Schrödinger's Chat(GPT)
Lies, Damned Lies, and Large Language Models
New Threads
Welcome to the first installment of Quantum Heap— a place for hyper-fixations and random fluctuations!
This newsletter is part of a larger resolution to restore some semblance of focus into my life. It is so easy for me to go through periods of what I think of as “constant consumption.” At times I feel as though I am on autopilot. My attention endlessly flitting between apps on my phone or the tabs on my screen— always trying to catch some thrill of being “up-to-date.” If you’ve spoken to me lately, you’ve likely heard me groaning about these feelings. I have been trying to do something about it, however, and one new interest of mine in particular has helped restore my concentration and in no small part my overall feeling of self worth. I wouldn’t be able to write this if not for crochet.
At the start of the year, my friend, Connor, jokingly said he could see me getting really into crochet. I didn’t think anything of it until a few weeks when I was talking to my therapist about how much I missed having tactile hobbies and she mentioned that I might like crocheting. It was enough of a coincidence that I figured it was worth giving it a shot. And, surprising no one (as Anna would say), I have got really into crochet.
Not only has crochet been a way to spend more time off my phone, but it attracts complimentary hobbies. While the heavy repetition makes it easy to sit quietly and let my mind wander freely, I often listen to an audiobook or music while working. Since I’ve started regularly crocheting, I have found it easier to devote more time to things outside of crochet that have been hard for me lately like spending an evening reading a book— and most especially writing.
I’ve wanted to write more regularly (and share it) for a long time, but plenty of fear and excuses have stopped me from following through until now. I’m still scared of course, but I’m ready to give it a shot.
While I am sure I’ll write much more about crochet as I take on increasingly complicated projects, I intend to write these posts about any and all of my interests without enforcing too much structure or division. That is why I chose the name Quantum Heap. The most sincere vision I have for this project is to catalog a disorderly collection of uncertainties, ephemera and passions.
I hope you’ll find something for yourself here.
Schrödinger’s Chat(GPT)
Skepticism is the name of the game, now more than ever.
The latest crop of AI chatbots like ChatGPT, image tools like Midjourney, and countless others have thrust conversation around these incontestably remarkable AI models into the mainstream. When Jane Pauley is talking about it, you can be sure it’s a big deal. A lot has been written about AI systems’ potential for malicious use, and recently an open letter from a number of tech leaders was published calling for a six-month pause in development of these tools. One thing is already certain: The models that currently exist have already exponentially increased the amount of bullshit available at our fingertips.
In the short amount of time ChatGPT has been around, I have gone from the initial excitement one has with a new toy, to now feeling much more wary of its usefulness for a number of purported uses—specifically using them as one would a search engine, as Bing and Google are barreling forward with fusing their large language models with their search engines.
“Search for anything. Ask follow-up questions. Make refinements in chat. Hone in your search and get the answers you're looking for.” [emphasis added]
The above quote from a Microsoft Bing marketing site seems like harmless PR, but I think it unintentionally points out the dangers of creating a vastly more subjective (if not outright fabricated) version of the heart of the internet—search.
Whether you feel as though targeted advertising is invasive, or you’re worried about various political actors in the US who are pushing for increasing censorship left and right and looking to make sweeping legal changes around internet usage, these concerns are only going to be exacerbated by a search engine that knows you more and more intimately, going far beyond current privacy concerns and with the growing threat of greater government entanglement.
I don’t mean to sound dramatic, but I invite you to go to the link and read it for yourself. It’s what is missing that bothers me. There is no reference to accuracy. To truth. The closest term they use is comprehensive.
I can see the rebuttal that they have warnings and disclaimers of various kinds if you dig deeper into their FAQs, but the race is on and the tech is being pushed forward with the typical “move fast and break things” attitude. I agree that right now, individuals are going to need to be more responsible and skeptical than ever about verifying their search results when they are using these experimental systems. Take for example Brandolini’s Law, the colloquial internet term for how much of an uphill battle it can be to refute false information online. We’re practically telling users to use an unprepared while a hurricane descends. Particularly when we’re already seeing people create Do Anything Now (DAN) commands that continue to do subvert guard rails the developers put in place.
My main concern for the average user is not actually around the subverted versions of, however. My concern is just how convincing these chatbots sounds by design. I fear their ability to be close enough to correct to stop you from acting to actually validate it. As Ted Chiang notes in The New Yorker, that blurriness of close enough is inherent to the models. These are not search engines— they are hallucinating savants. Let you’re guard down and they’ll dazzle you with their words. But don’t be sucked in. These models are as slick as the slickest con man. And what happens when real con men are acting in partnership? After all, these programs are trained to be confident above all else.
Finally, just for fun, some of that billion dollar bravado in action*:
Only critique is that this wasn’t a text directly to me, so I couldn’t tapback and reply in thread. Really enjoyed this, looking forward to future quantum defrazzlements.
Great read!!
A friend just asked me for a random screenplay summary. A tumultuous love story between Mike Ditka and one of his players (there was a picture of him nearby). He put it into ChatGPT, it generated a scene in seconds, we asked it to make the scene more romantic, and it spit it out. I’m glad I write obscure indie shit cause less niche writers are screwed. It sounded just as good as every big box office movie out there