AI
Revolutionizing the Future: How Pocket-Sized AI Models are Shaping a New Computing Era
To go back to this article, access My Profile and then select View saved stories.
Knight Will
Compact AI Models May Herald the Dawn of a New Computing Age
Upon its release in November 2023, ChatGPT was initially available solely via cloud services due to the sheer size of its underlying model.
Today, I'm operating an equally powerful artificial intelligence software on a Macbook Air, and surprisingly, it's not even heating up. This reduction highlights the swift progress scientists are making in optimizing AI algorithms to be more streamlined and resource-efficient. Additionally, it demonstrates that increasing size isn't the sole method to enhance the intelligence of machines significantly.
The AI model currently enhancing my laptop with capabilities akin to ChatGPT is named Phi-3-mini. This model is a member of a new series of smaller AI models developed by Microsoft's team. Despite its small size, allowing it to operate on a smartphone, I opted to experiment with it on a laptop and interacted with it using my iPhone via an app named Enchanted, which offers a chat environment much like the original ChatGPT application.
In a publication detailing the Phi-3 series of models, researchers from Microsoft assert that the model I utilized compares positively with GPT-3.5, the OpenAI model that powered the initial launch of ChatGPT. This assertion is grounded in evaluations of its effectiveness across a variety of AI benchmarks, which are intended to assess common sense and reasoning abilities. From my personal experimentation, it indeed appears equally proficient.
This week, during its yearly developer event, Build, Microsoft unveiled the Phi-3 model, an innovative “multimodal” technology designed to process audio, video, and text. This introduction follows closely on the heels of OpenAI and Google, who recently showcased their groundbreaking AI assistants powered by multimodal models that operate through cloud services.
Microsoft's diminutive series of AI technologies indicates a growing ability to develop a variety of useful AI applications independent of cloud-based resources. This shift could lead to innovative applications by enhancing their speed and privacy. (For example, Microsoft's new Recall function, which leverages AI to make every action on a user's PC searchable, relies on these offline algorithms.) Additionally, the unveiling of the Phi series sheds light on the current state of AI development and potential paths for its enhancement. Sébastien Bubeck, a Microsoft researcher involved in the endeavor, explained that the initiative aimed to explore how a more discerning approach to the training data of AI systems might refine their capabilities.
Massive language models such as OpenAI's GPT-4 and Google's Gemini, which fuel chatbots and various applications, are commonly loaded with vast amounts of text extracted from books, websites, and virtually any available source. This approach has sparked legal debates, yet OpenAI and similar entities have discovered that enhancing the volume of text input into these models, along with boosting the computational power for their training, can reveal previously untapped potential.
Authored by Carlton
Authored by Emily Mullin
By Andy Greenberg
Authored by Scott Gilbertson
Bubeck, intrigued by the ai-allcreator.com">kind of "intelligence" demonstrated by language models, opted to explore whether meticulously selecting the data inputted into a model could enhance its performance without the need to significantly increase its training data.
In September last year, his group utilized a model that was about 1/17th the scale of OpenAI's GPT-3.5, and they trained it on high-quality artificial data produced by a more substantial AI system. This data encompassed details from particular fields, such as programming. The performance of this smaller model was unexpectedly impressive. "To our astonishment, we found that our approach enabled the model to outperform GPT-3.5 in coding tasks," he mentions. "This outcome was quite unexpected for us."
The team led by Bubeck at Microsoft has also unveiled new findings through this method. Their research demonstrated that training a significantly small model with children’s tales enabled it to generate uniformly logical responses, a notable achievement since AI systems of similar scale usually yield nonsensical results when trained using standard methods. This outcome further implies that AI applications, which appear to lack sufficient capability, can become functional if they are taught using appropriate content.
According to Bubeck, the findings suggest that enhancing the intelligence of future AI technologies will involve more than merely increasing their size. Furthermore, it appears that smaller-scale models like Phi-3 will play a crucial role in the evolution of computing. Operating AI models directly on devices such as smartphones, laptops, or PCs minimizes delays or disruptions that can arise when data must be processed in the cloud. This approach ensures that personal data remains on the user's device and may pave the way for entirely new AI applications that are not feasible with a cloud-based approach, including AI functionalities that are seamlessly integrated into a device’s operating system.
Anticipation is high for Apple to reveal its much-anticipated approach to artificial intelligence at the upcoming WWDC event. The company has often highlighted its unique combination of in-house hardware and software, enabling AI processes to be conducted directly on its devices. Instead of engaging in a direct competition with giants like OpenAI and Google over the development of vast cloud-based AI systems, Apple could take a distinctive route by concentrating on miniaturizing AI technology to be more accessible and portable for its users.
Suggested for You…
Delivered to your email: Dive into Will Knight's Fast Forward for the latest on AI progress
Millions of deepfake videos are inundating Indian electorates.
Incarcerated individuals acquired tablets behind bars, only to discover an unful
The primary obstacle impeding the progress of the heat pump
Eternal Sunshine: Discover the Perfect Sunglasses for Any Adventure
Dave Paresh
The name in
Lauren Goode
No text provided to
Knight Will
Lauren Good
Roger Reece
Knight Will
Additional Content from WIRED
Evaluations and Tutorials
© 2024 Condé Nast. All rights reserved. Purchases made through our site may result in WIRED receiving a share of the sales, as we have affiliate agreements with various retailers. The content on this site is protected and cannot be copied, shared, broadcast, stored, or utilized in any form without explicit written consent from Condé Nast. Choices regarding advertisements.
Choose a global location
Discover more from Automobilnews News - The first AI News Portal world wide
Subscribe to get the latest posts sent to your email.