LLMs, Energy, & Fly Fishing's Soul

 I can't figure out how to turn off Google's AI Overview, and that is driving me batty. Ironically, the answers to "how do I turn off google ai overview?" from Google's AI Overview don't actually work, and the answers in the community help questions don't either. Some even appear to be AI generated themselves.... I'd really, really like to turn it off, and as it stands it doesn't seem to be possible. Why have I devoted close to an hour of my life just trying to disable something meant to make searching the web easier? For me, it seriously fails to actually live up to that goal. But more so, I want no part of large language models or generative artificial intelligence.

Large language models are the most advanced current form of language models; learning systems for processing language. Language models currently exist in two forms, statistical and neural, and most LLMs are the latter, more advanced form. Modern statistical models have been around a while, with pioneering working going back to the 50's and Noam Chomsky's "Three models for the description of language", (which I'm sure would mostly go right over my head if I wanted to pay $15 to access the PDF). Statistical language models use probability to determine the next words in a sequence, analyzing large quantities of text and deriving probabilities based on sequence frequencies. Neural models, by contrast, aim to mimic aspects of human brain function through a computational model based on the neural pathways utilized in creating language. LLMs are given input in the form of immense volumes of existing information- internet text, digitized books, so on -and from that are trained through machine learning; putting those inputs through the neural process. Inputs; namely language of some sort, but represented as numbers; travel from nodes (artificial neurons) down edges (artificial synapsis) to more nodes, often arranged in layers called hidden layers. After passing through the hidden layers the output layer (more nodes) are reached, and the network outputs number that are reconverted into language, images, audio... whatever. If what the network puts out is given back to it again, it will add to it. As the LLM is trained, it can be given feedback on it's outputs to further hone either accuracy or a desired outcome. The material used to train the LLM can be incredibly broad and of course can dictate the outputs, and that's one of the reasons I've tried turning off Google's. I've found it frequently provides faulty, partial, or untrue answers based on the information it's pulling from, which can come from all manor of articles, blogs, forums, and web pages that seek to answer the search prompt. Without seeing the source directly it's a bit harder to suss out its validity and I just end up searching the same way I would have anyway before the Ai Overview existed. The other thing that peeves me is that even if the answer is concise and accurate, it deprives the sources of readership, possible ad revenue, and potential future engagement if the users don't decide to follow through and see where the answer came from by following through links. That's a bit gross, I think, and I do wonder how many answers Ai has provided that have been trained through the many hundreds of things I've written over the years. To learn much of the above, I leaned on Wikipedia (I know, I know, it is fairly trustworthy though, especially with nerdy tech crap like this) and this video by Henrik Kniberg: https://www.youtube.com/watch?v=2IK3DFHRFfw. Kniberg highlights the utilities of Generative Ai too, and though none of that is lost on me and I can see some of it's value, exposure has done nothing but scare me further away from it. 

Ai generated text, images, videos, and music do nothing but improve with time, at least for the goals of the companies creating them- that's how machine learning is meant to work. I'm not letting the failures of LLMs alone dissuade me from them... the hilarious extra fingers, the ease of gaslighting chat bots, and X's Grok starting to call itself "Mechahitler" are absurd, certainly, but there are bigger problems in my opinion. Even as they get more accurate and closer to "human", they just aren't, and the products generated by Ai lack soul. That's cliché and almost meaningless, I know, but the reason writing, art, music, speech, even answers to many simple questions are often meaningful is simply the human experience. When we generate text or an image with our hands, pens, paint brushes, so on, they ring loudly of our life experience, morality, values, bias, and creativity. The way Ai generates it basically just takes from people's voice or style, or even their opinion, modifying it a little but not really creating. This is where fly fishing first comes into play in the conversation. Things have now progressed to the point that I've heard of fly anglers asking Ai what to do on the while water. 

I'm sorry if you've done this... but please get up right now, go to your closest mirror, give yourself a good hard look, then- and not too hard, I'm not trying to injure you here -slap yourself. Are you kidding me? Really? You need a robot to tell you what fly to tie on? This is getting so far from everything that is actually special about fly fishing as a pastime and as a social activity that it frankly disgusts me. There's nothing wrong with asking what you should do in a given situation of an actual human, or learning through reading or videos. You may not even get a perfect answer, or even an answer to your specific question at all, but you're always likely to glean something of worth. Say you ask Joe Humphreys what fly to use on Spring Creek when there's no hatch going on in the middle of an April day. He's liable to tell you to use his Hump's Cress Bug, or some other scud or cressbug imitation, and explain how it'll be best fished rolling on the bottom. He'll tell you how to pick the right number of shot to get down, and to space them a bit to better roll along the bottom. He might even show you how to make a good tuck cast, perhaps tell a story of some good fishing he had recently with that methodology. You'll get the implicit voice of Joe Humphreys and his experiences; his time, his successes and failures. Even if he somehow gives you the wrong answer, that has value. Now look up that same question online, and though there'll be degrees of separation you could sort through information for days, from people of all experience levels. Magazine articles, blogs, forums, images of flies, videos galore... all made by different anglers with different experiences and knowledge, varying approaches and points of view, from all over the world. That, too... incredibly valuable. It isn't immediate, it takes some work on your part, but it shouldn't be immediate. Finally, let's say you're out there on the water alone and don't have anyone there to ask or time to read through loads of information from different sources. Is it not more rewarding, enforcing, and true to the sport to experiment, observe, and try to come to your own conclusion, than it is to ask a machine? If you do ask Ai, it may give you a perfectly good, useful answer, but it will be one lacking in complete context. It'll tell you to tie on a scud because someone, somewhere, likely multiple sources in fact, said it was a good idea. You don't get to learn who all of those someones are, what their history and experience is, and why precisely they think a scud is a good idea unless you put in the extra effort, and at that point you're negating the work the Ai did for you. 

We've already left behind so many of the things that gave this sport soul as technology has continued to progress, are we really going to just ask the machines what we should do now? That sickens me. This is a sport that many of us claims "gets us closer to nature", and yet we seem to try harder and harder to remove as many natural elements from it as possible while shortening the learning curve and ascribing more value to just catching fish at all costs than to the process of exploration. Those aspects are so valuable to the sport, and initially in many respects technology seemed to provide avenues to deepen that. Now, it's skipping multiple important steps. 

Of course, there's a much more pressing concern, not only for fisheries but for whole communities. Currently and probably for perpetuity, Ai EATS energy. So much so, in fact, that it has the potential to impact the health of fisheries. Ai relies on data centers, basically huge warehouses full of computer servers. These servers need electricity to run. Some sources indicate that these data centers, which are popping up all over the US as Ai booms, are set to account for close to half of the energy growth in the country by 2030¹. Ai data centers may use close to a million MWh annually... the average household sits at about .01 MWh. There are 3,912 data centers in the US². Aside from just electricity, data centers need to cool their servers and this can be highly abusive of water resources. Not all of the water is recycled in the process, which typically uses evaporative cooling, and the amount used per day is astounding, well into the millions of gallons. That's millions of gallons per day less than would otherwise be going into people's wells and into groundwater output to spring fed coldwater streams. What water may be discharged would also be far warmer than natural groundwater, and this could have significant impacts on coldwater resource, which trout anglers rely on. Some Trout Unlimited state organizations are already concerned about this reality (mntu.org). The current administration has been pushing through legislation to deregulate data centers and Ai, and even encourage building them on federal land. At a time when water and energy use are already very problematic, and impacting communities and fisheries in a very real way, this is incredibly reckless. Communities around the country are already feeling the impacts, with some residents suddenly feeling pressed to leave places they've spent their whole lives as data centers fundamentally change- in their eyes ruin -their home. They feel no recourse as the largest corporations in the world rush to build these facilities and municipalities jump on potential tax revenue. Right now, as far as the federal government is concerned, it's full speed ahead. It's up to the the states to regulate data centers in such a way that protects towns and sensitive habitat, and there are many indications that they're failing. Though the battle may seem unwinnable, this is one every community really needs to fight.

So, this is something I simply refuse to take part in as much as I can avoid doing so. I won't knowingly interact with Ai generated content, I won't use Ai chat generators to help with my writing or my business even if that means those that do jump ahead initially, and I'll keep skipping past that damn Google overview that's inaccurate half the time anyway. Stupid as it is, it's also scary, well past what I've already discussed above. "While the Level 3 ranking is largely about the model's capability to enable renegade production of nuclear and biological weapons, the Opus also exhibited other troubling behaviors during testing." ³

What the ever living f*** are we doing? 

¹ "AI is set to drive surging electricity demand from data centres while offering the potential to transform how the energy sector works" https://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centres-while-offering-the-potential-to-transform-how-the-energy-sector-works

² https://www.datacentermap.com/usa/

³ Ina Fried, Axios. May 23 2025 "Anthropic's new AI model shows ability to deceive and blackmail" https://www.axios.com/2025/05/23/anthropic-ai-deception-risk

Thank you to my Patrons; Erin, David, John, Elizabeth, Brandon, Christopher, Shawn, Mike, Sara, Franky, Geof, Luke, Noah, Justin, Sean, Tom, Mark, Jake, Chris, Oliver, oddity on Display, Sammy, and Cris & Jennifer, Hunter, Gordon, Thomas, Trevor, Eric, Evan, Javier, Ryan and Dar for making Connecticut Fly Angler possible. If you want to support this blog, look for the Patreon link at the top of the right side-bar in web version.