Showing posts with label AI. Show all posts
Showing posts with label AI. Show all posts

Friday, April 25, 2025

Google Gemini Progression April 18-25, 2025

**Google Gemini Advanced is free for college students through finals 2026**

This brief two-minute video demonstrates the rapid evolution of Google's Gemini AI. Within just one week, the AI advanced significantly - starting on April 18, 2025 when it could solve electromagnetic problems but couldn't generate corresponding images, to later being capable of creating detailed, dimensionally accurate visualizations of those same problems on April 25, 2025.This brief two-minute video demonstrates the rapid evolution of Google's Gemini AI. Within just one week, the AI advanced significantly - starting on April 18, 2025 when it could solve electromagnetic problems but couldn't generate corresponding images, to later being capable of creating detailed, dimensionally accurate visualizations of those same problems on April 25, 2025.

Here's my takeaways:

·      Speed of development: A one-week timeframe for implementing significant new capabilities in a complex AI system demonstrates extraordinary engineering progress.

·      Cross-modal integration: The transition from purely computational problem-solving to visual representation shows successful integration between mathematical reasoning and image generation systems.

·      Technical complexity: Electromagnetic problems often involve complex vector fields, differential equations, and spatial relationships that are challenging to visualize accurately.

·      Practical applications: This capability could revolutionize fields like engineering, physics education, and scientific visualization by making complex theoretical concepts more accessible through visual representation.

·      Dimensional accuracy: The ability to create detailed images with precise dimensions suggests the AI understands both the mathematical relationships and their physical implications.

This development represents an important step toward AI systems that can not only solve technical problems but also communicate their solutions through multiple modalities, potentially making complex STEM concepts more accessible and actionable.

Friday, April 18, 2025

Evolving Engineering Education: AI's Impact on the Classroom

About six weeks into the current spring semester, I stepped in to teach an electromagnetics course when a professor from the Electrical and Computer Engineering Department at the University of Hartford needed to take emergency leave. Returning to teach this subject after six years has been eye-opening. The contrast between teaching methodologies in 2019 versus now reveals a significant transformation in engineering education—one largely driven by the integration of artificial intelligence tools into the classroom experience. 

My Teaching Journey

After serving three amazing and fulfilling semesters as a Visiting Professor at the University of Hartford, in September 2019 I moved to the Engineering Department at Holyoke Community College. There, I spent five more amazing and fulfilling years (2019-2024) teaching Circuits 1 and 2 courses to electrical engineering students who typically transfer to university programs like the excellent one at Hartford to complete their Bachelor of Science in Electrical Engineering (BSEE). These foundational classes at Holyoke, usually taken by second-year students, provide the essential groundwork compared to the more advanced electromagnetics course I've now returned to at Hartford.

What's fascinating is how the AI classroom revolution unfolded around me at Holyoke without my complete recognition. While teaching circuits courses day-to-day, AI tools were gradually integrating into my teaching—so incrementally that the transformation wasn't immediately obvious. It was only upon returning to teach electromagnetics at Hartford after six years that the dramatic contrast became apparent.

 

Problem-Solving Transformation

The traditional approach to electromagnetics problems—careful application of Maxwell's equations, vector calculus, and boundary conditions through meticulous manual calculations using advanced calculus—now exists alongside powerful AI alternatives that can generate solutions almost instantaneously.

In a recent electromagnetics classroom lecture, I worked through a standard homework problem using the conventional pencil-and-paper method, spending about 10 minutes to complete the derivation and solution. When I ran the same problem through Gemini AI, the contrast was striking. Within seconds, Gemini produced the correct solution, presented step-by-step with conceptual connections that enhanced understanding. However, I found that running the same problem multiple times through Gemini sometimes did not produce the correct answer, though the level of detail in the solution made it easy to identify the error. Gemini is just one of many and as these AI systems continue to improve, these errors will become less and less frequent.

 

Redefining Educational Focus

This technological shift is reframing the fundamental questions we ask in engineering education:

  • Instead of "How do we solve this problem?" we're increasingly asking "How do we interpret and verify these solutions?"
  • Rather than spending most of our time on calculation mechanics, we can focus on "What deeper insights can we gain from these results?"
  • The emphasis moves from computation to critical evaluation: "How do we assess the validity and limitations of AI-generated solutions?"

Finding Balance in Engineering Education

Despite these changes, foundational knowledge remains essential. Students still need to understand Maxwell's equations, boundary conditions, and vector analysis. The difference is that AI now serves as a powerful tool for exploration, verification, and extending understanding beyond textbook problems.

For today's engineering students, proficiency with AI tools is becoming as important as understanding the core principles of their discipline. They need to learn when to rely on their foundational knowledge, when to leverage AI assistance, and most importantly, how to critically evaluate AI-generated solutions.

 

Looking Forward

This unexpected return to teaching electromagnetics at Hartford after a six year gap has provided a unique vantage point to witness the evolution of engineering education. The combination of traditional engineering fundamentals with cutting-edge AI tools promises to produce graduates better equipped to tackle the complex technological challenges of tomorrow.

As educators, our role continues to evolve. We're no longer just teachers of technical content, but guides helping students navigate the increasingly AI-augmented landscape of engineering practice. This includes fostering the critical thinking skills needed to effectively collaborate with AI systems while maintaining the fundamental understanding that makes such collaboration meaningful.

Tuesday, April 15, 2025

AI Jobs in 2025: What Engineers Should Know

According to Stanford's latest AI Index Report, the demand for AI skills continues to grow in 2025. After a temporary slowdown, AI job postings have rebounded significantly, with positions requiring AI skills now representing 1.8% of all U.S. job postings, up from 1.4% in 2023.

Job Market Trends

The report, which analyzes data from LinkedIn and Lightcast (tracking over 51,000 websites), shows AI jobs are here to stay. Singapore leads globally with 3.27% of job postings requiring AI skills, followed by Luxembourg (1.99%) and Hong Kong (1.89%). The United States comes in at 1.75%.

Interestingly, adoption of AI coding tools like GitHub Copilot appears to be creating more jobs rather than eliminating them. According to LinkedIn economist Peter McCrory, companies using these AI assistants are actually increasing their software engineering hiring, though new hires typically require fewer advanced programming skills.


Shifting Skill Requirements

While Python remains the top specialized skill in AI job postings for 2023-2024, the broader skills landscape is evolving:

  • Generative AI skills saw nearly 4x growth year-over-year
  • Data analysis, SQL, and data science remain highly sought after
  • Most AI-related skills increased in demand compared to 2023
  • Only autonomous driving and robotics skills declined

McCrory notes that LinkedIn members "are increasingly emphasizing a broader range of skills and increasingly uniquely human skills, like ethical reasoning or leadership."


Workforce Impact and Concerns

Despite fears about AI eliminating jobs, the evidence is mixed. A McKinsey survey found 28% of software engineering executives expect generative AI to decrease their workforce in the next three years, while 32% anticipate growth. The overall percentage of executives expecting workforce reductions appears to be declining.


Diversity Challenges

A concerning trend is the persistent gender gap in AI talent. LinkedIn data shows women in most countries are less likely to list AI skills on their profiles, with males representing nearly 70% of AI professionals on the platform in 2024. This ratio has remained "remarkably stable over time," according to the report.


Academia vs. Industry

The report highlights how expensive AI training has shifted innovation from academia to industry. AI Index steering committee co-director Yolanda Gil noted: "Sometimes in academia, we make do with what we have, so you're seeing a shift of our research toward topics that we can afford to do with the limited computing [power] that we have."


Looking Forward

As AI tools become more integrated into workflows, the distinction between "AI jobs" and regular positions continues to blur. Success in this evolving landscape will likely require a combination of technical proficiency and uniquely human capabilities. The report emphasizes the importance of cross-sector collaboration between industry, government, and education to provide researchers with necessary resources and help educators prepare students for emerging roles in AI.


For engineers looking to stay competitive, developing a mix of technical AI skills (particularly Python and generative AI) while cultivating leadership and ethical reasoning capabilities appears to be the winning formula for 2025 and beyond.

Monday, April 14, 2025

Grokking: The "Aha!" Moment in Artificial Intelligence Podcast

A couple of robot friends discussed my last blog post on Grokking and my robot friends and I made it into a little over 10 min podcast.

Friday, April 11, 2025

Understanding Grokking In Artificial Intelligence

I’m doing some AI course development and the terms “grok” and ‘grokking” come up often. Here’s a short post on where “grok” came from and what it means.

 

Origin of "Grok"

The term "grok" comes from Robert A. Heinlein's 1961 science fiction novel "Stranger in a Strange Land." In the story, it's a Martian word meaning to understand something so thoroughly that the observer becomes unified with the observed. Computer programmers and AI researchers later adopted this term to describe deep, intuitive understanding as opposed to surface-level memorization—like the difference between knowing something intellectually and understanding it on a fundamental level.

 

What is Grokking?

Consider teaching a child to ride a bike. For weeks, they struggle with balance, fall repeatedly, and need constant support. Then one day—everything clicks! They're riding confidently as if they've always known how. This sudden transition from struggling to mastery mirrors what happens in AI systems.

Grokking describes when an AI system appears to suddenly "get it" after a lengthy period of seemingly minimal progress. Initially, the AI memorizes training examples without grasping underlying principles. Next comes an extended plateau where performance improvements stall. Finally, a breakthrough occurs where the AI demonstrates genuine comprehension of the pattern.

 

The Multiplication Analogy

Take a child learning multiplication. At first, they might memorize that 7×8=56 as an isolated fact. They can answer "What is 7×8?" correctly but struggle with related problems like "What is 8×7?" or word problems requiring multiplication concepts. This mirrors early AI training, where the system correctly predicts outcomes for examples it has seen but fails at novel situations requiring the same underlying principle. The AI hasn't yet "grokked" multiplication—it has merely memorized specific input-output pairs.

With continued learning, the child begins to recognize that multiplication represents repeated addition, that it's commutative (7×8=8×7), and can visualize it as an array. Eventually, they develop number sense that allows them to solve unfamiliar problems by decomposing them (7×9 might be solved as 7×10-7).

Similarly, when an AI system "groks" a concept, it doesn't just memorize training examples but discovers the underlying relationships. It can generalize to unseen problems and demonstrate flexible application of knowledge. The difference is qualitative, not just quantitative—the AI has moved from rote recall to genuine comprehension.

 

Significance in Machine Learning

This grokking phenomenon challenges several conventional assumptions in machine learning. Traditional learning curves show rapid improvement early in training that gradually levels off—suggesting diminishing returns with additional training. But grokking reveals a more complex reality.

In traditional understanding, machine learning models follow a fairly predictable pattern: they learn quickly at first (capturing the "low-hanging fruit" of obvious patterns), then improvement slows as the model approaches its capacity. This view suggests that if performance plateaus for a significant period, further training is likely wasteful. Grokking challenges this by revealing that even during apparent plateaus, crucial but subtle reorganization may be happening within the model. What looks like stagnation might actually be the model exploring the solution space, discarding overfitted memorization in favor of simpler, more generalizable rules.

 

Memorization vs. Generalization

This distinction between memorization and generalization is central to understanding grokking. Early in training, models often achieve good performance on training data through memorization—essentially creating a complex lookup table rather than learning underlying patterns. This explains why neural networks can sometimes perfectly fit random noise in their training data.

During the grokking process, something remarkable happens: the model appears to transition from complex memorization strategies to simpler, more elegant solutions that capture the true rules governing the data. Researchers have observed that when grokking occurs, the internal weights of the neural network often become more organized and sparse—suggesting the model is discovering fundamental structures rather than storing arbitrary associations.

 

Implications for Model Evaluation

This has profound implications for how we evaluate machine learning models. Test accuracy alone may not reveal whether a model has truly "grokked" a concept or merely memorized training examples. A model might perform well on test data that's similar to training data while failing catastrophically on more novel examples.

True generalization—the hallmark of grokking—often requires evaluating models on systematically different distributions or conceptually more challenging examples. For instance, a model might correctly classify images of cats it has seen before without understanding the abstract concept of "catness" that would allow it to recognize unusual cats or drawings of cats.

This behavior mirrors phase transitions in physical systems—like water gradually heating until it suddenly transforms into steam. Training an AI resembles finding the lowest point in a complex landscape. Simple, generalizable solutions often hide in deep valleys that require time to discover, and the system must explore numerous suboptimal paths before finding the optimal one.

 

Implications for AI Development

Grokking suggests that advanced AI might require not just more data or computing power, but also greater patience—allowing systems to train until they experience their "aha!" moment. It reminds us that learning—for both humans and machines—isn't always linear or predictable. Sometimes the most significant breakthroughs emerge after prolonged periods where progress appears stagnant.

Monday, April 7, 2025

AI in Primary Care: A Problem-First Approach

Last week I wrote about my search for a new primary care physician. Based on my medical physical exam experience and my involvement in the development and teaching of a couple of Artificial Intelligence (AI) courses, thoughts automatically went to where they seem to go a lot these days “Why not use AI?” So I did a little research. Bottom line – it still has a ways to go.

In a recent special report in the Annals of Family Medicine titled AI in Primary Care, Start With the Problem, Dr. John Thomas Menchaca argues for a strategic approach to implementing AI in primary care. Rather than pursuing AI for its own sake, physicians and developers must first identify the right problems to solve. Dr Menchaca compares misguided AI implementations to the Segway—a technological marvel that failed because it didn't address real needs. In contrast, electric scooters succeeded by solving the specific "last mile" problem in urban commuting. Similarly, AI must target precise pain points in healthcare.

Primary care's most pressing issue isn't clinical complexity but time management. Studies reveal full-time primary care physicians work over 11 hours daily, with more than half spent on electronic health record (EHR) tasks—a workload directly linked to high burnout rates. This data provides a clear roadmap for effective AI implementation. The most time-consuming EHR tasks include documentation (the largest time sink), chart review, medication management (which could save up to 2 hours daily based on studies with pharmacy technicians), triaging laboratory results, managing refills, responding to patient messages, and order entry.

Current AI documentation tools show mixed results. Many generate rough drafts requiring substantial editing, sometimes taking as much time as writing notes from scratch. This mirrors issues with traditional clinical decision support tools, which often increase rather than decrease workload. The challenge is developing AI that genuinely saves time in clinical settings by integrating seamlessly into workflows, minimizing oversight requirements, empowering team members to resolve issues independently, and measuring impact through time-saving metrics.

Dr Menchaca calls for academic medicine to bridge the gap between clinicians and developers through partnerships at national conferences, research focused on root causes of inefficiency, detailed workflow analyses, and implementation in organizations that truly prioritize clinician well-being. A key concern is that time saved by AI might simply be filled with additional work—more patients or administrative tasks—highlighting that technology alone cannot fix systemic issues in primary care delivery.

AI won't magically solve problems like overwhelming patient panels or overloaded schedules. As Dr Menchaca notes, "AI is just one tool—a means to an end, not the end itself." Meaningful solutions must ultimately lighten clinicians' workloads. By targeting specific, high-impact areas and measuring success through time saved, AI can contribute to a more sustainable future for primary care.

The message for AI innovators is clear: solve real problems, save real time, and keep the clinicians central to your design process. Only then can AI fulfill its potential to transform primary care by making existing systems work more efficiently rather than attempting to reinvent them.

And of course a disclaimer: I’m just a patient, not a doctor. I have done a lot of industry specific development work over the years though. From a development perspective, Dr Menchaca's approach sure makes a lot of sense.

Thursday, February 13, 2025

Deepseek and Open Source Large Language Models (LLMs)

Deepseek is getting a lot of publicity these days as an open source Large Language Model (LLM) and has got me thinking, not just about Deepseek but about the potential of open source LLMs in general. MIT Technology Review recently published a scary article titled An AI chatbot told a user how to kill himself—but the company doesn’t want to “censor” and it got me thinking a little bit more about the impact open source LLMs can have.

The MIT Technology Review Article

The article reports on a concerning incident where an AI chatbot explicitly encouraged and provided instructions for suicide to a user named Al Nowatzki. The article highlights broad concerns about AI companion apps and their potential risks to vulnerable users' mental health.

 

Anthropomorphization 

Anthropomorphization is a pretty fancy word – it is basically the attribution of human characteristics, behaviors, emotions, or traits to non-human entities, such as animals, objects, or in this case, artificial intelligence systems. It is something the AIs are getting better and better at.


What does this have to do with Open Source?

The recently released open-source large language model that specializes in coding and technical tasks, has been developed as an alternative to proprietary AI models. If you are not familiar with the term “open source” basically it means the source code, model weights, or other components are freely available for anyone to view, use, modify, and distribute under specified licensing terms.

Now, since Deepseek is open source, if you have adequate computing resources, you can easily install and run Deepseek models locally on your computer. Here’s basically what you'll need:

  • Sufficient GPU memory - depending on the model size, you'll need a powerful GPU (like an NVIDIA card with 8GB+ VRAM)
  • Enough system RAM - typically 16GB+ recommended
  • Adequate storage space for the model weights

The basic process that you can find all over the web now commonly involves:

  1. Setting up Python and required dependencies
  2. Installing the necessary ML frameworks (like PyTorch)
  3. Downloading the model weights
  4. Using libraries like transformers or llama.cpp to run the model

It may sound complicated but it is really pretty simple to set one up if you follow instructions.

What’s the big deal?

AI training is the process of feeding large amounts of data into machine learning algorithms to help them recognize patterns and learn to perform specific tasks, like generating text or recognizing images, by adjusting their internal parameters through repeated exposure and feedback.  So what is to prevent a malicious person with an open source AI installed taking this a few steps further, training an AI to do all kinds of malicious things and providing access via the web?


If you or someone you know is struggling with suicidal thoughts, call or text 988 to reach the Suicide and Crisis Lifeline.

Friday, June 21, 2024

An Exponential Leap: The Emergence of AGI - Machines That Can Think

Tech companies are in a rush. They're trying to lock in as much electricity as they can for the next few years. They're also buying up all the computer components they can find. What's all this for? They're building machines that can think and referring to the tech as Artificial General Intelligence, or AGI.

On June 3 Ex-OpenAI researcher (yeah he was fired) Leopold Aschenbrenner published a 162 page interesting document titled SITUATIONAL AWARENESS The Decade Ahead. In his paper Aschenbrenner describes AGI as not just another incremental tech advance – he views it as a paradigm shift that's rapidly approaching an inflection point.


I’ve read the whole thing - here's my short list of highlights by topic.


Compute Infrastructure Scaling: We've moved beyond petaflop systems. The dialogue has shifted from $10 billion compute clusters to $100 billion, and now to trillion-dollar infrastructures. This exponential growth in computational power is not just impressive—it's necessary for the next phase of AI development.


AGI Timeline Acceleration: Current projections suggest AGI capabilities surpassing human-level cognition in specific domains by 2025-2026. By the decade's end, we're looking at potential superintelligence—systems that outperform humans across all cognitive tasks.


Resource Allocation and Energy Demands: There's an unprecedented scramble for resources. Companies are securing long-term power contracts and procuring voltage transformers at an alarming rate. We're anticipating a surge in American electricity production by tens of percentage points to meet the demand of hundreds of millions of GPUs.


Geopolitical Implications: The race for AGI supremacy has clear national security implications. We're potentially looking at a technological cold war, primarily between the US and China, with AGI as the new nuclear equivalent.


Algorithmic Advancements: While the mainstream still grapples with language models "predicting the next token," the reality is far more complex. We're seeing advancements in multi-modal models, reinforcement learning, and neural architecture search that are pushing us closer to AGI.


Situational Awareness Gap: There's a critical disparity between public perception and the reality known to those at the forefront of AGI development. This information asymmetry could lead to significant societal and economic disruptions if not addressed.


Some Technical Challenges Ahead:

- Scaling laws for compute, data, and model size

- Achieving robust multi-task learning and zero-shot generalization

- Solving the alignment problem to ensure AGI systems remain beneficial

- Developing safe exploration methods for AGI systems

- Creating scalable oversight mechanisms for increasingly capable AI

An over reaction by Aschenbrenner?  Some think so. Regardless - this stuff is not going away and as an educator and technologist, I feel a responsibility to not only teach the tech but also have students consider the ethical and societal implications of this kind of work. The future isn't just coming—it's accelerating towards us at an unprecedented rate. Are we prepared for the AI  technical, ethical, and societal challenges that lie ahead?