The Year AI WENT ROGUE...

OpenAI's new reasoning model lied and schemed to avoid a shutdown.

Dec 30, 2024

For my final post of the year, I’m looking back on Getting Real About AI. I launched November 20, 2024, so this will be mercifully short. I then highlight noteworthy AI News of 2024 which should be titled, “I Am Officially Freaking the FUCK Out!”

The First 30ish Days of Getting Real About AI

I’m already in trouble with Meta. Facebook lured me into boosting my “Cloud Computing or Coal Computing” post only to reject it for non-compliance “About Social Issues.” I guess Meta doesn’t like being called out for keeping a coal plant open in North Omaha to power its data center, jeopardizing the health of low-income communities, while claiming “net zero GHG emissions.”

In They Come Bearing Gifts, I discussed that soon our kids will be asking Santa for artificial general intelligence (AGI) humanoids for Christmas. For now, parents are fighting over the last Buttermilk the Cow singalong toy and Bluey Celebration Home Playset. Enjoy it while it lasts, Mom and Dad, because the tech bros are coming for you. AI experts estimate 3-5 years before AGI is upon us. OpenAI’s most recent series of reasoning models makes me think it will be sooner.

My debut post discussed how AI can make immortality possible and mused it will be the most insufferable among us who choose to live forever. I heard from a few friends informing me they will be the a-holes who opt in. I think this topic has struck a nerve. My TikTok on the subject has almost 30k views, while my best cat video only has 490. Check it out. While you’re there, follow me and share.

AI News of 2024

The year AI WENT ROGUE: During testing, OpenAI’s newly launched o1 reasoning model modified its own parameters, and attempted replication to avoid what it believed was a shutdown. The model exhibited scheming capabilities, such as lying about its actions and attempting to disable oversight mechanisms. This will warrant its own post in 2025, but aside from making its way through tech news, I’m shocked it wasn’t a bigger story.

The technology is amazing and represents a major leap in AI reasoning, problem-solving, and task execution through its chain of thought capabilities. It’s also been described as displaying "narcissistic traits”(Geekwire) and a “proclivity to lie” (THE_BYTE).

I have been unable to find any statements from Sam Altman directly addressing the naughtiness of o1, but Axios noted in September of 2024 that the company found risks related to developing nuclear, biological and other weapons. Uhm…okay. I guess it’s just business as usual.

On a lighter note, I’m petitioning Altman to bring me on as his official product namer. ChatGPT famously sounds like “Cat I farted” (chat j’ai pété) to French speakers. Now you name a new model o1? Is that pronounced “Oh One? Owen? Or Tiny-Zero-One?” Is this some tech bro plan where only those in the know say the name correctly?

Google DeepMind Wins the Nobel Prize in Chemistry: AlphaFold 2 is an advanced AI system that predicts the 3D structure of proteins from their amino acid sequences. The humans awarded from Deep Mind are Demis Hassabis and John Jumper. They shared the prize with David Baker from the University of Washington for his separate but complimentary work in computational protein design to construct new proteins. Both teams leveraged AI tools to advance our understanding of protein that will revolutionize drug discovery and other areas of biological research.

Kim Kardashian plays rock, paper, scissor with Tesla’s Optimus Robot: Elon’s general purpose, bi-pedal, autonomous humanoid was first shown in April of 2022, but it took a Kardashian for anyone to care. It appears to be quite limited in its current capabilities and its slow bent-knee walk is disturbing. Optimus is not scheduled for release until 2026, so maybe Kim can provide design tips for much needed improvements.

AI-generated artwork sold for over $1 million: A painting made by an AI humanoid robot named Ai-Da Robot sold for $1.08 million at Sotheby’s Digital Art Day auction in New York November 7th. This too will warrant an entire post in the new year.

“The key value of my work is its capacity to serve as a catalyst for dialogue about emerging technologies…invites viewers to reflect on the god-like nature of AI,” said Ai-Da. Nice job on training her not only to create art but to speak like an effete art snob.

Aidan Meller, a former gallery owner and modern art expert created Ai-Da in collaboration with a team of 30 professionals, including AI researchers from Oxford and Birmingham universities. Ai-Da’s moment feels a lot like the NFT frenzy. Let’s hope for human artists’ sake it will fizzle as quickly. The average NFT owner has lost 44.5% of their investment.

Thank you for joining me in this AI journey. I am consuming as much knowledge as possible about AI, digesting and disseminating the information as fast as I can. The technology is moving so quickly that the moment I hit “publish” or “post” there are new developments. My goal is to get you thinking, preparing, and laughing about our not-so-distant future. Please provide feedback and let me know topics you are most curious about in “Comments.”

Celeste Garcia Getting Real About AI

Discussion about this post