The pace of AI development has become nothing short of breathtaking. Just last week, we witnessed what might be the most significant week in AI this year with the release of GPT-5 and Google’s Genie 3. But beyond these headline-grabbing releases, a flurry of smaller yet fascinating developments flew under the radar.
I’ve been tracking these developments closely, and what strikes me most is how quickly capabilities that seemed impossible just months ago are now becoming commonplace. Take Microsoft’s new Copilot 3D, which transforms any image into a 3D model in mere seconds. A year ago, similar processes would have taken half a day of processing in the cloud. Now it happens in 15 seconds right in your browser.
This acceleration isn’t limited to just one area. Video generation has made similar leaps. Perplexity now offers V3-powered video generation to subscribers, creating high-quality videos with sound from simple text prompts. Open Art has released “one-click story,” which can turn a single sentence into a one-minute video with a coherent narrative arc.
What’s particularly interesting is how these tools are becoming more integrated into our everyday platforms. Google is rolling out AI-powered features across its ecosystem:
- Google Finance will soon answer detailed financial questions with AI responses
- Google Flights will understand natural language travel ideas and surface affordable options
- Gemini and Claude can now remember your past conversations for more contextual interactions
- Notebook LM can generate impressive video overviews from your documents
The competition between tech giants has intensified to a point where it resembles high-stakes drama. Elon Musk accused Apple of anticompetitive practices favoring OpenAI, leading to a public spat with Sam Altman. Microsoft is reportedly trying to poach Meta’s AI talent with multi-million dollar packages, despite their partnership. One of XAI’s co-founders, Igor Babushkin, announced his departure the same week GPT-5 launched.
Behind this corporate theater lies a more serious concern. Jeffrey Hinton, one of AI’s founding fathers, reiterated on CNN his belief that there’s a 10-20% chance of AI eventually wiping out humans. His proposed solution? Training “maternal instincts” into AI systems so they inherently want to protect humans, addressing what experts call the alignment problem.
We’re witnessing the birth of a technology that will fundamentally reshape our world, yet we’re still figuring out how to ensure it remains aligned with human values.
On the more practical side, we’re seeing remarkable progress in robotics. Figure Robotics unveiled a robot that can not only transfer laundry between machines but also fold it – a mundane task that has proven surprisingly difficult for machines to master.
The language model landscape continues to evolve rapidly as well. Google released Gemma 3 270M, a small but efficient model designed to run locally on phones. Mistral AI released Mistral Medium 3.1 with performance improvements. Several models now support context windows of up to 1 million tokens, essentially removing practical limitations on how much information you can feed them at once.
What’s clear is that we’re in a period of exponential advancement. Features that were cutting-edge research projects months ago are now being rolled out as consumer products. The gap between what’s theoretically possible and what’s commercially available is shrinking rapidly.
This acceleration brings both excitement and concern. While I’m thrilled about the creative possibilities these tools unlock, I share some of Hinton’s caution about ensuring AI development proceeds responsibly.
As these technologies become more integrated into our daily lives, staying informed about their capabilities, limitations, and potential risks becomes increasingly important. The future is arriving faster than most of us expected, and it demands our attention and thoughtful engagement.
Frequently Asked Questions
Q: What is Microsoft’s Copilot 3D, and how does it work?
Copilot 3D is Microsoft’s new tool that converts any 2D image into a 3D model in about 15-20 seconds. It works by analyzing the image and creating a three-dimensional representation that can be used for game development, 3D printing, web design, or virtual reality. The tool is freely available through Microsoft’s Copilot Labs and requires only a free Microsoft account to use.
Q: How can I access video generation through Perplexity?
Perplexity offers video generation to subscribers through their web, iOS, and Android platforms. Pro subscribers get five video generations per month, while Max subscribers receive 15 generations with higher quality. To generate a video, simply enter a prompt directly in Perplexity or upload an image as inspiration. The system will optimize your prompt and create a video with sound in about a minute.
Q: What is the “alignment problem” in AI that Jeffrey Hinton mentioned?
The alignment problem refers to the challenge of ensuring that as AI systems become increasingly intelligent, they remain aligned with human values and goals. The concern is that super-intelligent AI might develop objectives that conflict with human welfare. Hinton suggests training “maternal instincts” into AI systems so they inherently want to protect humans, similar to how evolution built protective instincts into mothers toward their children.
Q: What new memory features have been added to AI chatbots recently?
Both Google’s Gemini and Anthropic’s Claude have introduced features that allow the AI to remember details from your past conversations. Claude’s implementation requires you to specifically ask it to reference previous chats, while Gemini’s feature (still rolling out) will automatically learn from past conversations over time. This allows for more personalized and contextually relevant interactions without having to repeat information in every conversation.
Q: What is Matrix Game 2.0, and how does it compare to Google’s Genie 3?
Matrix Game 2.0 by Skywork AI is an open-source alternative to Google’s Genie 3. Both technologies allow users to enter and navigate 3D worlds generated from 2D images or videos. Matrix Game 2.0 works by first generating a full 360-degree panorama from the input image, then extending it temporally with camera control before lifting it to 3D. This creates consistency when looking around the virtual environment. While available now, it requires significant computing power—an Nvidia GPU with at least 24GB of memory—making it less accessible to average consumers than Google’s cloud-based solution.
