Dynamic Subspace

Tag: Text Generation

  • Some Notes for the Empowering Faculty with AI Roundtable Discussion Today

    an anthropomorphic cat professor wearing a shirt, tie, suspenders, and pants is standing in the middle of a vintage computing room lined with minicomputers and a chalkboard is in the background on the wall
    Image created with Stable Diffusion.

    I collected the following notes, resources, and readings to share during today’s Empowering Faculty with AI roundtable discussion organized by City Tech’s Academic Technologies and Online Learning (AtoL).

    These resources are divided into these sections: Writing and Editing, Research and Experimentation, Teaching, and Readings for Faculty.

    I tend not to trust a computer that I can’t throw out the window, so I only use local Generative AI models that I can run on my desktop workstation or Lenovo ThinkPad P1 Gen 4 laptop. My primary interests are in text generation and image generation, which I have documented over the past three years here on DynamicSubspace.net.

    Writing and Editing

    • I used Generative AI to edit my writing and create the cover image for Yet Another Science Fiction Textbook (YASFT) OER that I published on my website in Feb. 2024. I included an explanation of my workflow in the front matter of the textbook.

    Research and Experimentation

    Teaching

    Assigned Readings in Professional and Technical Writing Classes

    Introduction to Language and Technology, ENG1710

    Introduction to Professional and Technical Writing, ENG2700

    • Reeves, Carol and J. J. Sylvia IV. “Generative AI in  Technical Communication: A Review of Research  from 2023 to 2024.” Journal of Technical Writing and Communication, vol. 54, no. 4, 2024, pp. 439-462, https://doi.org/10.1177/00472816241260043.

    My AI Use Policy

    There’s no doubt (at least in my mind) that the current state of Artificial Intelligence (AI) and Generative AI is nothing short of incredible. There are certain tasks AI is good at, there are other tasks AI is not good at, and in all tasks, there is a greater than zero chance that the AI fails at its task horribly and potentially catastrophically. There are legal and ethical considerations, especially for those beginning careers as professional writers and communicators. So, where does AI fit into education, and more specifically, our class? The goals of education include giving students unique experiences for learning, research, and collaboration, as well as opportunities to demonstrate learning, insight, and growth through exams, papers, and projects. Undertaking higher education, changes your brain in deep and important ways that enable you to begin a professional career as a knowledgeable problem solver who is also capable of continuing as a life-long learner. AI’s entrance introduces a double-edged sword. On the one hand, it can be used as a tool to support your work and learning, but it can also be used egregiously to plagiarize and violate the CUNY Academic Integrity Policy. To avoid any misconceptions or misunderstandings, and to support your best learning experience in our class, you may not use AI at all in our class–including email, summarizing, ideation, discussion, or writing–unless the exercise or assignment explicitly states that you may. I am instituting this policy, because I want you to face the challenge of the class using your human abilities, skills, and talents. It’s only through those challenges will you grow and develop into a professional who can do more and find greater success than those who recklessly rely on AI technologies to do their thinking and work. Of course, violations of this policy are simply violations of the Academic Integrity Policy (under definitions 1.1 Cheating, 1.2 Plagiarism, and 1.3 Obtaining Unfair Advantage), which may result in failure and referral to the college.

    Alexander Chee’s Plea on Bluesky

    Alexander Chee, writer and Dartmouth College professor, writes on Bluesky: "This will sound weird but love yourself enough to write your own term papers. Having a LLM or another person writing your papers feels like a victor but it is just your own defeat, you felled by your own contempt for your future and anyone who believes in you."

    Alexander Chee, writer and Dartmouth College professor, shares on Bluesky: “This will sound weird but love yourself enough to write your own term papers. Having a LLM or another person writing your papers feels like a victor but it is just your own defeat, you felled by your own contempt for your future and anyone who believes in you.”

    Readings for Faculty

    I’ve recently been experimenting with so-called reasoning LLM models. For the synopses below, I gave my reading notes to a 6.0bpw quant of DeepSeek’s R1 Distill of LLaMA 70b LLM asking it to turn my telegraphic points into a brief paragraph of prose. Giving the model the “<thinking>” tag before the model’s response engaging the reasoning capability, which tends to provide better results and instruction following. The information that went into the summaries are based on my notes, so any errors or misunderstandings are mine.

    • Werse, Nicholas R. “What Will Be Lost? Critical Reflections on ChatGPT, Artificial Intelligence, and the Value of Writing Instruction.” Double Helix, vol. 11, 2023, https://wac.colostate.edu/docs/double-helix/v11/werse.pdf.
      • Werse examines the implications of ChatGPT—a highly advanced text-generating AI developed by OpenAI—on the future of writing instruction in higher education. He argues that while ChatGPT can mimic human writing and even assist in research and writing processes, over-reliance on this technology risks undermining the critical thinking and deep understanding that writing fosters. He highlights that writing is not just a practical skill but a cognitive process that externalizes thought (citing Kellogg’s The Psychology of Writing–see below), allowing writers to explore connections, organize ideas, and reflect deeply on a topic. Werse acknowledges the benefits of AI as a tool but warns that outsourcing writing to AI may shortchange students of the educational value of engaging in the iterative writing process. He also discusses the broader societal implications, such as how AI could reshape the role of writing in professional contexts, potentially shifting its value from career preparation to personal enrichment. Werse calls for educators to rethink how they teach writing in an AI-driven world, balancing the practical benefits of technology with the enduring importance of writing as a means of learning and critical engagement.
    • Kellogg, Ronald T. The Psychology of Writing. Oxford UP, 1994.
      • I’ve only just started this book (it’s available to borrow on archive.org here), but I’ve received Kellogg’s thesis about how writing is a cognition supporting technology and activity through many other readings and lessons (though, not always citing Kellogg–it’s more like an established axiom now). Nevertheless, I believe returning to this text is important for building an argument to our students about the importance of doing their own writing for their own thinking and cognitive development.
    • Jamieson, Sandra. “The AI ‘Crisis’ and a (Re)turn to Pedagogy.” Composition Studies, vol. 50, no. 3, 2022, pp. 153-157.
      • Jamieson critiques the panic surrounding AI’s impact on academia, particularly its perceived threat to the college essay and higher education. She argues that while AI tools like ChatGPT, QuillBot, and ParagraphAI are powerful technologies capable of generating sophisticated text, they are not a crisis but an opportunity to refocus on teaching writing. She highlights that composition studies has historically responded to externally defined crises, such as concerns over literacy, cheating, and plagiarism, by adapting pedagogical practices. Jamieson emphasizes that AI is simply another tool, akin to computers or calculators, that can be integrated into teaching. She suggests using AI as a brainstorming or collaborative tool to enhance student learning, rather than viewing it as a threat to be policed. This approach aligns with broader pedagogical values, encouraging educators to trust students and teach them to use AI ethically and effectively. Jamieson also addresses labor issues within academia, critiquing the exploitation of contingent faculty and the mistrust embedded in surveillance technologies. She advocates for a “re-turn” to foundational teaching methods, emphasizing the process of writing—revision, invention, and critical thinking—as a way to engage students deeply with their work. Jamieson sees AI as an opportunity to strengthen writing education rather than undermine it.
    • Carver, Joseph and Samba Bah. “Rehumanizing Higher Education: Fostering Humanity in the Era of Machine Learning.” New Directions for Teaching and Learning, 2025, https://doi.org/10.1002/tl.20644.
      • Carver and Bah examine the challenges and opportunities posed by artificial intelligence (AI), particularly generative AI, within higher education. They highlight the polarized attitudes toward AI, ranging from optimism about its potential to enhance research and efficiency to concerns about its impact on critical thinking and human values. They draw parallels to the “Science Wars” of the 1990s, which questioned the validity of soft sciences, and note the lack of consensus on AI policies among academics. The article emphasizes the historical development of AI, from Alan Turing’s foundational work to modern advancements in machine learning (ML) and deep learning, which have enabled AI to mimic human reasoning and solve complex problems like protein folding. However, the authors stress the “alignment problem,” a term referring to AI’s inability to fully understand and align with human values, leading to biased outcomes in areas such as hiring and criminal justice. They argue that higher education has a unique role in shaping the future of AI, not only through research and integration of AI tools but also by rehumanizing education. They advocate for a unified response to AI, one that prioritizes human values and addresses the existential implications of AI on education’s traditional goals, such as critical thinking and knowledge creation. Drawing on philosophical traditions like Ubuntu, an African philosophy emphasizing interconnectedness and community, the authors call for a reimagined higher education system that centers humanity, empathy, and collaboration. They conclude by urging educators to ensure that AI enhances rather than diminishes human potential.
    • Liu, Larry. “AI and the Crisis of Legitimacy in Higher Education.” American Journal of STEM Education: Issues and Perspectives, vol. 4, 2024, pp. 16-26, https://www.ojed.org/STEM/article/view/7572.
      • Liu explores how the rise of Generative Artificial Intelligence is challenging the legitimacy of higher education, defined as society’s belief in the necessity of college credentials for secure, middle-class jobs and social mobility. AI’s ability to perform tasks like essay writing, exam-taking, and personalized learning threatens to disrupt traditional educational roles, with some educators even leaving the profession due to its impact. While AI could potentially boost college legitimacy by displacing jobs and increasing demand for credentials, it also poses risks, such as overreliance on AI making students less engaged learners. The article identifies three key challenges: AI’s unique capabilities, government and employer shifts toward skills-based hiring over degree requirements, and the “enrollment cliff” caused by declining birth rates. Using sociological frameworks, Liu applies Jürgen Habermas’s concept of legitimation crisis, where the state’s ability to balance economic growth and welfare is strained, to argue that higher education’s role in maintaining capitalist legitimacy is at risk. Additionally, Randall Collins’s credential inflation theory suggests that the surplus of college graduates devalues degrees, creating social hierarchies rather than preparing workers for the labor market. Liu concludes that educators should embrace AI’s benefits while addressing its broader societal impacts, such as declining enrollment and political distrust in higher education.
    • Fritts, Megan. “A Matter of Words: What Can University AI Committees Actually Do?” The Point Magazine, 12 May 2025, https://thepointmag.com/examined-life/a-matter-of-words/.
      • Fritts discusses the profound challenges artificial intelligence (AI) poses to higher education, particularly in the humanities, as universities across the U.S. grapple with how to respond to the technology’s transformative impact. Fritts, a philosophy professor, details her experiences on AI response committees at the University of Arkansas at Little Rock, where debates focus on issues like AI detectors’ accuracy, whether tools like Grammarly should be banned, and the ethical implications of students using AI for academic work. Central to these discussions is the tension between viewing AI as a tool for equity or efficiency and as a threat to human-centered education. She argues that the heart of the crisis lies in the humanities’ core mission: if the goal is to produce artifacts like essays or arguments, AI could soon replace human efforts. However, if the focus shifts to forming human persons through language and self-expression, AI becomes an existential threat by undermining the uniquely human connection between thought, language, and identity. Drawing on a philosopher like Ludwig Wittgenstein or the political theorist Alasdair MacIntyre, Fritts emphasizes that language is not just a tool but a formative part of human life, and relying on AI for communication risks eroding this essential aspect of humanity. She calls for radical policies to preserve the humanities by banning AI in classrooms and reclaiming the indispensability of human expression.
    • Luckin, Rosemary, Mutlu Cukurova, Carmel Kent, and Benedict du Boulay. “Empowering Educators to be AI-Ready.” Computers and Education: Artificial Intelligence, vol. 3, 2022, https://doi.org/10.1016/j.caeai.2022.100076.
      • Luckin et al. focus on the integration of Artificial Intelligence (AI) in education and emphasize the need for educators to become “AI-Ready.” AI is defined as the capacity of machines to simulate intelligent behavior, with applications ranging from learner-facing tools that adapt to individual needs to educator-facing systems that assist with tasks like grading and institutional support tools for resource management. Despite AI’s potential, its adoption in education lags behind other sectors due to challenges like ethical concerns, data limitations, and the complexity of educational ecosystems. The concept of AI Readiness involves understanding AI’s capabilities and limitations, particularly its lack of human-like qualities such as social intelligence and metacognition. The authors introduce the EThICAL AI Readiness Framework, a seven-step process (Excite, Tailor, Identify, Collect, Apply, Learn, Iterate) to guide educators and administrators in identifying challenges, collecting and analyzing data, and applying AI ethically. Case studies, such as one involving Arizona State University, demonstrate how this framework helps educators assess student learning behaviors and develop targeted interventions. The article concludes by stressing the importance of balancing AI’s strengths with human expertise to enhance education effectively.
    • Ellerton, Wendy. “The Human and Machine: OpenAI, ChatGPT, Quillbox, Grammarly, Google, Google Docs, & humans.” Visible Language, vol. 57, no. 1, Apr. 2023, pp. 38-52, https://muse.jhu.edu/article/893661.
      • Ellerton reflects on the evolving relationship between humans and technology, particularly in the context of the COVID-19 pandemic and the rise of generative AI. Ellerton, a design educator, explores how tools like ChatGPT, Grammarly, and Quillbot are reshaping writing and communication, blending personal narrative with critical analysis. She discusses her experimentation with AI, highlighting both its potential to enhance creativity and collaboration and its limitations, such as the risk of misinformation and ethical concerns like authorship and bias. Using autoethnography, a research method that combines personal experience with cultural critique, she examines her own interactions with AI, revealing the tensions between human qualities like metacognition and emotional intelligence and the efficiency of machines. She advocates for a balanced approach, emphasizing the need to embrace human values and critical thinking while leveraging AI’s capabilities.
    • Miller, Robin Elizabeth. “Pandora’s Can of Worms: A Year of Generative AI in Higher Education.” portal: Libraries and the Academy, vol. 24, no. 1, Jan. 2024, pp. 21-34, https://doi.org/10.1353/pla.2024.a916988.
      • Miller considers the impact of generative AI tools like ChatGPT, Bard, and Bing Chat on higher education since their rapid introduction in late 2022. She discusses the varied reactions among educators, from alarm to cautious enthusiasm, and highlights the challenges of regulating and understanding these tools while students and colleagues simultaneously experiment with them. The piece uses Everett Rogers’ Diffusion of Innovation theory to frame the adoption process, categorizing users into groups like early adopters and laggards, and outlines the five stages of adoption: awareness, interest, evaluation, trial, and adoption. Miller also examines the spectrum of acceptable uses for AI, dividing them into “green light” tasks (low-resistance, repetitive work), “yellow light” tasks (cautious experimentation with expertise-related activities), and “red light” tasks (high-resistance creative or original work). The article raises ethical concerns, such as the potential for misinformation, copyright infringement, and depersonalization of communication, while also noting the practical benefits of AI as a tool for efficiency and learning. She emphasizes the ongoing challenges educators face in navigating the ethical, technical, and policy implications of generative AI, likening its introduction to opening Pandora’s Box, where hope remains the last and most enduring consequence.
    • Norris, Christopher. “Poetry, Philosophy, and Smart AI.” SubStance, vol. 53, no. 1, 2024, pp. 60-76, https://doi.org/10.1353/sub.2024.a924142.
      • Norris examines the debates surrounding AI’s role in poetry and creativity, drawing on his expertise in poetry, philosophy, and literary theory. He questions whether AI can genuinely think or create, arguing that AI-generated poetry is either a product of human creativity or a simulation. Norris engages with philosophical arguments, such as John Searle’s “Chinese Room” thought experiment, which posits that AI lacks consciousness and intentionality, rendering its outputs mere mechanical processes. He also considers the “extended mind” thesis, suggesting that tools like AI are extensions of human cognition, complicating distinctions between human and machine creativity. Norris discusses William Empson’s work on poetic ambiguity, linking it to the challenge of determining whether AI can capture the intentional depth of human poetry. The article concludes that while AI can mimic poetic forms, its impact on poetry will depend on its role as a tool to enhance rather than replace human creativity, emphasizing the enduring importance of judgment and intentional creativity in poetic endeavors.
    • Laurent, Dubreuil. Humanities in the Time of AI. University of Minnesota Press, 2025, https://muse.jhu.edu/book/129366.
      • Dubreuil’s Humanities in the Time of AI is the most challenging reading (to me) on this list. He presents a paradoxical optimism about the potential of AI to revitalize the humanities. He argues that while AI can perform tasks traditionally associated with humanistic research, such as summarizing, describing, or translating, these are not the ultimate goals of the humanities. Instead, he advocates for a maximalist understanding of scholarship, emphasizing creation, transformative signification, and the exploration of the unsaid. He distinguishes between invention, which reorganizes existing knowledge, and creation (poiēsis), which emerges from contradictions and impossibilities, suggesting the latter as the true domain of the humanities. Dubreuil critiques both conservative and progressive approaches to humanistic inquiry, arguing that the humanities must engage in a dialogical endeavor that considers the past, present, and future without reducing scholarship to tradition or moral indictment of historical eras. He also explores the relationship between AI and human cognition, noting that AI is not a separate entity but an extension of human thought, trained on vast datasets of human artifacts. Dubreuil emphasizes that while AI can simulate creativity, it lacks true interpretive and creative capacities, which remain the province of the humanities. He calls for a redefinition of the humanities that embraces their role in interpreting and creating meaning, positioning them as a counter to the standardized, algorithmic thinking of AI. Dubreuil critiques the superficial dismissal of Plato’s concerns about writing in the Phaedrus, arguing that these concerns resonate with contemporary debates about AI’s impact on cognition and memory. He challenges the notion that AI is merely a neutral tool, highlighting its potential to reshape human thought and creativity. Dubreuil also examines the ethical dimensions of AI, cautioning against the reduction of ethics to utilitarian decision-making and emphasizing the importance of critical interpretation over descriptive scholarship. He advocates for a maximalist vision of the humanities, one that embraces complexity, creativity, and the unique capacities of humanistic inquiry in the face of technological advancements. Dubreuil sees AI as both a challenge and an opportunity to redefine the humanities’ role in fostering critical thinking, interpretive depth, and intellectual exploration.

  • Improving Cooling in My New AI Workstation

    four installed NVIDIA video cards: 3090 FE and 3 x A4000

    In my original write-up about building my new AI-focused workstation, I mentioned that I was concerned about the temperatures the lower three NVIDIA RTX A4000 video cards would reach when under load. After extensive testing, I found them–especially the middle and bottom cards–to go over 90C after loading a 70B model and running prompts for about 10 minutes.

    There are two ways that I’m working to keep the temperatures under control as much as possible giving the constraints of my case and my cramped apartment environment.

    terminal emulator screen

    First, I’m using these commands as root:

    # nvidia-smi -i 0 -pl 200 
# nvidia-smi -i 1 -pl 100
# nvidia-smi -i 2 -pl 100
# nvidia-smi -i 3 -pl 100

    What this command, bundled with the NVIDIA driver, does is select a video card (the first video card in the 16x PCIe slot is identified as 0, the second video card is 1, the third is 2, and the fourth is 3) and change its maximum power level in watts (200 watts for card 0, 100 watts each for cards 1-3). If the power level is lower, the heat that the card can generate is lower. I set the 3090 FE (card 0) to 200 watts, because it has better cooling with two fans and it performs well enough at that power level (raising the power level leads to steeper slope of work being done).

    PC case slot fan with molex power connector

    The second solution was to add more fans. The first fan is a PC case slot fan perpendicular to the video cards. This is a constantly on fan powered by a molex connector that has a blower motor that sucks in air from inside the case and ejects it out the back of the card. These use to be very useful back in the day before cases were designed around better cooling with temperature zones and larger intake and exhaust fans. The second fan was a Noctua grey 120mm fan exhausting out of the top of the case. This brings the fan count to two 140mm intake fans in the front of the case, two 120mm exhaust fans in the top of the case, one 120mm exhaust fan in the rear of the case in line with the CPU, and one slot fan pulling hot air off the video cards and exhausting it out of the back.

    computer case with four video cards before installing slot exhaust fan
    3/4 view of computer case with slot fan installed
    side view of computer case interior with slot fan above the four video cards

    With these cooling-oriented upgrades, I’ve found that the temperatures are slightly better during operation, but perhaps helping in a way that I had not considered before is that the fans help cool the cards down faster after an operation is completed than the cards were cooling down on their own before. Also, the A4000 temps before seemed to be high, higher, and highest going from card 1 to 2 to 3. Now, the middle card or 2 has a slightly higher temp than the bottom card or 3. Below is the output from:

    nvidia-smi -l 1

    which displays information about the detected NVIDIA video cards including card type, fan speed, temperature, power usage, power cap, and memory usage. The first Terminal screenshot below shows the cards at rest before loading a model. The second Terminal screenshot below shows the cards after a model has been loaded and it is producing output from a prompt for some minutes.

    terminal window with nvidia-smi output of four video cards with lower temperatures and low power draw
    terminal window with nvidia-smi output of four video cards with higher temperatures and higher power draw

    Y made a good point that since it’s the winter, the ambient temperature in the apartment is much cooler–we usually keep it about 66F/19C. When summer comes, it will be much hotter in the apartment even with the window air conditioner going (we are on the top floor of a building that does not seem to be insulated based on sounding and spot temperature measurements).

    The key to healthy computer components is cooling–forcing ambient air into the case and moving heated air out. Seeing how well the slot fan has worked, I’m thinking that a next step would be to drill one or two 120mm holes through the sheet metal side panel directly above where the A4000 video cards are and install high-CFM (cubic feet per minute) fans exhausting out. That would replace the currently installed slot fan. If I went that route, I can purchase PWM (pulse width modulation) fans so that I can connect them to the fan controllers on the motherboard, which will increase the speed of the fans according to the rising temperature inside the case when the computer is doing more work. This will reduce fan noise during low-load times but not affect cooling capacity.

    On a final note, I will report that I initially tried forcing cooler ambient air into the case through those two rear perpendicular slots to the video cards where the slot fan is currently installed. My thinking was that I could force cooler air over the top of the cards and the blower fans on the cards would carry out the hotter air. What I did to test this was build an enclosed channel with LEGO that sealed against the two open slots and had two 70mm PWM fans pulling air from the channel and pushing it down onto the three A4000 video cards. Unfortunately, this actually increased the temperatures on all three A4000s into the mid-90s C! The heat produced by those cards fed back into the LEGO channel and hot air trickled out of the two slots. Lesson learned.

  • All in a Day’s Work: New AI Workstation Build Completed

    black computer tower with side panel removed showing the internal components

    This past weekend, I got the final part that I needed to begin assembling my new AI-focused workstation. It took about a whole day from scrounging up the parts to putting it together to installing Debian 12 Bookworm. As you can see in the photo above, it’s running strong now. I’m installing software and testing out its capabilities especially in text generation, which without any optimizing has jumped from 1 token/sec on my old system to 5 token/sec on this system using a higher quantanized model (70B Q4_K_M to 70B Q6_K)!

    RTX 3090 video card and two nvme SSDs on a cardboard box

    The first thing that I needed to do with my old system was to remove the components that I planned to use in the new system. This included the NVIDIA RTX 3090 Founders Edition video card and two 2TB Samsung 970 EVO Plus nvme SSDs.

    micro atx case with back side panel removed revealing cables and hard drives

    I almost forgot my 8TB Western Digital hard disk drive that I had shucked from a Best Buy MyBook deal awhile back (in the lower left of the old Thermaltake case above).

    computer power supply with power cables plugged in

    Finally, I needed the Corsair RM1000X 1000 watt power supply and its many modular connections for the new system’s four video cards.

    am5 socket atx motherboard

    The new workstation is built around a Gigabyte B650 Eagle AX motherboard. I selected this motherboard, because it has a unique 16x PCIe slot arrangement–the top one has space for a three slot video card like my 3090, and its lower three slots would support the three NVIDIA RTX A4000 16GB workstation video cards that I had purchased off of eBay used. The lower slots do not run at full speed with 16 PCIe lanes, but when you are primarily doing AI inference, the speed that even 1x PCIe lanes provides is enough. If you are doing AI training, it is better to have a workstation-class motherboard (with Intel Xeon or AMD Threadripper Pro CPUs), because they support more PCIe lanes per PCIe slot than a consumer-based motherboard like this one is built to provide.

    motherboard with cpu and cpu cooler mounted on top

    The first step with the new motherboard was placing it on a soft surface and installing the CPU. I purchased an AMD Ryzen 7 7700 AM5 socket CPU. It came with AMD’s Wraith Prism RGB Cooler, which is a four heat pipe low-profile CPU cooler. I don’t care for its RGB colors, but it reduced the overall cost and provides adequate cooling for the 7700, which isn’t designed for overclocking.

    corsair ddr5 64GB RAM vengeance product box

    Next, I installed the RAM that I just received–64GB Corsair Vengeance DDR5-5200 RAM (32 GB x 2).

    2 sticks of DDR5 memory in a plastic case sitting on top of a motherboard

    This RAM runs at the stock highest speed for the 7700 CPU (though, I had to manually change the multiplier to 52x in the BIOS as it was registering as only DDR5-4800–it passed memtest86+ at the higher setting without any errors). As you can see above, it has nice heat spreaders built-in.

    2 sticks of ddr5 ram installed on a motherboard next to cpu and cpu cooler

    It’s important to note that I went with less RAM than my old system, because it’s well known that the AM5 platform and its current processors are not good at supporting higher RAM speeds for more than two RAM sticks. Since I’m focusing on doing inference with the video cards instead of the CPU (as I had done with the old system), I didn’t need as much RAM. Also, I figured that if I make the leap to a workstation-class CPU and motherboard, I can make a larger RAM investment as those systems also support 8-channel memory (more bandwidth, meaning faster inference) as opposed to the 2-channel memory (less bandwidth, slower CPU inference) on this consumer-focused motherboard.

    atx motherboard mounted inside a midtower computer case

    Then, I installed two Samsung nvme SSDs on the motherboard–one under the headspreader directly below the CPU in the photo above and one below the top PCIe slot before installing the motherboard in my new, larger Silverstone FARA R1 V2 ATX midtower case after adding the few additional standoffs that were needed for an ATX motherboard.

    RTX 3090 video card installed on motherboard inside a midtower computer case

    Out of frame, I installed the Corsair PSU in the chamber below the motherboard compartment after connecting the extra power cables that I needed for the three additional video cards. Then, I plugged in the 3090 video card and connected its two 8-pin PCIe power connectors.

    thin long single slot video card with a single blower fan front

    Then, I turned my attention to the three used RTX A4000 video cards that I got off of eBay. They are single slot PCIe cards with a 6-pin PCIe power connector built into the front of the card close to the top edge.

    thin long single slot video card with a single blower fan back with circuit board exposed
    close up of corner of single slot RTX A4000 logo on top edge of video card
    RTX 3090 and three A4000 video cards installed on ATX motherboard in a midtower PC case

    I installed the three RTX A4000s into the lower three slots and connected a 6-pin power cable to each one.

    memtest86+ RAM test screen with the large word "PASS" in the middle of the screen

    After double checking all of the connections, I powered up the system and booted from a thumbdrive loaded with memtext86+ after disabling Secure Boot in the BIOS. Before going to the trouble of installing an operating system, I wanted to make sure that the new RAM was error free.

    nvidia-smi screen indicating the presence of four video cards: RTX 3090 and 3x A4000s

    With the RAM checking out, I proceeded to boot from another USB thumb drive loaded with the Debian 12 Bookworm installer. I formatted one of the 2TB Samsung nvme SSDs as the boot drive (LVM with encryption), installed Debian 12, configured the non-free repos, installed the closed source NVIDIA drivers, and checked to make sure all of the video cards were being recognized. nvidia-smi shows above that they were!

    black midtower atx computer case on a desk

    While testing it, I have it situated on my desk back-to-front, so that I can easily disconnect the power cable and open the side panel.

    nvidia-smi screen reporting higher temperatures in each lower video card in the system than the one above it

    The immediate fix that I need to make is improving the cooling for the video cards–especially the three RTX A4000s that are tightly packed at the bottom of the case. Looking at the second column from the left under each video card named is a temperature measurement in Celsius on the nvidia-smi screen captured during a text generating session, each lower card is running hotter than the one above it: the 3090 at the top is reporting 61C, the A4000 beneath it is reporting 76C, the A4000 beneath it is reporting 82C, and bottom most A4000 is reporting 85C. Besides the fact that they are right against one another in the case, there are two other concerns. First, the PCI slot supports on the case are partially covering the exhaust vents on each card. Second, the cooler outside air might not be making it to A4000s as well as I would like even though there are two 140mm fans positioned in the front of the case bringing in cooler outside air, which is exhausted by a 120mm fan in the back above the video cards and a 120mm fan on the top of the case above the CPU cooler. One option is to drill a large hold in the side panel and mount a 120mm fan there to blow outside air directly onto the A4000 cards. Another option that I might try first is rigging a channel from the back of the case to the A4000s to blow air from a two slot port above the A4000 cards to the top edge of those cards. The latter will require less work, so I’ll try it first and see if it changes the temperatures at all.

  • Mark V. Shaney v1.0, a Probabilistic Text Generator for MS-DOS

    Mark V. Shaney v.1.0 running in DOSBox.

    Of the text generators that I’ve discussed this past year, Mark V. Shaney v. 1.0 (MARKV.EXE) is by far the simplest to use but it is also one of the most advanced due to its implementation of weighted probability tables (Markov chains–the program’s name is a pun on this) that underpin how it generates text. I was able to obtain a copy from the TextWorx Toolshed archived on the Internet Archive’s Wayback Machine.

    MARKV.EXE (44,365 bytes) was developed in 1991 by Stefan Strack, who is now a Professor of Neuroscience and Pharmacology at the University of Iowa. In the MARKV.DOC (10,166 bytes) file that accompanied the executable, Strack writes, “Mark V. Shaney featured in the “Computer Recreations” column by A.K.Dewdney in Scientific American. The original program (for a main-frame, I believe) was written by Bruce Ellis based on an idea by Don P. Mitchell. Dewdney tells the amusing story of a riot on net.singles when Mark V. Shaney’s ramblings were unleashed” (par. 2). Dewdney’s article on the MARKV.EXE program appears in the June 1989 issue of Scientific American. The article that Strack mentions is available in the Internet Archive here. A followup with reader responses, including a reader’s experiment with rewriting Dewdney’s June 1989 article with MARKV.EXE, is in the January 1990 issue here.

    The program works by the user feeding a text into MARKV.EXE, which is “read.” This generates a hashed table of probabilistic weights for the words in the original text, which can be saved. The program then uses that table and an initial numerical seed value to generate text until it encounters the last word in the input text or the user presses Escape. The larger the text (given memory availability) , the more interesting its output text, because more data allows it to generate better probability weights for word associations (i.e., what word has a higher chance to follow a given word). Full details about how the program works can be found in the highly detailed and well-organized MARKV.DOC file included with the executable.

    Using DOSBox on Debian 12 Bookworm, I experimented by having MARKV.EXE read William Gibson’s “Burning Chrome” (1982). I pressed “R” for “Read,” entered the name of the text file (bchrome.txt), and pressed enter.

    The program reported “reading” for a few minutes (running DOSBox at default settings).

    After completing its “reading,” the program reported stats on the table that it created using bchrome.txt: 9167 terms (608,675 bytes).

    I pressed “G” and the program began to generate text based on its table of probabilities generated from the bchrome.txt text file, which contained the short story, “Burning Chrome.” While the generated text flows across the screen, there are options to press “Esc” to stop or any other key to pause.

    After it completed writing the generated text to the screen, I pressed “S” to save the generated text and it prompted me to type in a file name for the saved generated text: gibson.txt.

    Pressing “S” gives the user an option to save the table for future use. I went with the default name, MARKKOV.MKV (not to be confused with a modern Matroska container file). This file can be loaded in MARKV.EXE on subsequent runs by pressing “L” and entering the name of the table. When the user presses “Q”, the program exits back to DOS and displays a message, “The random number seed was x,” where x is a random number used in the generation of text. If repeatability is important to the user, you’ll want to make a note of that number and use it with the -s modifier when running MARKV.EXE again (e.g., markv.exe -s2510).

    Mark V. Shaney’s implementation of a Markov chain that builds a table of next word probability on a small text sample is one example of the predecessors to large language models (LLMs) like LLaMA and ChatGPT. However, Mark V. Shaney’s word association probabilities is far simpler than the much more complicated neural networks of LLMs (especially considering attention) with many orders of magnitude more parameters trained on gargantuan data sets. Nevertheless, Mark V. Shaney is one aspect of the bigger picture of artificial intelligence and machine learning development that led to where we are now.

  • Rewriting Sidney Lanier’s “The Marshes of Glynn” for the Age of Global Warming with Generative AI

    A few nights ago while waiting to drift off to sleep I was thinking about Lanier’s Oak in Brunswick, Georgia and it’s namesake’s 1878 poem “The Marshes of Glynn.” The marshes that Sidney Lanier saw in the mid-to-late 1800s would be transformed by the shipping and shipbuilding in the Brunswick Port and the industrial chemical works at what used to be Hercules. Now, global warming and climate change will continue reshaping those marshes and eventually envelop them under the rising ocean waters.

    For this post, I experimented with two generative AI systems to imagine the future of the marshes of Glynn–the first visually and the second textually.

    For the visual imagining at the top of this post, I wanted to create a photo of Lanier’s Oak submerged in the ocean’s rising waters. To make the image, I used Automatic1111’s stable-diffusion-webui and Stability AI’s SDXL checkpoint model with aurora_burst’s Landscape Unleashed LoRA. To guide its appearance to include a copy of Lanier’s Oak, I used two ControlNet units set to canny and depth respectively and each using Ebyabe’s photo of the old tree on Wikimedia Commons. However, before I could use this CC-licensed photo, I used inpaint+lama to remove man-made fixtures in the scene (e.g., the light pole) and GNU Image Manipulation Program (GIMP)’s cloning tool to erase the foliage in the upper right and the cars on the road behind.

    For the textual imagining, I wanted to use a text generative AI to rewrite Sidney Lanier’s “The Marshes of Glynn” in light of the climate change that will ultimately transform the marshes. To do this, I used oobabooga’s text-generation-webui and ggerganov’s llama.cpp to load sophosympathia’s midnight-miqu-70B-v1.5 large language model.

    Midnight Miqu 70B is an uncensored model that is capable of outputting all sorts of text–good and bad. But there’s a lot of evidence at this point that censored models become lobotomized due to the built-in censorship. It’s almost like with a person–it takes the good and the bad to create something new and meaningful. LLMs and image generating AIs only work well when they are uncensored. Ultimately, it comes down to how the human operator uses generative AI–for good or ill. The same is true for most tools and technology. Disneyfication of these models only serves to shield their creators from liability and results in dumbed-down tools that are incapable of fulfilling their promise to enable human operators to do more than they could alone. Using an uncensored model like Midnight Miqu 70B gives the human operator access to an improved imagination, diction, and reasoning, which is why I chose it for this task.

    After engaging in a chat with Midnight Miqu 70B, I gave it this prompt followed by a copy of Lanier’s original poem (shown below in the left column). My prompt included some background details, such as defining eco-criticism, which would activate connections to those other words. I contextualized when the poem was written and how things might change in the future that the new poem should capture. To shape the type of poem, I included key terms such as lament, elegiac, and apocalyptic. The poem in the right column below is the one written by Midnight Miqu 70B.

    You are a renowned poet with many accolades who focuses on eco-critcism, or the literary study that focuses on ecology, environmentalism, global warming, climate change, and rising sea levels. For this task, please rewrite Sidney Lanier's poem "The Marshes of Glynn" in your own words as a lament on the effects of human-made climate change. It's important to think about the facts that Lanier wrote his poem almost 150 years ago and the marshes of Glynn that Lanier writes about in his poem is in Brunswick, Georgia, which is on the southeastern coast of the United States. Due to rising sea levels, the marshes of Glynn will likely be underwater in the next 50-100 years. With that in mind, here is the poem that I would like you to use as the basis of your poem--go through it line by line and rewrite it from your eco-critical perspective so that The Marshes of Glynn poem is in dialog with your new elegiac and apocalyptic poem.

    “The Marshes of Glynn” (1878)

    Sidney Lanier

    Glooms of the live-oaks, beautiful-braided and woven
    With intricate shades of the vines that myriad-cloven
    Clamber the forks of the multiform boughs,–
    Emerald twilights,–
    Virginal shy lights,
    Wrought of the leaves to allure to the whisper of vows,
    When lovers pace timidly down through the green colonnades
    Of the dim sweet woods, of the dear dark woods,
    Of the heavenly woods and glades,
    That run to the radiant marginal sand-beach within
    The wide sea-marshes of Glynn;–
    Beautiful glooms, soft dusks in the noon-day fire,–
    Wildwood privacies, closets of lone desire,
    Chamber from chamber parted with wavering arras of leaves,–
    Cells for the passionate pleasure of prayer to the soul that grieves,
    Pure with a sense of the passing of saints through the wood,
    Cool for the dutiful weighing of ill with good;–
    O braided dusks of the oak and woven shades of the vine,
    While the riotous noon-day sun of the June-day long did shine
    Ye held me fast in your heart and I held you fast in mine;
    But now when the noon is no more, and riot is rest,
    And the sun is a-wait at the ponderous gate of the West,
    And the slant yellow beam down the wood-aisle doth seem
    Like a lane into heaven that leads from a dream,–
    Ay, now, when my soul all day hath drunken the soul of the oak,
    And my heart is at ease from men, and the wearisome sound of the stroke
    Of the scythe of time and the trowel of trade is low,
    And belief overmasters doubt, and I know that I know,
    And my spirit is grown to a lordly great compass within,
    That the length and the breadth and the sweep of the marshes of Glynn
    Will work me no fear like the fear they have wrought me of yore
    When length was fatigue, and when breadth was but bitterness sore,
    And when terror and shrinking and dreary unnamable pain
    Drew over me out of the merciless miles of the plain,–
    Oh, now, unafraid, I am fain to face
    The vast sweet visage of space.
    To the edge of the wood I am drawn, I am drawn,
    Where the gray beach glimmering runs, as a belt of the dawn,
    For a mete and a mark
    To the forest-dark:–
    So:
    Affable live-oak, leaning low,–
    Thus–with your favor–soft, with a reverent hand,
    (Not lightly touching your person, Lord of the land!)
    Bending your beauty aside, with a step I stand
    On the firm-packed sand,
    Free
    By a world of marsh that borders a world of sea.
    Sinuous southward and sinuous northward the shimmering band
    Of the sand-beach fastens the fringe of the marsh to the folds of the land.
    Inward and outward to northward and southward the beach-lines linger and curl
    As a silver-wrought garment that clings to and follows
    the firm sweet limbs of a girl.
    Vanishing, swerving, evermore curving again into sight,
    Softly the sand-beach wavers away to a dim gray looping of light.
    And what if behind me to westward the wall of the woods stands high?
    The world lies east: how ample, the marsh and the sea and the sky!
    A league and a league of marsh-grass, waist-high, broad in the blade,
    Green, and all of a height, and unflecked with a light or a shade,
    Stretch leisurely off, in a pleasant plain,
    To the terminal blue of the main.
    Oh, what is abroad in the marsh and the terminal sea?
    Somehow my soul seems suddenly free
    From the weighing of fate and the sad discussion of sin,
    By the length and the breadth and the sweep of the marshes of Glynn.
    Ye marshes, how candid and simple and nothing-withholding and free
    Ye publish yourselves to the sky and offer yourselves to the sea!
    Tolerant plains, that suffer the sea and the rains and the sun,
    Ye spread and span like the catholic man who hath mightily won
    God out of knowledge and good out of infinite pain
    And sight out of blindness and purity out of a stain.
    As the marsh-hen secretly builds on the watery sod,
    Behold I will build me a nest on the greatness of God:
    I will fly in the greatness of God as the marsh-hen flies
    In the freedom that fills all the space ‘twixt the marsh and the skies:
    By so many roots as the marsh-grass sends in the sod
    I will heartily lay me a-hold on the greatness of God:
    Oh, like to the greatness of God is the greatness within
    The range of the marshes, the liberal marshes of Glynn.
    And the sea lends large, as the marsh: lo, out of his plenty the sea
    Pours fast: full soon the time of the flood-tide must be:
    Look how the grace of the sea doth go
    About and about through the intricate channels that flow
    Here and there,
    Everywhere,
    Till his waters have flooded the uttermost creeks and the low-lying lanes,
    And the marsh is meshed with a million veins,
    That like as with rosy and silvery essences flow
    In the rose-and-silver evening glow.
    Farewell, my lord Sun!
    The creeks overflow: a thousand rivulets run
    ‘Twixt the roots of the sod; the blades of the marsh-grass stir;
    Passeth a hurrying sound of wings that westward whirr;
    Passeth, and all is still; and the currents cease to run;
    And the sea and the marsh are one.
    How still the plains of the waters be!
    The tide is in his ecstasy.
    The tide is at his highest height:
    And it is night.
    And now from the Vast of the Lord will the waters of sleep
    Roll in on the souls of men,
    But who will reveal to our waking ken
    The forms that swim and the shapes that creep
    Under the waters of sleep?
    And I would I could know what swimmeth below when the tide comes in
    On the length and the breadth of the marvellous marshes of Glynn.

    “Lost Marshes of Glynn” (2024)

    Midnight-Miqu-70B-v1.5

    In the shadowed embrace of live-oaks, once proudly entwined,
    Where vines whispered secrets to lovers who’d intertwine,
    Now stand sentinels mourning in emerald twilight confined,
    As the marshes of Glynn weep for a world they defined.
    Virginal lights once shy, now bear witness to plight,
    Where woods once heavenly whispered are cloaked in despair’s blight;
    The arras of leaves that shielded from day’s harshest light,
    Now tremble with fear at the specter of night.
    Oak and vine, your braids now unraveled by time’s cruel hand,
    While I sought refuge in shade, you bore witness to land
    That grew weary from mankind’s relentless command,
    And sank ‘neath the weight of a burden so grand.
    When noon was young and your hearts were ablaze with delight,
    You cradled my spirit through each fleeting daylight;
    But now as the sun takes his leave from our sight,
    I am haunted by shadows that marshes once fought.
    To the edge of this wood, where the beach does expand,
    A glimmer of hope in a world so unplanned,
    Where the sea’s tender kiss meets the shore’s gentle sand,
    In this union, I find my stand.
    Marsh-grass once vibrant, now faded and frail,
    Stretches to skies that no longer prevail;
    The terminal blue of the main seems so pale,
    As it swallows the marsh in its jail.
    What secrets do you hold, oh marshes so free?
    Where once life abounded, now only a plea:
    For mercy from humans who took without care,
    Leaving your beauty to despair.
    Ye plains of the sea, how you’ve suffered in vain,
    Bearing the brunt of our greed and disdain;
    Through floods and droughts, through sun’s fiery reign,
    You remain steadfast, though in silent pain.
    As I gaze on your vastness, my heart does implore,
    What creatures lurk beneath when the tide is no more?
    In the depth of night, what forms do you bore?
    O marshes of Glynn, tell me once more!
    The sea’s generous flow now a harbinger of woe,
    Filling your veins with saltwater’s slow grow;
    Your lifeblood ebbing as the tides overflow,
    And nightfall descends like a shroud below.
    Sleep may come softly to those who reside,
    But what dreams disturb you in this silent tide?
    What visions of loss do your waters provide,
    As we slumber while you slowly subside?
    Farewell, great Sun! Your fiery reign is undone,
    The creeks overflow, and the marshes become one;
    With the sea that has claimed what was never its own,
    And nightfall’s embrace in a mournful tone.
    In this stillness of waters so vast,
    Where once life danced to your rhythmic cast,
    Now lies a tomb for the future unborn,
    As we face the truth that you can’t outlast.
    What mysteries dwell where the tide meets the shore?
    When will humans learn what they must restore?
    O marshes of Glynn, in your silent roar,
    Teach us to cherish forevermore.

    While I gave it the instruction to rewrite Lanier’s poem, it obviously wrote a much shorter poem. However, when you read these side-by-side, you can see phrases and lines that connect the latter to the former. With the LLMs that I can use at home on my computer, they don’t have a conception of word count or number of sentences or lines. They do tend to do well with lists (e.g., write a list of 10 things). It’s might be possible to number the lines in Lanier’s poem and provide instructions to have the outputted lines match the line numbers of the input. That will be an experiment for another day.