AI predictions for 2025 (and beyond?)

I'm having a hard time finalizing this year's prediction because the timeline is tricky. Last few years I noticed most my predictions (70-80%) come true as expected and the rest take some extra time.

So I may have left out some that happen, and included some that don't happen until 2026 or later.

Challenges are the "jagged-frontier" of artificial intelligence and adoption make it difficult to get the timeline right.

I expect the main obstacle for progress to remain adoption.

The impact difference between the users at the cutting edge and the rest will increase. And that's a hint of what to do to decrease that difference for yourself.

Thing 1 - AI predictions for 2025 (and beyond?)

AI and Intelligence Expansion

• More and more devices start getting voice interfaces and intelligence (i.e. a microwave, fridge, washer and dryer, or whatever)
• AI agents can do useful real-life stuff. If you can do it on the phone call or via web browser, AI can get you 80% or more of the way.
• Open-source models become the most used models in the world. Maybe even a local model.
• Small-ish local models get deployed to mobile devices for various purposes. You might not even know you're seeing an AI-influenced information.
• Affordable AI inference hardware for hackers and tinkerers - add GPT-4 level intelligence to any of your projects for a few hundred bucks or less.

AI in Everyday Life

• Driverless cars (and autonomous driving in general) go more mainstream. This means expanding the operating area of current vehicles as well as more autonomous vehicles on the roads.
• Tesla gets embedded Grok (or whatever else xAI decides to name it).
• More AI used in education - as primary and secondary contact.
• Robots with limbs (quadrupeds, robots with one or more arms, etc.) start becoming commonplace with early adopters.
• Building autonomous robots becomes more accessible, and there are hobbyists who do it.

AI and Innovation Leadership

• xAI gets more traction and starts rocking the boat of the rest. They just have too much compute and talent not to make waves. And data, don't forget the data.
• Some will fight it (even on the service provider side), some will embrace it - guess who wins?

AI in Science and Research

• AI in sciences. Purpose-built transformer models for various niches (genomics, proteins, material science, whatever).
• AI that helps verify/improve science papers.
• Maybe an agent that reads and implements papers as a way to reproduce results - if reproducible and good results, guarantees publishing (might not happen this year though).
• We get research published that wouldn't happen without AI's help

We are so early...

Adoption will increase (i.e. your mom will now have heard about it and even tried it). But nowhere near enough. Most people still won't use it as much as they could.

Thing 2 - AI of the week

• Prime Intellect + researchers from USC released METAGENE-1. Open-source state-of-the-art 7B parameter Metagenomic Foundation Model. Enabling planetary-scale pathogen detection and reducing the risk of pandemics in the age of exponential biology.
• NVIDIA introduced Cosmos - world model development platform that consists of world foundation models, tokenizers and video processing pipeline to accelerate the development of Physical AI at Robotics & AV labs
• Humva AI launched a new Custom Avatars feature that allows users to create AI-generated versions of themselves with remarkable flexibility
• Kokoro is a frontier text-to-speech (TTS) model for its size of 82 million parameters (text in/audio out) - and it sounds good!

Thing 3 - Minutes That Steal Your Life

We think that time is ours to waste. We even say, “We have two hours to kill” or speak of dead time between projects. The irony! Because time is the one that’s killing us. Each minute that passes is not just dead to us, it brings us closer to being dead.

- Ryan Holiday (Daily Stoic)

How much more would you get out of life if you acted with a sense of urgency?

Cheers, Zvonimir