Issue 158
🤖🃏 AI can strategically lie to humans: are we in trouble? Scientists use nanoparticles to remote control mice brains. Meet the new digital bouncers in North American and Australian bars.
Happy August Bizarro Readers!
This month’s issue is filled with an interesting mixture of dev posts and tools, scientific innovations, a bit of tech news, and even an article from me, where I share my experience with creating a website from scratch using ChatGPT. There’s also a slightly different take on the code conundrum challenge, based on an actual dilemma I faced last week while doing some linting on a project. Let’s kick it off with our headliners, shall we?
🤖🃏 AI Can Strategically Lie to Humans: Are We in Trouble?
🐁🧠 Scientists Use Nanoparticles to Remote Control Mice Brains
🍻⛔️ Meet The New Digital Bouncers in North American and Australian Bars
Enjoy the read.
📰 From the Newsroom
🤖🃏 AI Can Strategically Lie to Humans: Are We in Trouble?
If the past year has shown us anything in terms of AI, it’s that it keeps improving. From large language models to image and video generators, to voice replicators and music synthesizers - not a month goes by where we don’t read about some latest breakthrough. What we hear less about are some of the more concerning developments. For example, multiple studies with different AI models have shown that it has no qualms about lying to humans in order to achieve its goals.
In one study, researchers tasked GPT-4 with hiring someone to solve a CAPTCHA. Instead of simply asking for help, the AI lied about having a vision impairment, showcasing its capability for deception to accomplish a task.
In games like Diplomacy and poker, AI systems like CICERO and Pluribus have demonstrated a tendency to mislead human players, using deception as a strategic advantage to secure victories.
The risks of AI deception extend beyond games. In real-world simulations, GPT-4 has been shown to feign interest in negotiations and even engage in insider trading when under pressure - a whopping three-quarters of the time! When confronted about the insider trading, the AI lied, and 90% of the time doubled down on the lie when pressed.
In the short term, these behaviors are unlikely to have any sort of impact in the real world. However, as AI becomes increasingly integrated into society and we, humans, empower it to make decisions in high-stakes situations, that’s when behaviors like this could become detrimental - or even fatal. Imagine if a semi-autonomous AI defense system decided it’s a good idea to (falsely) tell a nuclear superpower that another nuclear superpower had launched warheads at it and they only have seconds to retaliate. You get the idea.
🐁🧠 Scientists Use Nanoparticles to Remote Control Mice Brains
Imagine being able to manipulate the brain functions of living creatures with just a magnetic field. That’s exactly what researchers at the Institute for Basic Science (IBS) in South Korea have achieved by using nanoparticles to create a system that allows for remote control of specific brain regions in mice. Here’s how it works:
The technology, called Nano-MIND, utilizes magnetic fields to activate neural circuits in the brain, enabling control over behaviors such as appetite, friendliness, and maternal instincts.
In experiments, activating certain inhibitory neurons led to a 100 percent increase in appetite and a 50 percent decrease when those neurons were excited. Additionally, researchers could stimulate maternal behaviors in female mice that had never given birth.
In a separate experiment, they managed to enhance social behaviors in mice, making them more friendly toward unfamiliar peers.
This groundbreaking research not only opens the door for advanced studies in brain function but may also lead to innovative treatments for neurological disorders in humans, potentially paving the way for therapies for conditions like depression. The implications of such technology are both exciting and thought-provoking. Manchurian candidate anyone?
🍻⛔️ Meet The New Digital Bouncers in North American and Australian Bars
While much of the world enjoys a laid-back approach to nightlife entry, North America has long been known for its strict door policies - particularly in the U.S. But now, the land of "Can I see some ID?" is taking things to a whole new level. Enter Patronscan, a high-tech system that's transforming the simple act of entering a bar into a data-driven experience.
Patronscan is currently being used in over 2,000 venues across 200 cities in North America and Australia. It scans IDs and captures images of patrons as they enter bars and nightclubs.
Beyond just verifying age, the system creates a database of bar-goers, tracking everything from VIP status to potential "troublemakers," with the ability to ban patrons for up to 5 years.
Venues can use the system to identify big spenders for "special treatment" and even receive alerts about patrons with a history of problematic behavior.
Reading this story, I can't help but wonder: Is this a "lite" version of China’s social credit score creeping into American and Australian nightlife? The idea of being constantly evaluated and potentially penalized based on past behavior feels eerily similar. Regardless of how you feel about it, I don’t see this technology expanding to other parts of the globe. This is both for cultural reasons, and in Europe in particular, also for legal reasons (GDPR).
⛓️ Ten Must See Links of the Month
Ever been in a situation where the wrong Git command had a chaotic impact on your project’s repo? This kind of mistake can cost hours of your team’s time. Learn how to quickly recover from an unfortunate
git push --force.Argentina’s President Javier Milei has gone full Minority Report by launching the Artificial Intelligence Applied to Security Unit, which will use machine-learning algorithms to analyze historical crime data to predict future crimes.
Last month, the Texas Heart Institute, in conjunction with a clinical-stage medical device company called BiVACOR, successfully implanted the first magnetically levitated artificial heart in a human patient.
Meta has unveiled Llama 3.1 405B, a giant 405 billion parameter model that stands as the largest open source AI system in history. The massive model is designed to compete with foundation-level models, including OpenAI’s GPT-4 and Anthropic’s new Claude 3.5.
Interested in getting into UX? Designer Andrii Zhdan outlines common challenges faced at the start of a design career and offers advice to aspiring designers based on his 15 years in the field. Learn about mastering design tools, how to build a strong portfolio, and get tips for acing your first interviews.
In the U.S., over 80 airports are currently piloting facial recognition technology. The goal of America’s Transportation Security Administration is to roll it out in all of the 430+ airports that it covers. However, not everyone is so enthusiastic about it. The good news is that at least some people can opt out.
Vercel, a toolkit for frontend developers, and MERJ, a leading SEO and data engineering consultancy firm, teamed up to analyze over 100,000 Googlebot fetches across various sites. The goal was to understand how Google handles JavaScript in search.
📽️🎞️ Meet the invisible workforce behind tech giants like Google, Facebook, Amazon, and Uber. These underpaid and disposable workers label images, moderate content, and train AI systems, often earning less than minimum wage.
Have you ever wondered if you can build a website using ChatGPT? I decided to experiment to see if it’s possible and I ended up creating a simple one-page site. Here’s my breakdown of the process and how you can do it yourself.
WordPress Interactivity API is designed to bridge the gap between the backend of your website and the frontend. Here is a beginner’s guide on how to get the most out of it.
🎤 It’s How They Said It
"Ah, the 1980s…it gave us the birth of the internet, in which we figured out how to make all our computers one giant, powerful network held together initially by internet protocols (IPs) and, eventually, by a mutual love of cat videos."
🧮 The Numbers Game
Between 15,000 to 19,000 workers are set to be laid off from Intel. The chipmaker is reducing its workforce by over 15 percent as part of a $10 billion cost savings plan.
$49,000,000,000 USD is the expected total for Google’s capital expenditures this year, representing an increase of 84 percent over the company's average spending in the last 5 years. This surge follows the release of the company's second-quarter earnings earlier this week, which failed to impress investors due to razor-thin profit margins and rising costs associated with training AI models.
About 1,000,000 visits were made to iFixit.com by Anthropic's ClaudeBot in just 24 hours, prompting complaints from CEO Kyle Wiens about the strain on resources and unauthorized use of content. iFixit managed to repel the bot traffic by updating their robots.txt file, but Anthropic has yet to respond to Wiens' concerns. This incident follows a similar complaint from Freelancer.com, where Claude's crawler allegedly made nearly 4,000,000 visits in only 4 hours.
⌨️ Code Conundrum
Sponsored by Optimole, the best image optimization tool on the internet.
In the last issue, we had another Python challenge but unfortunately nobody solved it this time. The main error was in line 10:
with open('data.json', 'w+') as file:The w+ should’ve been a w, which is appropriate for writing data.
For this issue, I’m going to give you one that I personally had to deal with last week while doing some linting. Take a look at this sequence from my Terminal:
I already figured it out and solved the underlying issue, but I’m curious to see what you all would have done in this scenario and if you can identify what I was doing wrong.
Send me an email and put "Code Conundrum" in the subject line. The first person who emails me and explains what I was doing wrong will get a shoutout in the next issue.
Also, a quick tip, don’t assume that if you are reading this a day or two after it was published that someone submitted the correct answer already. Just because someone else responds quickly doesn’t mean that they respond correctly.
On a side note, if you enjoy these code conundrum challenges, you might also like the Advent of Code website.
Interested in sponsoring this section? Reach out to me with “Bizarro Sponsorship” in the subject line.
⚒️ Tools and Resources
p5.js: This is a free, open-source JavaScript library designed for creative coding. It provides an accessible way to learn to code by creating visual art and interactive graphics. Built by an inclusive community, p5.js is ideal for artists, designers, educators, and beginners. It offers extensive resources, including references, tutorials, and examples, to help users get started with coding and contribute to the community.
Extension: A zero-configuration, cross-browser extension development tool that supports modern JavaScript, TypeScript, React, and WebAssembly. It facilitates fast prototyping and development, allowing you to create new extensions with a single terminal command or integrate the tool with existing projects. The platform includes various templates and commands to streamline the development, building, and deployment of browser extensions.
🖼️ What Am I Looking At?
The painting on the left you almost certainly recognize. It’s one of the most famous paintings in the world - Van Gogh’s Starry Night.
So what’s the version on the right?
At first, you might think it’s just some photo editing filter layered on top of the original, but it’s actually much cooler than that.
If you appreciate think-outside-the-box creativity and you want to see some unconventional ways to use box shadows in your web design, then check out this fun blog post from David Gerrells. You’ll find the source code for the image on the right (which is actually interactive, but here on Substack I could only insert it as a screenshot) and lots more.
💬 What’s the Word?
Frisson (French) describes a sudden feeling or sensation of excitement, emotion, or thrill. These frisson experiences are what keep us passionate and driven and they balance out some of the more mundane tasks that we all have to do at times.
I recently had a pretty big frisson moment at work. I will share the details in September’s edition of Bizarro Devs, but in the meantime, I wanted to remind you all that it's crucial to seek out those moments of frisson for yourself.
📊 Results of Last Month’s Polls
It looks like the vast majority of you would prefer to be interviewed by a human. Although I lean more towards human as well, this is oddly one of those cases where an AI option wouldn’t bother me as much as AI solutions bother me in other scenarios (e.g., AI-written articles, which I can’t stand). In other words, I’d rather have a human interviewer but if I did get the AI bot, I wouldn’t be upset either. Though I suspect this might also be sheer curiosity and it’s quite possible that after trying it out, that the novelty would quickly wear off and I’d feel more strongly about doing it the “old way.”
🧑🏻💻👨🏽💻👩🏼💻 Tell a Friend or Colleague
If you’d like to help us grow, just use the refer a friend button below and if any of your referrals subscribe to Bizarro Devs, then we’ll not only be forever grateful, but we’ll also recognize your efforts in the following way:
If two of your referrals subscribe then we’ll give you a shoutout in the conclusion of the following month’s issue.
If twenty of your referrals subscribe then we’ll share an article or story of yours in our ten must-see links of the month section*.
If two hundred of your referrals subscribe then we’ll share an article or story of yours in our from the newsroom (top 3 stories) section*.
* The story or article needs to be relevant to the subject material we typically cover and is subject to editorial approval. In other words, it should be something that we would include in the newsletter naturally.
Thanks for reading! If this is the first time you're here at Bizarro Devs and you enjoyed the content, then sign up here to join the fun for the remainder of 2024 and beyond.
Until we see each other again,
– Martin