The AI race is getting expensive
AI labs are spending billions to build massive data centers, some as big as entire city blocks. These machines use huge amounts of electricity just to train larger and larger models.
Tech companies hope that more computing power means better results, but many experts now say that approach is hitting a wall. Despite all the flash, these massive systems may not actually be the smartest way to move forward in artificial intelligence anymore.
One researcher is going against the trend
Sara Hooker used to be the VP of AI research at Cohere and worked at Google Brain before that. Instead of following the scaling crowd, she co-founded Adaption Labs to prove there’s a smarter path.
Hooker believes the idea of endlessly scaling models is no longer useful and is slowing progress. Her new company is built around efficiency and real-world learning, not size. It’s a bold move in a field obsessed with being the biggest.
Meet the startup shaking things up
Adaption Labs is focused on a different kind of AI, one that adapts like a person. Instead of just learning from huge datasets in the lab, their systems are designed to learn from the real world while in use.
That means the AI would improve over time, just like people do when they make mistakes. Hooker and co-founder Sudip Roy are challenging the core belief that more hardware always equals better performance.
Why scaling alone may not work anymore
Many researchers are now saying that simply adding more data and computing power doesn’t lead to smarter AI. There’s growing evidence that this strategy is delivering smaller gains while costing more.
At some point, bigger models stop being useful and just become more expensive. That’s the key idea behind Adaption Labs. Hooker wants to find smarter ways to improve AI performance without relying on constant upgrades to size and hardware.
AIs that learn like we do
Imagine an AI that learns from its own mistakes in real time, just like a person would. You stub your toe once, and next time, you walk around the table. That’s the kind of learning Adaption Labs is aiming for.
Most current AI systems don’t work that way; they need to be retrained from scratch. Hooker wants to break away from that cycle and build systems that grow and change naturally from experience.
The current method is too costly
Most companies can’t afford to keep up with today’s AI costs. Fine-tuning an AI model to meet specific business needs can cost millions, especially with firms like OpenAI.
That’s where Adaption Labs sees an opening. If AI can be made to adapt on its own, that would cut out expensive retraining and consulting. Hooker’s team is working on smarter tech that doesn’t need to be babysat or rebuilt every time something changes.
Reinforcement learning isn’t enough
Reinforcement learning, or RL, is a method where AI learns by trying things and getting feedback, but it mostly works in controlled labs. Once AI is in the real world, it often can’t adjust to new problems or fix its own mistakes.
It ends up repeating errors because it doesn’t keep learning. Hooker sees this as a major flaw. Her goal is to build models that can truly adapt, even outside a lab environment.
Making AI more accessible for everyone
Right now, AI tools are controlled by a small group of big companies with deep pockets. They create models and sell access to them, but they’re expensive and hard to customize. Adaption Labs wants to lower that barrier.
If they succeed, smaller businesses and developers could get smarter, more flexible AI without the high price tags. That could lead to a more balanced, more creative tech world where more people can innovate.
Hooker’s bet on real-world learning
Hooker believes the future of AI lies in how well it can respond to the world around it. Instead of stuffing more information into models, she’s focused on how they can adjust and improve from real use.
The human brain doesn’t need millions of examples to learn; it learns quickly and efficiently. That’s the inspiration behind Adaption Labs. They want AI that grows through action, not just memorization, leading to more useful and reliable systems.
AI spending is hitting wild levels
Research labs are pouring money into new models at shocking rates. One recent project exploring RL improvements cost over $4 million just to run. These budgets are getting harder to justify, especially as performance gains shrink.
Hooker argues that there’s a better way, one that doesn’t require endless resources. Adaption Labs is trying to show that smarter design and real-time learning can outperform massive budgets and brute-force computing.
Tiny models are winning big
Small, efficient AI models are starting to outperform their larger counterparts on many tasks like math, coding, and problem-solving. Hooker led similar efforts while at Cohere, proving that compact systems can be powerful and cost-effective.
This is great news for smaller companies and developers, because it means AI doesn’t have to be massive to be smart. It’s a trend Adaption Labs is betting will grow even more in the next few years.
The AI world is rethinking everything
There’s been a noticeable shift in how the tech community talks about AI. Even popular AI podcasters and researchers are now questioning the long-term value of scaling. People who once championed bigger models are expressing doubts about their usefulness.
Hooker is part of this growing wave of thinkers asking, “What if we’ve been going the wrong way?” It’s not just a trend, it’s a major pivot that could shape the future of AI.
A chance to shift the power
Today’s AI world is dominated by a few companies, but Adaption Labs could change that. By building adaptable systems that are cheaper and smarter, they could put advanced tools into more hands.
That means AI could better serve smaller organizations, schools, and communities, not just big tech. Hooker’s team wants to change who has control over AI and who it works for. It’s a mission that could shake up the entire industry.
Bringing global talent into AI
Hooker has a strong track record of hiring people from regions often overlooked by the tech world, especially in Africa. She wants Adaption Labs to be a global-first company, not just based in Silicon Valley.
The goal is to build a team that reflects the world, not just the usual tech crowd. That kind of diversity could help the company build smarter AI that works for more people, in more places, in more ways.
Curious how other AI teams are rethinking the game? Check out what Gemini 2.5 Pro is experimenting with here.
Why this could change everything
If Adaption Labs succeeds, it could flip the script on what AI needs to be powerful. Instead of giant models that require millions to run, we’d have smart, flexible systems that learn like we do, efficiently and in real time.
That would make AI more useful, less wasteful, and way more affordable. Hooker is betting big on this future, and if she’s right, it could lead to one of the biggest shifts in AI yet.
Want to see how other AI power moves are shaking things up? Take a look at who Meta just brought on board here.
Think AI needs a reset, too? Drop your thoughts in the comments and hit like if you’re excited about smarter, leaner systems like this.
Read More From This Brand:
- AI just started making social norms
- How students use AI to stay on track
- The new Meta AI tool that claims to decode your thoughts
Don’t forget to follow us for more exclusive content right here on MSN.
This slideshow was made with AI assistance and human editing.
This is exclusive content for our subscribers.
Enter your email address to instantly unlock ALL of the content 100% FREE forever and join our growing community of smart home enthusiasts.
No spam, Unsubscribe at any time.
Lucky you! This thread is empty,
which means you've got dibs on the first comment.
Go for it!