Applied AI Weekly News

Justin Grammens

Welcome, Applied AI Weekly Readers, to the latest issue! I'm thrilled to once again share with you the most interesting articles on artificial intelligence that I've found this past week.

Here are the highlights: DeepSeek's AGI ambitions, OpenAI's emotionally advanced GPT-4.5, and AI tools detecting research errors. Notably, 25% of new Y Combinator startups now use 95% AI-generated code. The Pentagon has partnered with Scale AI for military planning, while Amazon unveiled Alexa+ with enhanced LLM capabilities and orchestration features. Also featured is Light-R1-32B, an impressive open-source math model that outperforms competitors despite only $1000 in training costs.

Before we go there and get into the details, here are a few things I wish to share.

Upcoming Events

Our next Applied AI event is TRASH TALKING - Applying AI and IoT to the Refuse and Recycling Industry. We'll be talking about how Artificial Intelligence (AI) and the Internet of Things (IoT) can revolutionize refuse management and recycling in America.
For the Spring 2025 Applied AI Conference, I'm thrilled to announce that will be partnering with MinneAnalytics.
Yahoo! The next conference I'll be speaking at is the closing keynote at the Illinois Associations of Home Care & Hospice. I'll be speaking on the new wave of agents that will transform the way home care hospice providers provide care.

A Bit of Thanks

Thank you to everyone who came out and supported me and the panel at the Future of Healthcare Conference at the University of Saint Thomas. Here's some photos of our panel at the event with Minnesota healthcare leaders.
I'm continuing to offer free consultations and workshops with many business leaders on how AI is changing the landscape of how you will run your business today and into the future. Connect today and book a meeting with me.

Now that we have that covered, please enjoy the articles I have spent time finding and curating for you this past week.

Finally, please do reach out if there is anything you feel I might have missed. Enjoy!

News

The Meaning of Artificial General Intelligence Remains Unclear

When Chinese AI startup DeepSeek burst onto the scene in January, it sparked intense chatter about its efficient and cost-effective approach to generative AI. But like its U.S. competitors, DeepSeek’s main goal is murkier than just efficiency: The company aims to create the first true artificial general intelligence, or AGI.

Hands-On With GPT-4.5, OpenAI’s Most Powerful Model Yet

When comparing AI benchmark tests from competitors’ models as well as OpenAI’s “reasoning” releases, the benefits of using GPT-4.5 are not immediately clear. However, in the model’s system card and in a previous interview with WIRED, the OpenAI researchers who worked on GPT-4.5 claimed improvements can be felt in the anthropomorphic aspects of the model, like a stronger intuition and a deeper understanding of emotion.

AI Tools Are Spotting Errors in Research Papers

Late last year, media outlets worldwide warned that black plastic cooking utensils contained worrying levels of cancer-linked flame retardants. The risk was found to be overhyped — a mathematical error in the underlying research suggested a key chemical exceeded the safe limit when in fact it was ten times lower than the limit. Keen-eyed researchers quickly showed that an artificial intelligence (AI) model could have spotted the error in seconds.

Sponsor

Innovating Across the Americas

Stay ahead with insights from our experts across Canada, the U.S., Mexico, and South America. From application and data modernization to AI/ML, platform engineering, and solution architecture—we’re shaping the future of digital services. Let’s build what’s next, together.

Improving – It’s what we do.™

Business

A Quarter of Startups in YC's Current Cohort Have Codebases That Are Almost Entirely AI-Generated

With the release of new AI models that are better at coding, developers are increasingly using AI to generate code. One of the newest examples is the current batch coming out of Y Combinator, the storied Silicon Valley startup accelerator. A quarter of the W25 startup batch have 95% of their codebases generated by AI, YC managing partner Jared Friedman said during a conversation posted on YouTube.

Yum's Taco Bell Shows off AI Tool for Fast-Food Managers

The use of AI is "already beginning around labor and inventory," said Dane Mathews, Taco Bell's Chief Digital and Technology officer, at the event. To demonstrate its current and planned use of AI technology, Chief Operating Officer Jason Kidd showed Wall Street analysts a video skit with a restaurant manager talking to a human character playing the role of an AI assistant, which Yum calls Byte AI Restaurant Coach.

Healthcare

Could Artificial Intelligence Increase Clinician Burden?

Vendors say their AI scribes and assistants will save doctors time. But the need for careful oversight, along with hospitals’ inclination to see as many patients as possible, could mean clinicians’ schedules won’t be freed up by AI tools.

Health Expert Warns of Leaning Too Heavily on AI for Social Connections

Recent research from app intelligence provider Appfigures found that AI companion mobile apps were seeing over 652% year-over-year revenue growth in 2024, attracting $55 million in consumer spending over the course of the year, for instance. The U.S. was the top market for these apps last year, accounting for 30.5% of total consumer spending.

Industrial

Pentagon to Give AI Agents a Role in Planning, Operations

The American military has signed a deal with Scale AI to give artificial intelligence, as far as we can tell, its most prominent role in the Western defense sector to date – with AI agents to now be used in planning and operations.

The value of the contract, awarded as part of the US Defense Innovation Unit's Thunderforge project, wasn't specified, though given its considerable scope it's likely to be a large one.

Consumer

Google Is Expanding AI Overviews Worldwide, and Adding an AI Mode

The AI-ification of Google Search continues to accelerate: the company announced on Wednesday that it will start showing AI Overviews for even more kinds of queries and that users around the world, even those who are logged out of Google, will start seeing them too.

Introducing Alexa+, the Next Generation of Alexa

At the foundation of Alexa’s state-of-the-art architecture are powerful large language models (LLMs) available on Amazon Bedrock, but that’s just the start. Alexa+ is designed to take action, and is able to orchestrate across tens of thousands of services and devices—which, to our knowledge, has never been done at this scale. To achieve this, we created a concept called “experts”—groups of systems, capabilities, APIs, and instructions that accomplish specific types of tasks for customers.

Development

Quantum AI Computing

Imagine a world where the rules of what we know about computing, reality, and even existence are turned on their heads. You’re scrolling through this blog, but somewhere out there, another you is doing the same. Welcome to the fascinating journey into quantum computing, where possibilities are as abundant as the stars in the sky! This post delves into the heart of what quantum computing is and how devices like Google’s groundbreaking quantum computer are about to make our current tech seem outdated and trivial.

Exploring LoRA as a Dynamic Neural Network Layer for Efficient LLM Adaptation

LLMs need constant updates — legal AI must learn new laws, finance chatbots need fresh market data, and medical models should adapt to new research. But traditional fine-tuning is expensive. LoRA helps, but most versions are static, using a fixed rank for updates. I propose a smarter approach: a dynamic LoRA that adjusts rank based on data complexity, making fine-tuning more efficient.

New open-source math model Light-R1-32B surpasses equivalent DeepSeek performance with only $1000 in training costs | VentureBeat

Researchers have introduced Light-R1-32B, a new open-source AI model optimized to solve advanced math problems. It is now available on Hugging Face under a permissive Apache 2.0 license — free for enterprises and researchers to take, deploy, fine-tune or modify as they wish, even for commercial purposes.