helping people with eating disorders in the U.S. The organization, NEDA, fired their human helpline staff and replaced them with an AI-powered chatbot. Shortly after we released last month’s issue with that story, NEDA had to stop using the chatbot. Unsurprisingly, the chatbot ended up giving out bad advice.
If this is the first time you're here at Bizarro Devs and you liked what you read, then sign up here to get all future editions.
Until next month,