AI Insight
The AI stories you don't want to miss this week!
(yawn) Hey everybody. Just a normal day in the US of A today. Did I miss anything?
OK let me be real for a second. I am seriously distracted today and I’m obviously not alone. I’m surprised I got the podcast out at all if I’m honest. There’s a lot to sort through post-election, but all I can really say at this point is that a thriving democracy relies on citizens showing up to the polls and voting for the candidate that represents the direction they want to see the country go in. Needless to say, the results last night were pretty darn decisive. As someone who loves my country, I MUST accept that though my person didn’t win, a wide majority of the country wanted the other guy to win, they showed up and voted, and win he did. And that’s how its all meant to work, right? So, no sour grapes over here. (ok maybe a few) I gotta respect the win, and I’m going to do my part to understand all that I don’t know about the needs of those who chose to vote him into power for the next four years. It’s as simple as that. And that’s about as political as I’m willing to get here, at least right now. Stay sane and healthy, ya’ll.
Alright, let’s get to some AI news! On today’s episode of the AI Inside podcast, I invited my friend Mike Elgan from the Machine Society newsletter to join me in Jeff Jarvis’ absence. You can watch that conversation here, and while your at it, here are some of my expanded thoughts on the big news items from the show including a few stories that didn’t make it.
ChatGPT's New Search Powers Take Center Stage
OpenAI has introduced ChatGPT Search, powered by a specialized version of GPT-4o, providing real-time information access through the platform. The service includes licensing deals with major news organizations like AP, Reuters, and Axel Springer. While currently available to paid users, it will roll out to free users in the coming weeks. Users can access the feature through the globe icon labeled "Search the web" in ChatGPT.
Big Tech's AI Momentum
Meta's AI has reached an impressive milestone with 500 million monthly active users, driving increased engagement across its platforms. The chatbot integration has boosted time spent in its main app by 8% and Instagram by 5%.
Meanwhile, Microsoft's AI business is on track to become its fastest product to hit a $10 billion annual run rate, with the company confirming $13 billion in total funding commitments to OpenAI.
Robots Get Smarter with HPT
MIT researchers have developed Heterogenous Pertained Transformers (HPT), a breakthrough technique for training general-purpose robots. This system combines multiple data sources, including simulations, real-world robots, vision sensors, and robotic arm positions, into a unified language model. The extensive dataset includes over 200,000 robot trajectories, aiming to create a "universal robot brain".
While on the topic of robotics, Mike Elgan also wrote two pieces about why he believes humanoid robotics are not necessary and ill advised for the problems they purport to solve. We also have a very healthy discussion on this topic on the podcast.
AI Regulation Takes Shape
Anthropic has published its stance on AI safety, advocating for targeted regulation of fundamental model properties rather than use-case specific rules. The company suggests an 18-month timeline for implementation.
Meanwhile, Reuters reports that major AI models are struggling to meet upcoming EU regulations, with even top performers like GPT-4 and Claude 3 scoring only 83% on compliance metrics.
Meta's Military Pivot
In a significant policy shift, Meta will now make Llama AI models available to US government agencies and defense contractors. The company is partnering with major players like Amazon, Microsoft, and Lockheed Martin, with applications ranging from aircraft maintenance to national security missions.
Amazon’s Next-Gen Alexa Pushed to 2025
Amazon's plans to revolutionize Alexa with advanced AI capabilities have hit a speed bump, with the launch now pushed to 2025. If this story feels familiar, it's because we're seeing what I'd call the "greatest hits" of AI challenges playing out in real-time: frustratingly slow response times, hallucinations, and difficulty handling basic tasks that the current Alexa manages just fine. It's giving me serious déjà vu of Google's Assistant/Gemini transition.
AI in Hollywood
The upcoming film "Here," starring Tom Hanks and Robin Wright, showcases AI's transformative potential in filmmaking. Rather than using traditional CGI for de-aging effects - which typically requires hundreds of VFX artists and millions of dollars - the production employed AI tools that could achieve similar results "with the click of a few keystrokes."
Meanwhile, Runway ML continues to push boundaries with their Advanced Camera Control for Gen-3 Alpha Turbo, enabling sophisticated camera movements and effects.
Here are a few other interesting stories
What a Trump Victory Means for Tech - nytimes.com
Google’s AI system could change the way we write: InkSight turns handwritten notes - venturebeat.com
Introducing Swarm.js: Node.js Implementation of OpenAI Swarm! - openai.com