According to a team of scientists, ChatGPT-like large language models (LLMs) can now create smaller AI tools without human intervention.
It is no secret that 2023 was the year of artificial intelligence (AI). A slew of exciting developments were reported in the AI space this year.
The pace of AI innovation in 2023 was astounding – but even more impressive was how people applied this technology to make a real difference in their careers, communities, and countries. Here are some of my favorite examples. pic.twitter.com/78l9mhMUFx
— Satya Nadella (@satyanadella) December 14, 2023
For instance, Google and Microsoft unveiled their own AI-powered chatbots, Bard and Bing. Likewise, OpenAI’s widely popular AI chatbot ChatGPT got more powerful with GPT-4.
Google also introduced Gemini AI, which outperformed ChatGPT in multiple benchmark tests and is currently powering Bard. Aside from this, the company is reportedly working on a program codenamed “Project Ellmann” that can use Gemini AI to tell your life story.
As if that weren’t enough, the search giant recently announced its AI studio in a bid to keep up with other tech companies. So, it is safe to say that the year has been all about AI advancement.
AI used to create new AI
Now, a team of scientists from MIT and the University of California campuses along with AI technology company Aizip, claim they can get large AI models, like the one that powers ChatGPT, to replicate automatically.
“Right now, we’re using bigger models to build the smaller models, like a bigger brother helping [its smaller] brother to improve,” CEO of Aizip Yan Sun told Fox News.
The top executive believes this is the first step towards a bigger job of self-evolving AI. “This is the first step in the path to show that AI models can build AI models,” he added.
Yubei Chen, who is one of the researchers, a U.C. Davis professor and Aizip co-founder, said they were surprised to find that the largest model can be used to automatically design the smaller ones.
Chen believes the large and the small models will collaborate together in the future and build a complete intelligence ecosystem.
According to Chen, AI can be used to spawn models that are capable of “identifying human voices among ambient noise, monitoring pipeline data to proactively prevent integrity issues and analysing satellite and ground-based sensor data to track wild animals”.
He noted that the technology is a breakthrough because they have designed a fully automated pipeline for the first time. It can design an AI model without requiring human intervention, Chen added.
Chen went on to say that the researchers have demonstrated the first proof of concept that shows one type of model can be automatically designed from scratch. In other words, this AI model can do everything from data generation to model deployment and testing without any human involvement.
Sun and Chen demonstrated a human activity tracking device, which used AI to gather and analyse motion data.
Unlike giant language models, which cost a bomb to run, these smaller AI models (A.K.A. TinyMLs) cost less and are more portable. TinyML can be used for specific tasks. For instance, TinyML can be used to improve home gadgets, hearing aids or facial recognition.
Talking about their mission, Sun said their goal is to bring intelligence to everyday life while making things safer and smarter.
Microsoft shows off its small language models
Meanwhile, Microsoft has given us a glimpse into its own small-scale AI models that can give ChatGPT a run for its money. According to the folks at Business Insider, this is the Redmond-based tech giant’s “OpenAI backup plan”.
To those unaware, Microsoft has invested a lot of money into OpenAI this year hoping to cash in on the skyrocketing popularity of ChatGPT.
In fact, the company has integrated OpenAI’s technology into some of its products including Bing. However, following the turmoil at OpenAI which saw CEO Sam Altman’s shock firing and reinstatement, Microsoft appears keen to show it doesn’t lack powerful technologies.
According to Business Insider’s Ashley Stewart, Microsoft’s executives have separated the company from OpenAI. Microsoft unveiled a new iteration of an AI model called Phi-2 earlier this month. The company’s researchers say this model is small but mighty.
