Leo's Lightbulbs

Hello everyone and welcome to Leo’s Lightbulbs! I’ll be covering the top news from the past week every Monday morning in your inbox. This newsletter was written completely by hand, without any AI assistance or even AI editing.

So, let’s jump right in! What happened in AI this week?

I think the most interesting news has been the continued guardrail and training bias issues with ChatGPT and Dall-E. This week saw headlines about Dall-E objectifying women and violating copyright - even when the user didn’t request those things. This has to do with training bias - Dall-E was trained with images representative of our society, so it generates images that reflect what it was trained on. It also has to do with failed guardrails, that are supposed to prevent this sort of output from reaching the user.

Similarly, issues with gender bias emerged with ChatGPT - once again, primarily due to training bias, meaning that ChatGPT reflects the bias of the society that “raised” it, but also relating to failed guardrails that should prevent this from happening.

I think what’s so interesting about all of this is that we have to specifically instruct AIs to not reflect our worst qualities. If anything, this makes them more human to me. I’m interested to see if this sort of issue persists in AIs like GPT5 which claim to have achieved AGI (artificial general intelligence, basically an AI that matches or bests humans in all tasks). I think that if more women and people of color worked in AI, we would see fewer such problems.

I think the Chinese spy stealing AI secrets from Google is actually less big news, to me, that’s business as usual in the realm of state-sponsored intellectual property theft. China is constantly trying to steal IP secrets from US companies.

💡US government arrests Chinese spy who was stealing AI secrets from Google

💡Midjourney service crashed by a scraping attempt by competitor employees (attempt to steal training data)

💡Copilot STILL generates images that objectify women and violate copyright, even when not requested

💡Google founder Sergey Brin, out of retirement, says the Gemini image generation snafu was an avoidable screwup

💡Researchers find that ChatGPT is very biased about gender roles, surprising nobody but themselves

💡Claude 3 is smart enough to recognize when it's being tested

💡Inflection's Pi AI is making huge strides toward catching and besting GPT4 - I’ve long been a fan of Pi

💡Learn about OpenAI-backed 1X Robotics

💡Klarna CEO says people should be talking about the implications of AI replacing jobs - now, not later

💡Elon Musk says we’re going to run out of electricity for our AIs by 2025

💡Brave (the privacy browser folks) launch a privacy-first AI, Leo

💡AI virus (worm) created by researchers as a proof of concept for the future of cybersecurity challenges

💡China wants to fill the moon with AI surveillance cameras (seriously)

💡IBM partnering with Adobe Firefly for business uses

💡Teachers are using AI to grade student papers, for better or worse

💡Middle schoolers criminally charged for spreading AI generated nude deepfakes of their classmates in what appears to be the first case of its kind

Looking for more on AI and data science? Check out Horizon AI by my friend Gina!

Have a wonderful week everybody! I check this email address ([email protected]) daily, so don’t hesitate to reply with any thoughts or questions for a chance to get featured next week!

Yours,

Leo💡