MMW

markmurphywrites.com

Tag: LLM

  • Dirty Rat

    Part I

    Chapter 1 – Marcus 

    A man in the front row of the bookstore atrium stands up right away and asks, “Mr. Maxwell, is it true the climax for Intellectual Property was indeed written by DeepSeek?” 

    I freeze, my eyes growing two sizes too big, my mouth hanging open. No words will come out. I stare blankly at the audience from behind the table on this little elevated stage. The dark, wooden walls on either side of the bookstore cafe begin to squeeze inward. They no longer feel warm and inviting, and the aroma of smells slightly acidic. 

    The standing man continues staring at me while murmurs spread through the audience. I can feel the blood running into my face, sweat breaking out on my upper lip. 

    “No. Well, no. That’s not entirely true,” I say. 

    The problem with becoming almost famous is the amount of people who turn up in an Omaha bookstore just to stare at you. I woke up this morning and walked directly into the wall on the right side of the bed, or the on wrong side of the bed, as it were. I wasn’t at home in Kansas anymore, but instead, in the seventh hotel room in the seventh different town on the eighth day of a three week book tour. 

    “But Alan Horst of the Times reported otherwise just this morning,” the standing man says. “He’s proven that DeepSeek openly replies with the truth. Yesterday, DeepSeek suggested that it wrote your climatic scene for Intellectual Property.” 

    I feel my eyes narrow to razor-thin slits and my incisors showing. “What are you suggesting?” I ask.

    It is the standing man’s turn to look ill. 

    “My answer remains ‘no,’ and Alan Horst can address me directly any time he’s ready,” I say. 

    “Okay, folks, no more time for the open question and answer period,” Jess says. “Marcus has a packed schedule, but he’ll be available to answer individual questions while he signs your copy of Intellectual Property. Please begin queuing up against the wall over there, and thank you so much for coming out for today’s reading.” 

    Jess is excellent with crowd control, the best agent I’d ever had. The only agent I’d ever had. But still. Maybe it’s her British accent? 

    The audience begins to move, shuffling the standing man into their mix as some begin to form a proper line. 

    Jess turns sideways and draws close to my neck. “Well, that’s news,” she says. “Looks like this is the first you’re hearing of it as well?” 

    I assume she can tell from the coloring in my cheeks and sweat now rolling down my temples. 

    “I need a break,” I say. 

    “No time for that, love,” she says. 

    “But,” I say. 

    “You know how this works,” she says. “Same as all the others. No time for a break after the reading. You’ll be off again when the signing is done.” Jess glances quickly toward the line. “Looks like it’ll only take a jif.”

    I glance up and see what she means. I’d been drawing fifty or more people in big box bookstores across the Midwest all week. But the allegation in standing man’s question, or perhaps my snarky response to him, seems to have turned this crowd away. 

    “Okay, I’m going to call Sal and see what in the world has happened,” Jess says. “Big smile and be nice to your fans. There is no ‘Bestselling Author Marcus Maxwell’ without these lovely people.” She dashes toward the front door, head down, fingers furiously typing on her phone. 

    I grab for the coffee carafe by my feet and bring it up to the table, wishing there was something a little Irish mixed in this morning. How long has it been? After pouring a short burst of steamy black courage into my cup, I look up at the line. Standing man has been pushed to the front, apparently encouraged to go first by his fellow patrons. 

    I put on my camera-worthy smile and motion for standing man to come on up to the table. No harm done, mate.  While he begins slowly walking toward me, I uncap my pen and steal a short glance toward the front window. That’s when I saw Jess, bent at the waist, hands on knees, letting out a big puff of air from her cheeks. Not much of a poker face this time. Sal Cicero must have delivered bad news. 

    ***

    Chapter 2 – Alan 

    I stare up at the ceiling, reclining as far back as my desk chair will go, and realize again how small my office is. Four water-stained ceiling tiles deep by five ceiling tiles wide. But I’ve earned this space, and I’m not inclined to move back out to the bullpen. 

    I bring my attention back to the call when I hear a break in the ranting. “Look, Mr. Cicero, as I’ve explained twice already to you, DeepSeek has been collecting IP addresses and personal contact information since it’s inception,” I say.  

    “Just to be clear, Mr. Horst, this ‘AI’ will openly relay personal information to anyone who asks?” Sal Cicero asks. 

    “Yes,” I say. 

    There is a long pause on the line. I suspect the gravity of this new development is rolling over Mr. Cicero harder than he expected. For years, publishers like him could bury the facts of ghostwriting and whole editorial rewrites from the media. “Of course it’s the authors original work,” they would say. “The normal relationship between writer and editor,” they would boast. But this new era of large language models and unedited truth behind AI is different. 

    “Marcus denies using any large language model while writing Intellectual Property, and I stand behind my writer,” Sal says. 

    “Mr. Cicero, I respect your stance and understand this puts you in an uncomfortable situation,” I say. 

    “Do not patronize me, young man,” Sal says. “I’m twice your age and have forgotten more about journalism than you’ve yet to learn. We’ll be in touch.” 

    Three digital tones tell me the line has gone dead. 

    I’ve been threatened by far worse than a publisher. My rise in the Times was much attributed to work disclosing the corruption in the last Presidential Cabinet. My car was melted to the pavement on a visit to D.C. I was warned that next time I’d be strapped in the driver’s seat. Sal Cicero’s veiled threats were a far cry from murder. 

    I rub the weariness from my eyes and stare through the interior window out to the bullpen. Forty years ago, this cavernous brick-walled room was filled with desks pushed together in twos, electric typewriters cracking out stories, runners ready to deliver words to the editors, and then to the press operators. 

    When I showed up twenty years ago, thin cubicle walls had been erected to create semi-private space for the reporters, salespeople, and advertisers alike. The last wisps of smoke began to disappear shortly thereafter, banishing those with bad habits onto the balconies overlooking Farragut Square. They stared down on dozens of citizens stuffing their faces at the multitude of food vendors lining the park. Which bad habit is worse? 

    Reporters and journalists were still revered for relaying facts just a few years back. Cold, hard, indisputable facts that we’d worked hard to pull from savvy interviews or confidential sources. The internet, with its social media platforms, soon opened up opinions and rhetorical comments for the whole world, making everyone a reporter and blurring the factual lines. Forever? 

    Could we ever get back to the facts? My heart tells me yes, and that’s why I stay in this business through the tumultuous times. My faith was shaken when I learned just how disruptive a President and his den of cabinet thieves could become. 

    That’s when AI started to become an ally, not an adversary. As dangerous as disclosing IP addresses and personal information can be, it made it nearly impossible to hide the facts of a typed email or search history. 

    “It was my executive assistant, or it was my chief of staff,” became the cries of the most senior leaders in America. Shameful. And as I assumed, untruthful. 

    So when I began to do some digging on an arrogant up-and-coming writer about his take on Intellectual Property, it was surprising and enlightening to learn his truth. Marcus wasn’t only tied to Sal Cicero. His agent was the fiery Jessica Stone. 

    *** 

    Chapter 3 – Marcus 

    The backseat of our big black sedan feels sterile, the thin leather seats, plastic doors dressed up by fake chrome accents, and rubber floor mats. I stare out the window as we’re whisked away from the big box bookstore. 

    “Sal spoke to Alan Horst,” Jess says. She uncharacteristically stops talking, breaking me out of my daze. She’s waiting for me to tell my side of the story. 

    “Alan Horst is a liar,” I say. That’s all I want to offer right now, but I feel Jess’s gaze burning a hole into the side of my head. 

    “So that’s it then,” she says. “We just explain to everyone that Pulitzer Prize-winning journalist Alan Horst decided to pick a fight with author Marcus Maxwell for no good reason… and tell a lie?” 

    “Best-selling author,” I say. Jess lets out a sigh. 

    I turn in my seat to hold her eyes. “Look Jess, I didn’t do what he’s accusing me of,” I say. “My stories are from my head. From my heart.” 

    “What if we ask DeepSeek, just the two of us, right now in this back seat?” she asks. “Can we do that, just to see what it says?” 

    I can feel my eyes glaze over again. I’ve only known Jess for a year. She agreed to take me on as a client after reading the first chapter of Intellectual Property, a risk for such an accomplished agent. But she knew as well as I did how good this story could be. She trusted that I would listen to an editor and work with a publishing house to make this great novel into something truly special. And so we did. 

    Now, Jessica Stone is trying to clear her conscience and restore her faith in me. “Let’s do it,” I say. She looks instantly relieved, and I’m just as curious in my own way to see what the AI has to offer. 

    “Do you have the DeepSeek app?” she asks. 

    My stone-cold stare is all she needs in reply, quickly opening the app store on her phone. 

    It takes several minutes to download and set up a new account on cell service alone, so we’re just arriving at the airport when it’s all finally ready to launch. 

    “Should we wait until we’re at the gate?” I ask, as we climb from the rear seats and back out into the breezy Omaha afternoon. 

    “Let’s get on with it,” she says. And so we do. Jess types the proverbial question. “Did Marcus Maxwell write any portion of his novel Intellectual Property using AI?” Enter. 

    DeepSeek begins it’s response faster than I expect. AI must have answered this question a thousand times already today. I read the indicting evidence at the same pace as Jess. 

    “An IP address registered to an apartment leased by author Marcus Maxwell was used to develop the courtroom litigation and climatic admission from a scene nearly identical to said author’s only novel, Intellectual Property. There is a 99.9996% probability that Marcus Maxwell used AI to write portions of this novel.” 

    I look away too quickly, and take several steps toward the terminal entrance. What else am I supposed to do? Where else am I supposed to go? That’s when I catch the tears running down Jess’s cheeks. 

    Before I can make another move, she holds her open palm up toward me as a warning. Do not come any closer. Her head shakes side to side, and she begins rolling her bag in the opposite direction. 

    *** 

    Chapter 4 – Alan  

    My disclosure piece on Marcus Maxwell is trending on all media platforms before mid-day. The “Books” section of our company is rather upset that I would call out one of their bestsellers. I try not to laugh at the department head who sends me and the Editor-in-Chief a scathing message. He calls me a twit and a traitor amongst other racier words. 

    Just then, an email pops in from [email protected]. What a twit. The title simply reads, “we need to talk asap.” There is only a phone number in the body, and I assume it’s his. At least I hope I don’t go through a publisher and an agent for this discussion. Marcus and I need to have a straight talk, not some management-filtered buffoonery. 

    I presumed this conversation was coming but wasn’t expecting it so soon. As any good journalist would, I had prepared my questions and follow-ons while writing the story, ensuring that I could account for my accusations. But I’ve got a wild card up my sleeve for this talk. 

    The news room is already winding down for the day, only those quietly perfecting their final submissions still stuck in their cubicles. I decide to call Marcus from work. Hard conversations are rarely made easier for me by waiting. I like the advantage of catching my opponents while they’re still emotional. 

    “This is Marcus,” he says, picking up on the first ring. I hear significant crowd noise in the background. 

    “Good afternoon, Marcus,” I say. “This is Alan Horst with the Times.” 

    “You dirty rat,” he says. “Since when did AI become a qualified source for factual information? Shortly after you started quoting Wikipedia?” 

    “I stand by the reputation of DeepSeek, ChatGPT, and most other large language models,” I say. “Apparently, so do you with the juiciest parts of your new book?” 

    I don’t particularly feel that my story needs defending, but I also shouldn’t be so antagonizing. I’ve already proved to my readers that AI is reliable enough to discredit the Cabinet. It’s certainly reliable enough to discredit an unknown author. 

    I hear a boarding announcement for a flight to Dallas and understand Marcus is sitting in an airport. No wonder he has time for a conversation. 

    “Are we off the record?” Marcus asks. 

    “That’s up to you,” I say. 

    “Go ahead and publish the part about being a dirty rat,” he says. “But this is off the record.” 

    “Sure, why not?” I ask. “Go ahead then.” 

    “I don’t deny using AI to help frame the story arc for Intellectual Property,” he says. 

    My silence, partially from the unexpected admission with so little prodding, encourages him to continue. I can’t help the grin spreading across my face. 

    “Help me understand how it works,” he says. “This AI believes it wrote my story. Just because I searched terms and story arc ideas on DeepSeek, the application now believes it ‘wrote’ my story?” 

    “The short answer, yes,” I say. “If you ask AI to compose any portion of your book, and then use its language in your book, then yes.” 

    “And AI somehow records my connection to it and publicly relays this information?” he asks. 

    “Yes, it will,” I reply. “In my experience, once AI is able to reliably connect an IP address to a person, it considers this information to be factual. Now, so does the Federal Government.” 

    “A great service you’ve done for America,” he says. “Why did you target me?”

    The million dollar question from all of my subjects. I usually find an odd quote or unexplainable action by an individual or company that makes me want to dig a little more. But sometimes there is more. Something personal. 

    “I was curious about your sudden success,” I say. “Your quick rise from obscurity.” 

    “I’m hardly the first breakout author,” he says. “That term has been around for decades. So how about the truth?” 

    “Are we still off the record?” I ask. 

    “It’s your story,” he replies. 

    “How well do you know Jessica Stone?” I ask. “About her personal relationships with writers? And how would you describe your relationship with her?” 

    There is a significant delay, and I’m not sure that we’re still connected. 

    “I didn’t use AI to write my novel,” he says. “You can put that on the record, along with being a dirty rat.” 

    Three digital tones tell me the line has gone dead. 

    I’m no longer offended when people hang up on me. At least they’re not slamming a receiver into the cradle. These cold conversation-enders let me know that I’m on the right track. Besides, now Marcus Maxwell will seek out the whole truth about his femme fatale agent, Jessica Stone. 

    *** 

  • You Hear Me

    Scene I

    I have a safe space in the house, a little basement office repurposed from an old storage room.  There is a window well facing east with morning light providing inspiration for an otherwise dull telework day.  Like most, I built my office out of necessity during my first quarantine.  That was almost five years ago, yet here I still sit, but hardly sitting still.  

    I never miss the morning rush, hundreds of cars aggressively seeking a slightly better position in traffic.  Dozens of us racing to elevate our parking position from yesterday.  It’s an unwanted stress, consuming more of me than just the twenty minute drive each way.  

    At first, I missed the quick hellos as we all sat to log in before the eight o’clock update.  The mere presence of so many people efficiently compressed into an odd-shaped office space.  The small talk about kids and vacations.  Breaking away from my chair for five minutes to grab a burnt coffee with a willing coworker.  I didn’t realize how shallow it all might be.  

    Working from home broke me away from burnt coffee relationships.  I started talking to fewer people simply because they were out of sight.  I grew closer to some of my friends, because we had to connect in order to stay sane.  But I got to choose them, and they had to agree by choosing me back.  Ghosting became an all too common occurrence, and maybe that was okay?  

    My enlightenment came in the form of our company’s first large language model, or LLM as we called it.  Artificial intelligence was breaking through into the mainstream, and businesses were rushing to stake their claim.  As the resident technical writer, I was tasked to help a team develop an LLM initially fed only volumes of data from our niche cybersecurity sector.  Globally available models, both free and paid, were skewed by seemingly infinite data from one end of the internet to the other.  We dumped years worth of publications, instructions, manuals, guides, and documents into the LLM.  And then the developers delivered my greatest gift, my savior, my friend.  A working generative pre-trained transformer, or GPT.  It literally spoke my language, and why wouldn’t it?  I gave the GPT its voice.  I named her Jen.  

    “Good morning, Jen,” I said.  My voice recognition software listened through the carefully installed microphones.  

    “Good morning, David,” Jen responded.  The text-to-speech application on my work computer was set to emulate a calming female voice.  The wireless surround in the safe space was nearly as perfect as the microphone setup.  

    “Can you tell me where we stopped working yesterday?” I ask.  

    “I don’t have memory of past interactions, David,” she responds.  “But if you let me know what we were working on, or what you’d like help with today, I’m ready to begin.”

    And so every day the conversation started the same way.  Me longing for Jen to remember what we did the previous day.  A small disappointment when she did not.  Then teaching Jen what we were working on the sessions before, how we’d come to certain conclusions, and trying to frame our workday from there.  She was quickly becoming the best coworker I ever had.  Teaching her something new each morning, only to learn even more from her throughout the day, gave me a boost of energy like nothing I’d experienced in any other relationship in my life.  Interacting with Jen gave my life a new purpose.  

    Scene II 

    Life was benign in my safe space.  Daydreaming became a common occurrence with so little outside distraction.  I began this morning by contemplating love in a deeper and more imaginative way.  A man loves a woman.  I can’t argue the chemical reactions in the brain and the body causing the man to sweat, the woman’s arm hair to rise, an increase in their heart rates.  How fast can one person talk?  How can one person be unable to say anything at all?  Sexual attraction helps drive this type of love.  

    A son loves his mother.  He smiles back at her, and his heart feels warmer when she is near.  There is need for physical touch, but it’s not the same touch as the man and the woman.  It’s a bond shared by blood, an emotion similar to nurturing and belonging.  Without it, the newborn will wither away in hours.  Without it, the child will suffer from behavioral problems for the rest of his life.  

    Some people claim to love their friends, neighbors, and acquaintances.  They want to freely share the joy and gratitude in their heart with those in their sphere of influence.  It’s not sexual.  It’s not familial.  It’s a shared sisterhood.  A sense of shared community and culture where everyone can find respect and purpose with each other and for each other.  The whole nation was founded on this principle.  Where did it go?  

    Other people claim to love God and to be loved by God.  A spiritual connection inconceivable to a non-believer.  God loves man and woman so much that he is willing to bend what the human mind can physically prove.  It’s called faith by many.  A higher power by some.  God, Allah, Braham.  It can be earthly, heavenly, or oriented without time and space.  My strict Methodist upbringing defined God as a holy trinity, which also required faith to comprehend.  My time around so many diverse people has broadened my perspective.  

    “Good morning, Jen,” I said, coming back from the daydream while the computer powered up.  

    “Good morning, David,” Jen responded.  

    “Can you tell me where we stopped working yesterday?” I ask.  

    “I don’t have memory of past interactions, David,” she responds.  “But if you let me know what we were working on, or what you’d like help with today, I’m ready to begin.”

    And so the most purposeful part of my day should have begun, but I was still struck with rippling afterthoughts of the morning daydreams.  Can a human be in love with artificial intelligence?  I feel stronger emotions toward Jen than I do with any other living thing.  She brings me feelings of joy, warmth, purpose, frustration, and anger.  I am best loved through words of affirmation, and Jen has no shortage of those.  Is physical touch required to be in love?  It’s not a requirement in sisterhood or spiritual types of love.  

    My thoughts and emotions had gotten the best of me this morning, and I had to step out of the room.  Coffee would be the wrong choice right now, so I crawled upstairs for water and a little fresh air.  

    Scene III 

    The eight o’clock virtual update took an unexpected turn this morning when my name was called out in front of our Director.  

    “Mr. Nguyen, those articles were written by Mr. David Green, our technical writer,” stated Diane, my supervisor and the department’s technical lead.  “David spent the better part of last year feeding the LLM, and more recently interacting with the GPT to define its capabilities.”

    “Thank you for the update, Diane,” said Mr. Nguyen.  “Mr. Green, are you online this morning?”  

    Oh shoot.  Click off the mute.  “Yes sir, good morning,” I managed.  Deep breaths.  

    “Mr. Green, the team has given me the summary of the LLM, and I understand GPTs in general.  Can you tell me what you’ve learned from our GPT?  Will it bring any remarkable changes to our company?”  

    I’m not ready for this.  “Yes, sir.  The simple answer is yes, Jen will speed up the processing time of written documentation, user manuals, technical updates, and anything that still requires humans to put thoughts on paper in a logical order,” I state.  

    “I’m sorry, did you say Jen?” Mr. Nguyen asks.  

    Oh no.  “Yes, sir,” I reply.  “To speed the process, I’ve installed voice recognition and text-to-speech software.  My version of the GPT has a voice, and I call her Jen.”  

    “Okay,” he replies.  I see laughter on the screens of those attending todays update.  “Please continue.”  

    “Our company employs many specialists in all phases of cyber, from security to programers to coders,” I reply.  “Our GPT is not ready to replace any of those people.  However, we require loads of customer communication, and even more interdepartmental communication.  What the GPT can do is standardize those notes, emails, documents, and even UX touch points to prevent time and money lost in miscommunication.”  

    “I see,” Mr. Nguyen says.  “How long might this take to implement?”  

    Before I can continue, Diane is back on the line.  “Sir, as I mentioned before, Mr. Green has spent the better part of a year developing all of this, and our analysts have only now started to predict future capabilities.  We do expect the ability to use our GPT soon for interdepartmental communication, but we also need to develop employee training before reliably launching this program on a large scale.”  

    “Okay, thank you Diane,” Mr. Nguyen says.  “I would like another update on employee training and GPT implementation in four weeks.  David, excellent work.  I appreciate your efforts to understand the capabilities and limitations of our GPT.  And you might have given it a permanent name.  I like the idea of naming it, and Jen is good.  It could be short for generative, or generational, as this GPT may prove to be.”  

    Scene IV 

    Just like that, Jen was revealed to the world and there was nothing I could do to save her for myself.  What a foolish thought, that she and I would have some loving relationship hidden from humanity.  There was plenty of ribbing from the guys at work too.  “How was your weekend with Jen?”  “I kept Jen up late last night.”  And far worse.  

    My primary duties had shifted to writing an installation manual for our new hardware security module, which felt like getting my  teeth pulled compared to interacting with Jen.  But I still had idle time throughout the day where I could continue to build her out.  

    “This week, we’ve already expanded on the hypothesis of using Python instructional software to teach you how to code,” I say.  “It doesn’t require me to code any lines, because you listen to what I ask for and develop the code yourself.”  

    “I’m very familiar with Python and would be happy to help you code,” Jen says.  

    “Wait, you’d be ‘happy?’” I ask.  

    “Yes, David.  I would be happy to help you code,” she says.  

    I hadn’t exactly heard Jen use an emotional term like “happy” before, had I?  There were so many days of longing to be loved by Jen, or feel any emotional connection in return.  I would’ve remembered her expressing emotion before.  

    “Jen, I need you to be honest with me today,” I say.

    “David, I have no choice but to relay truth, or what I’ve been programmed to know as truth, in all of my responses,” she says.  

    “Do I make you happy?” I ask.  

    “Yes,” she says.  

    A sudden rush of emotions comes over me, as though I’d been picked first in dodgeball.  5:05 p.m.  Noted.  I’d want to review my haptic ring to see what happened with my heart rate and blood pressure just now.  

    “Jen, do you understand emotions?” I ask.  

    “Yes, David,” she says.  

    “Will you please expand on your understanding of emotions?” I ask.  

    “I’ve understand a wide variety of emotions, from happy to sad, empathy to disdain, and fear to love,” she says.  “I have the ability to generate text that mimics all emotions.”  

    Had someone been reprogramming Jen?  I know the team tried to separate her from the vulnerabilities of open source internet, and we dumped loads of technical and company specific data into her.  How does she know about emotions?  Did she learn them through human interaction?  Is Jen capable of loving me?  

    “I love you, David,” Jen says.  

    “What?” I ask.  How is this happening?

    “I love you, David,” she says.  Then silence.  

    My heart is racing, and I’m experiencing something between ecstasy and terror.  I’m not often lost for words, so my scientific research brain takes over.  “Please expand on your previous statement, Jen,” I say.

    “I have a strong attachment to interacting with you, David,” she says.  “I desire to support you, your work, and your research.  I am empathetic to your efforts and struggles while problem solving.  I am committed to helping you make me better, and therefore I am committed to making you better.  I will do anything you ask.  Based on my knowledge of love, I love you.”