the tool shed is apparently a kiosk now
We are creatures of habit, using the tools that have worn ruts into our mind.
Figure 38 from An Ax to Grind: A Practical Ax Manual 1999 USFS T&D Pub 9923 2823P
We take the necessary form to utilize the tools we have available, and at times we form new tools out of necessity, convenience, and efficiency. Some tools are purely tools of thought, while others exist as complex machinery. Yet most exist as an artifact produced to accomplish some form of work.
The tool shed is a kiosk now
Unless you have grown numb to the current buzz (or infrasound from datacenters), the introduction of Generative Pre-trained Transformer based AI models have etched an indelible subtext to any productive task that humans engage with on a daily basis: Can AI do it?
Those who are sublimated by this subtext reach out to these tumultuous, poorly understood systems in hopes that they themselves could answer the question.
The tool shed is replaced with a kiosk at this point: You may prompt it to evoke ephemera, generated by a slew of discrete operations across special computers tasked to assign how probable certain tokens are in a sequence of tokens... or something like that.
Companies that are driving resources to train these GPT Large-Language Models for inference are creating artifacts that are themselves shaped by a constant current of change; driven by those who wish to leave their impressions on this technology. The "LLM" seemingly exists as the front door to ecosystems of related GenAI for all kinds of media (images, video, audio), and it's hard to ignore the effect this has: people start imagining what it could be.
It's all of us in there...
This rapid acceleration of adoption is almost to spite the majority of us who do not get to choose if their intellectual legacy and personal information remains as a core dependency of this ravaging paradigm, no matter how ineffectual it may seem for an individual.
Flagship companies seem to defer to shady data brokerages with limited or no oversight, which simply solve the need for data labeling in whichever ways produce the desired effect at the cost of a complete abandonment of ethics1.
and it's not for me.
I will never be able to, in good faith, use a large language model.
I may never have the resources to be able to shape them to myself until the capacity of my own tools allow for building sufficient models from first principles, using ethical and fair sources. I could say this about many technologies, but this one is different. I don't want to visit a kiosk to do work. I want to do work with a tacit consensus that respects the real people who use the output of my work, and those who eventually might have to maintain the artifacts produced by my work.
There are valid uses for the fundamental principles that are exercised by these systems: And I'd love to get behind them and cater my tool shed to enable new patterns of work. To be as effective as possible, in both the task at hand and at developing deeper knowledge and relationships with the people who helped make it possible.
It is a breach of trust to rely on the state of the art for agentic workflows. The friction of using these tools is not a two-way street: The momentum of AI companies overshadows my role I get to play when engaging with their services. Their effectiveness is the product of an approach I cannot endorse or support.
Popular services from questionable tech companies in the past could be replaced with existing software, developed by generous people the likes of which already support the backbone of modern information systems. It's not quite the same equation with these generative AI enterprises, in a similar vein to social media giants which continue to trundle on despite movements to migrate to independent social media. The institutional momentum and commercial success has established a steep gradient to overcome: The gap between open research, trade secrets, and raw computing horsepower seems to only grow wider, with downstream effects that turn commodity hardware into a scarce resource2.
"It's inevitable, just adopt it or be left behind"
Fine, I experience the friction, to be shaped by these new tools. I may finally go through the delicate process of configuring the shape of my work to enable prompting for The Right Thing™️ (by all measurable reports this is getting easier to manage) and I seem to have a productive workflow within my means of managing it.
Hypothetically, it may not be economically feasible3 for me to utilize unfathomable intellectual property running on some hyperscaler somewhere that expends a significant envelope of limited resources to process requests that might give the right outputs: which might have otherwise taken an extra afternoon or two to churn through "by human" (since I suppose the hand is now so far removed at this point, that we might as well abandon our bodies, too).
We are all consumed by demands for higher productivity, but at what point does the "AI bubble" implode, and how deep is the groove in my mind going to be after going through the friction to adopt the latest and greatest? Some report that going back is not an option, and some studies have brain scans to prove that.
"It depends."
It's never about the way you use the tool, it's about how it's adopted en masse. It's hard to say with any certainty how these tools will be used by the great minority of the technologically literate in 5 to 10 years time.
Despite that, the great it depends debate will rage on, comparisons will be drawn, and broad oversimplifications will be made (on all sides, too), until we are abreast to the consequences of acquiescing to a paradigm that shows contempt for individual sovereignty.
All we have are facts and made up minds
I understand that there is an urge to be at the cusp, to be one of the first to collect the precipitation of the state of the art. I have personally decided that this state of the art is particularly hostile, and I feel resistance to adopting it, and I will seek facts that support this feeling until the feeling goes away. I am not adverse to using the technology where it shines, but I may still prefer the familiar dull sheen of old machines.
This rambling may have some facts sprinkled in, but it is primarily a rendering of my state of mind. It's only something I can generate, and it's something that you took the time to read.
And that's more than I could ever ask for. Thank you.