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.

Tuesday, January 7, 2025

Characterizing Student Perspectives on Departmental Culture

The AAAS-IUSE initiative, funded by the NSF Improving Undergraduate STEM Education: Directorate for STEM Education (IUSE: EDU) posted an interesting read last month. If you are not familiar – the AAAS-IUSE Initiative supports faculty, students, and the greater undergraduate STEM education community by disseminating research and knowledge about STEM teaching, learning, equity and institutional transformation.

 

The post that caught my eye discusses how traditional metrics for evaluating student success (like graduation rates) are insufficient because they only measure outcomes at the end of a student's academic journey. 

 

It introduces what the authors are calling the student DELTA survey (s-DELTA) as a new tool to measure departmental culture from students' perspectives, with the goal of improving diversity, equity, inclusion, and accessibility.


Key points:

  1. Departmental culture significantly impacts student experience, and departments often have more distinct cultures than institutions as a whole. This makes departments an effective focus for implementing change.
  2. The s-DELTA survey is based on six core principles, including student partnership, collective outcomes, data-informed decisions, collaboration, continuous improvement, and commitment to equity.
  3. The survey structure asks students to evaluate both their current department and their ideal department using a six-point Likert scale, allowing for comparison between reality and aspirations.
  4. Initial findings showed significant differences between current and ideal department ratings, particularly in areas related to equity and inclusion. Gender-based analysis revealed that women and men often had different experiences of departmental culture.

A great start. While the s-DELTA survey needs further psychometric testing, it can provide a valuable tool for:

    • Tracking student perspectives over time
    • Informing departmental policy development
    • Supporting diversity and inclusion efforts
    • Helping departments understand and improve their cultural impact on student experience

Worth a look for academics. Here’s a link if you are interesting in reading the original post: https://aaas-iuse.org/characterizing-student-perspectives-on-departmental-culture-the-student-delta-survey/

Wednesday, December 4, 2024

Quantum Communications - Part 3: Photon Polarization and Superposition

 Building on Part 2’s discussion of polarization, let's look a little deeper at how light enables quantum communication. Superposition in quantum mechanics means a photon can exist in multiple polarization states at once, unlike classical objects that must be in one definite state. Looking at the diagram below, a photon isn't limited to being just vertical (1) or horizontal (0), but can exist in a mixture of both until measured. Think of it like a spinning coin - while spinning, it's neither heads nor tails but both possibilities at once. When we measure the photon's polarization (like catching the coin), it "collapses" into one definite state.

The diagram shows two measurement bases - rectilinear and diagonal. A photon in superposition measured in either base has a probability of being found in either state of that base. This property is crucial for quantum cryptography because any measurement by an eavesdropper forces the superposition to collapse, altering the photon's state and revealing the intrusion.

 

The 45° polarization state, shown in the diagonal base of our diagram, demonstrates one of the most fascinating aspects of quantum mechanics. While we might classically think of 45° as simply an angle halfway between horizontal and vertical, in quantum mechanics it represents something far more profound. When a photon is polarized at 45°, it literally exists in a perfect blend of horizontal and vertical states at the same time - not just leaning one way or the other, but fully in both states simultaneously.

 

This quantum behavior becomes clear when we measure these 45° polarized photons. If we measure using the rectilinear base (horizontal/vertical as shown in the left circle of our diagram), we get the following result: the photon will randomly show up as either horizontal or vertical with exactly equal probability. This isn't because we're measuring imprecisely or because the photon was "kind of" in both states - it was genuinely, mathematically, and physically in both states at once until the act of measurement forced it to pick one.

 

Think of it this way: if we send many 45° polarized photons through a horizontal polarizer, exactly half will pass through (registering as horizontal) and half will be blocked (registering as vertical). This isn't due to some classical "angled" behavior - it's a direct manifestation of quantum superposition, where the photon existed in both states simultaneously until we forced it to "decide" by measuring it.

 

This property is part of what makes quantum communication so secure: any attempt to measure these superposition states unavoidably disturbs them, making eavesdropping detectable.

Tuesday, November 26, 2024

Quantum Communications - Part 2: Polarization

In Part 1 of this series I discussed how quantum superposition allows particles to exist in multiple states at once until measured. This makes quantum information almost impossible to copy secretly, enabling ultra-secure communication systems that can detect eavesdropping attempts. Here we'll take a closer look at how this information is transmitted using light particles called photons. 


Light can be analyzed as either a ray or a wave, each model revealing different aspects of its behavior. The ray model treats light as straight lines traveling through space, useful for understanding reflection, refraction, and how lenses and mirrors work. The wave model shows light as oscillating electromagnetic waves, explaining phenomena like interference, diffraction, and polarization. While the ray model helps us design simple optical devices like eyeglasses, the wave model is necessary for understanding more complex effects like how polarizing filters work. Both models remain important in modern optics, with each being used depending on which aspects of light's behavior are most relevant to the situation at hand.

 

When it comes to polarization we need to think about light as a wave. Light waves oscillate perpendicular to their direction of travel, Unpolarized light ( like we get from natural sources like the sun or artificial sources like light bulbs) vibrates in all possible directions like a rope being waved up-down, side-to-side, and at every angle in between. A polarization filter works like a microscopic venetian blind with extremely fine parallel slits. When light encounters the filter, only the waves that vibrate parallel to these slits can pass through completely. Waves vibrating in other directions are either blocked entirely or have only their parallel components transmitted. The light that emerges from the filter is polarized, meaning all the waves are vibrating in the same direction. 


Here's a quick 3 second video I made demonstrating light passing through a polarizing filter.


Before the filter: 
  • Horizontal component (blue wave)
  • Vertical component (pink wave)
  • 45-degree component (green wave, thicker line) All components are present in the unpolarized light
The Filter: 
  • Oriented at 45 degrees
  • Only allows waves aligned with its transmission axis
After the filter: 
  • Only the 45-degree component (green wave) passes through
  • Horizontal and vertical components are blocked
  • The transmitted light is polarized along the 45-degree axis

Notice:

  • Only transmits light waves that oscillate parallel to its transmission axis (45 degrees in this case)
  • Waves at other angles are either blocked or have only their 45-degree component transmitted
  • The result is polarized light oscillating only at 45 degrees
This selective transmission property makes polarizing filters particularly useful in everyday applications. For example, polarizing sunglasses can effectively reduce glare because light reflecting off horizontal surfaces like water or roads tends to become partially polarized in the horizontal direction. The sunglasses, which have vertical polarizing slits, block this horizontal glare while still allowing other light through. A demonstration of how polarization works involves using two polarizing filters. When light passes through the first filter, it becomes polarized in one direction. If you then rotate a second filter 90 degrees relative to the first, no light gets through at all because the polarized light from the first filter is now perpendicular to the slits in the second filter.

 

So what does polarization have to do with quantum communications? In quantum communications individual photons can be prepared in specific polarization states (vertical, horizontal, or diagonal) to represent quantum bits. Due to quantum mechanics principles, any attempt to measure these polarization states disturbs them, making secure communication possible - eavesdropping can be detected. Polarization also enables quantum entanglement, where measuring one photon's polarization instantly determines its entangled partner's state, even at a distance.

 

While powerful, polarization methods face practical challenges as polarization states can degrade during transmission through optical fibers or atmosphere, requiring sophisticated error correction methods. For these reasons, quantum communications can use several alternatives to polarization for encoding quantum information. Time-bin encoding uses photon arrival times and works well in fiber optics where polarization degrades. Phase encoding utilizes phase differences between photon paths, while frequency encoding uses different photon frequencies. Orbital Angular Momentum (OAM) encoding exploits spiral patterns of light waves, potentially carrying more information than polarization. Path encoding, which uses different physical routes for photons, is useful in integrated photonic circuits. Each of these methods has its own advantages and the choice often depends on the specific application and transmission medium being used. For instance, time-bin encoding tends to be more robust for long-distance fiber communication, while OAM can potentially carry more information per photon. 


Thursday, November 21, 2024

Quantum Communications – Part 1

 Computers, communications, photonics, cybersecurity…… some of my favorite technologies all bundled together in quantum communications! But… what is it? How does it work? Let's take an introductory look.

In quantum mechanics, something called superposition allows systems to exist in multiple states simultaneously - like a spinning coin being both heads and tails at once, until observed. Only measurement forces it into a definite state. This principle affects quantum information through the no-cloning theorem, which states that it's impossible to create an exact copy of an unknown quantum state due to quantum mechanics' mathematical foundations. This feature enables quantum key distribution (QKD), which creates unbreakable encryption keys. Any eavesdropping attempt disturbs the quantum states due to the no-cloning theorem, instantly revealing the intrusion. While classical information can be copied perfectly, quantum information's resistance to copying both protects it and makes quantum teleportation the only way to transfer quantum states.

 

How about an example? Let’s now compare sending a classical letter and sending a quantum letter using two diagrams I’ve created. These diagrams split into two parallel workflows showing classical versus quantum communication highlight the key security advantages of quantum communication over classical methods.


 




The classical letter path shows a letter that can be intercepted, read, and copied without detection as it moves from sender to recipient through the postal system.


The quantum letter path illustrates how quantum letters behave differently:

  • The letter exists in superposition (blue state) until measured
  • Any attempt to read/copy disturbs the quantum state (changes to red)
  • This disturbance is detectable when received by the recipient, revealing tampering

We’ll dig a little deeper in future posts – for now think of quantum communication as the first step toward a quantum internet - one that operates on the powerful principles of quantum physics rather than classical physics. While we may not see quantum email on our phones anytime soon, the technology is steadily advancing from science fiction toward practical reality.

Thursday, November 14, 2024

Time Traveling Technology

Back then I was always in the PC camp; Apple had their market, but the price point kept me away and the PC was more about business.


Scrolling through vintage tech photos online, I feel a pang of regret for not documenting my own journey. A 1974 CP/M terminal catches my eye - those phosphor-green screens where we typed cryptic commands, unwittingly shaping the digital future. Back then, RS232 serial ports were our lifeline machine to machine, carefully configuring stop bits and parity. Direct connections between computers required null modem adapters or cables with crossover wiring - TX to RX, RTS to CTS, the careful dance of handshaking signals.


Before the internet as we know it, there was TYMNET - that pioneering packet-switched network connecting terminals across the country. The hiss and static of dial-up modems negotiating connections to local terminals providing access to remote systems through X.25 protocols. The world seemed smaller as we connected across continents through packet-switched networks, each X.25 virtual circuit opening new possibilities.


The memories flow: DOS in the '80s, watching that C:\ prompt blink as we crafted batch and config.sys files and navigated directory structures. Serial connections ruled - null modem cables letting machines talk, COM ports needed constant coaxing. Remember XMODEM and KERMIT file transfers? Hours spent watching progress bars crawl across the screen, praying the connection wouldn't drop. My first local area network ran on ARCnet, with its characteristic coax cables and T-connectors linked by 93-ohm terminators. Troubleshooting always started with "Wiggle the coax!" - somehow, that gentle twist of BNC connectors often restored life to the network.


Ethernet brought its own challenges - thick yellow "frozen garden hose" cables threading through office ceilings. Installing vampire taps demanded precision: clamp them just right,  the perfect depth to contact the core. One mistake meant cutting, splicing, and starting over. But a successful connection brought pure joy.

 

And let’s not forget Token Ring, IBM's answer to networking. Those distinctive MAUs enabling token passing, ensuring orderly network access. Type 1 cable with its hefty copper shielding snake through the walls, each connection requiring perfect crimping of those specialized connectors. The satisfaction of seeing all stations "actively monitor present" on the ring - when it worked, it was bulletproof but so expensive.


OS/2 was a brief chapter - true multitasking, the Workplace Shell, and an object-oriented interface years ahead of its time. Though the market passed it by, tech enthusiasts still remember its innovation.


Windows evolved rapidly: 3.1, then Windows for Workgroups 3.11 that ran on top of DOS and revolutionized business and industry computing. Finally, sharing files and printers became accessible. Configuring peer-to-peer networks and IRQ settings was complex but rewarding - each successful file transfer felt like a small miracle.


1995 brought Windows 95, and I captured that excitement in my book "Windows 95 for Engineers." Features we take for granted - the Start button, taskbar, plug-and-play - seemed revolutionary then.


Looking at my modern setup, these milestones feel surprisingly close. Fifty years of technology have flowed past like silicon sand, each innovation building on its predecessors, each memory marking my path through the digital age.


I wonder what that CP/M dude would think of today's world. Would they see their work's legacy in our systems? Technology time moves strangely - crawling in anticipation, yet lightning-fast in memory.

Friday, October 25, 2024

Testing A Fuse Using Your Cell Phone

 

Capacitive touch screens dominate modern mobile devices due to their precision and reliability. These screens operate on principles of electrical capacitance - the ability of a system to store an electrical charge. The screen comprises a layer of electrically conductive indium tin oxide (ITO) protected by insulating glass. This arrangement creates a uniform capacitive field across the display surface. When a conductive object, such as a human finger, contacts the screen, it alters the local capacitance by several picofarads. High-precision sensors detect and measure these minute electrical changes to determine the exact contact location.


If you don't have a meter handy - here’s a 12 second video demonstrating how to test a fuse with your phone. Notice one end of the fuse is in contact with my fingers while the opposite end makes contact with the phone screen.

 

By comparison, resistive touch screens, commonly not found on phones but found in industrial interfaces and ATMs, employ a pressure-based mechanism. This alternative technology sacrifices the responsiveness of capacitive systems for universal input compatibility.