Dr. Gupta's discovery began with the observation that after interacting with Large Language Models (LLMs), her postdoctoral research assistants were listless and bored; hardly their usual energetic young selves. Not to mention useless for doing theoretical physics research. Most likely she was noticing a spurious correlation, but Gupta was intrigued and set up an informal experiment to disprove the link.
Experimental Group A directly interacted with LLM software and had the strongest effect. Control groups C (interacting with non-LLM software), D (non-LLM software in use, non-interactive) and E (no software) were unaffected. Most startling was the outcome for Experimental Group B, who were in a room with a computer executing LLM software, but with no inputs and outputs. For members of Group B, the degree of flattened affect was indistinguishable from group A.
Amused, she published her findings on arXiv under a pseudonym and moved on. That is, until it was cited by Emi Suzuiki in a paper that found that the performance of a synthetic nematode connectome was measurably degraded when it occupied the same shared hosting server as LLM software.
Gupta and the Suzuki collaborated on their next paper, establishing a causal relationship between LLM software usage and impaired neural activity, both in natural and simulated neurons. After use of LLM software was discontinued the neural activity initially appeared to return to baseline after a period of time. The problem was that Suzuki measured a higher baseline than Gupta, even for the identical synthetic connectome on identical hardware.
Gupta's lab, as it turns out, was in a building shared with the business school. The MBAs had long since incorporated use of LLMs into every aspect of their work: writing business plans, evaluating business plans, drafting e-mails, summarizing e-mails, generating dataing profiles, sifting through dating profiles for good matches. You name it, they were doing it with LLMs.
The evidence gathered by Gupta and Suzuki soon became crystal clear: neural activity and deep learning software both use, and potentially exhaust, some physical feature of spacetime not previously identified. Their next paper coined the term "Qualium", detailed construction of a 7-neuron device to measure the ambient level, and clearly showed that the Earth's baseline level of Qualium had been depleted by 2% since Suzuki's paper.
The fact is, we now face two ecological catastrophes: anthropogenic climate change ,and now Qualium exhaustion. World leaders have spent the last few decades saying much and doing little about the former. What, if anything, will they do about the impending Qualium shortage? In both cases, taking real action seems to risk the "line go up" nature of the global economic system. Well, I for one hope that we run out of Qualium before we cook in an atmosphere of CO2. By all acccounts the ego-death of Qualium deprivation is a painless one. I just regret that my animated corpse will probably continue serving the ends of capitalism until it, too, dies. I'm just glad I won't be there to experience it.
Entry first conceived on 8 June 2023, 22:51 UTC, last modified on 9 June 2023, 0:04 UTC
Website Copyright © 2004-2021 Jeff Epler