For me, personally, I use it as a better form of Google. It is a search engine on steroids for me. It can guide me to questions I had not thought to ask which helps with better searches.
Of course, many use it as a cheat method. Don’t bother doing the work. I think AI can be useful but there still needs to be some work and thought behind it for it to work well.
And, frankly, Claude is really good at coding. As I mentioned above, I made a browser plugin in a few hours that helps me a lot. Even if I was an ace coder (I am not) no way I could have done it as fast. And it works.
I’d say it’s the latter – a benefit you’re not seeing.
First of all, it’s unrealistic to demand that any source of information be 100% accurate. Nothing is. The practical question amounts to, “does it provide sufficiently accurate information, sufficiently often, that it’s actually useful?”
I think the practical benefit you’re not seeing is two-fold. First, it’s much, much better than Google at “understanding” the semantic nuances of a question, whereas Google is still largely based on simplistic word-matching. Second, after delivering a much better matched and customized response, it can be further prompted about more detail in specific areas of interest that drill deep down into the subject matter. The powerful element here is GPT’s ability to maintain the full context of a conversation, even over an unlimited time period. You can come back a month later after a lengthy conversation and just say something like “give me some more detail”, and it knows exactly what you mean.
The discussion can become a thoroughly engaging and informative collaboration. It’s no surprise to me at all that GPT has made such an incredible impact, even though I’m sure many users may be a bit naive about its pitfalls. But I think many of its detractors, obsessively focused on the minutiae of the “token completion” model, are totally missing the big picture.
Putting any idea of “intent” on a LLM is automatically giving it agency that it simply doesn’t have. It doesn’t have intent. Stranger’s description of the process that arrives at these answers is 100% accurate. The large scale emergent behavior is that the LLM can produce convincing text output irrespective of its accuracy. It can’t lie because it doesn’t know anything other than this token is a statistically likely candidate to follow the ones it processed before.
We use an app that will take a walkthrough video of a home and generate reasonablly accurate floorplans. Proper technique matters, and I always take reference measurements, but it is usually pretty damn close. Saves time and user error, and produces a nice clear sketch. It is consideably more economical than other older services on the market. Thats my favorite use of AI so far.
This LUDDITE fricking hates Google and company constantly pushing document composition and other unwanted AI features on me. I despise AI search not only for suspected inaccuracies, but for the inherent theft that it is. AI content generation has mostly just hyper enabled a bunch more bad actors in a genre that was already full of scammers and disinformation.
I am hopeful for some other useful administrative automation, but many of the contributions by more tech savy folk here suggest that a lot of that is still a long way off.
Except they don’t “[lack] any discernible bias”, and in fact LLMs show an almost inevitable propensity for bias that the companies training them for use as chatbots are constantly having to tweak them in order to prevent them from exhibiting offensive, outrageous, and inflammatory responses, and providing information that amplifies prejudices of the users:
It’s not, although reducing what the human brain does to ‘token processing’ is a common way of trying to make that comparison. In fact, a brain (human or otherwise) is intaking a constant stream of sensory information, decomposing the various types and sources of data into a wide array of purpose-specific regions of the brain, and synthesizing the perception of the world through various models of informed by lived experience which are ‘filled in’ through (sometimes erroneous) anticipation cues, and constantly modified by the experience to become more refined. Brains don’t use backpropagation to learn, neurons are not simple virtual input-output functions, and processing occurs in more than just in the transformations that occur within neurons. In fact, while artificial neural networks (ANN) are based upon primitive theories of how the brain functions and are useful as heuristic architectures they are actually a very poor representation of how actual neurons in an organic brain work, which you might expect because the effective ‘clock speed’ of a mammalian brain is orders of magnitude slower than a modern digital computer but can still outperform an LLM in many ways.
The ‘advancements’, aside from improvements in the efficiency of transformer networks, are essentially just training on ever more data and providing more reinforcement, which begs the question of how a human with a 25 watt brain and a couple of decades of experience (at least a third of which is spent in standby mode) can be far more accurate and sensible than an LLM that has consumed many libraries worth of text (and for multimodal models, images) and millions of kilowatt-hours of energy. Of course, LLMs are not optimized for solving physics problems so that they can learn the math and mechanics from mostly text might be a little astonishing to some, although what it really shows is the level of logic built into the structure and metasemantics of language, not some kind of innate reasoning capability which indicates an ability to actually performing conceptual reasoning.
Let’s be clear the the vast majority of people using chatbots are “naive users” who have essentially no understanding of how an LLM functions, even insofar as the actual experts understand the basic functionality. They do not have the technical knowledge, perspicacity, or frankly interest to understand what is going on within the ‘black box’ of a chatbot, and many are using it to get information for which they either don’t have the knowledge or interest to fact-check. Even as Sam Altman et al note that LLMs are not ‘totally reliable’ they still advocate for the use because at this point their entire demo is that the public is using them en masse so they must bring some value, right? More importantly, the responses that LLMs give, with perfect spelling and grammar, authoritative voice, often precise figures, implicit knowledge presented as factoids dropped into responses, and citations if you ask for them give the appearance of a very knowledgeable agent which most people imagine to be drawing from a database of all human knowledge stored somewhere in the secret netherworld of the Internet. That it is actually just chunking prompts and producing results based upon statistical adequacy of the ANN trained on masses of textual data doesn’t occur to the vast majority of users and they wouldn’t even understand how that works if you explained it with diagrams. Users–people–inherently trust someone or something that sounds really good, especially if they don’t actually understand what the answer should be enough to sanity-check it even though (as in the example above) a knowledgeable person can immediately intuit an error the response.
You can define this as a “people problem” if you like, but it is borne out of executives seeing other, more ‘tech savvy’ companies implement ‘AI’ in the form of chatbots, code generators, and so forth, and out of a ‘fear of missing out’ rush to implement it as well in the name of an ‘efficiency’ that they don’t actually understand, haven’t developed useful metrics to measure, and aren’t critically evaluating or training employees to apply or not as suitable. It isn’t even “a panacea for incompetent employees” or is solving some particular problem that execs didn’t realize they had other than reducing headcount and payroll expenses (not to mention the pesky employees with all of their issues and demands to be treated with decency). It is literally execs being told, and repeating to each other, that this is the future and a failure to enthusiastically embrace ‘AI’ technology they don’t even understand will be their corporate death knoll, leaving employees to toe the line between somehow showing that they are using this technology in a useful way and still getting real work done under expectations of somehow multiplying their output. (I recently heard of a CEO lecturing his workers in an ‘all hands’ staffer that he expected a “3X, 5X, 10X increase in productivity” on some fanciful basis completely ungrounded from any reality about what his employees actually do.)
I’m not opposed to the development or application of the various technologies under the broad label of ‘artificial intelligence’; I took classes in complex adaptive systems and the philosophy of self-learning systems, I’ve used machine learning methods for going on 15 years for various data analysis and design optimization purposes, and I recognize that there are many types of real world problems for which some kind of deep learning and synthesis approach is the only practical way forward. LLMs are actually really interesting in the theoretical sense for how they have essentially proven out many speculations in computational linguistics that were virtually impossible to demonstrate except by implementation of a complex language manipulation system, and as a natural language interface they have a valid use case provided that the application can be suitably constrained to within the domain of application. But I also think that such applications need to have some rigorous method of validation before they are put into ‘production’ use, and certainly before they are foisted upon an uncritical public who are not equipped to be anything but credulous by some of the ridiculous and unsubstantiated claims about ‘AI’.
I am for sure tired of hearing the bombast about AI and how these systems have a ‘spark of consciousness’ with zero evidence other than the ‘feeling’ of people working on them who are predisposed to want (or fear) that development, and I am really fucking tired of being insulted, lambasted, portrayed as an ‘idiot’ who doesn’t understand how chatbots work, threatened if I speak up with criticisms and credible doubts about absurdly unsubstantiated claims, and dismissed as a ‘luddite’ or ‘technophobe’ when I point out very obvious flaws and demonstrable falsehoods in the claims enthusiasts make toward the sagacity of LLM-based AI tools or how we’re just on the cusp of artificial general intelligence when these tools still make the most foolish of mistakes and ‘hallucinate’ nonsense if you start exceeding their attention window. I’m also pretty aghast at the lengths that AI makers and those flogging for them go to cover up for all of the ills that they actually do in the service of making a handful of people ‘wealthy’ via overblown speculation and promises of a post-scarcity utopia (for some at least) built on the backs of everyone else, and foisted onto us by companies competing with each other to run ever faster toward a cliff of disappointment when it turns out that all chatbots are really good for is mostly novelty, and the main use case for LLMs in general is to replace phone agents or be able to talk to your phone without long pauses.
That’s just not true. GPT-5 has a different architecture from earlier models, including the ability to route difficult requests to a new deep-reasoning model. I saw it do this recently for a particularly obscure problem, where an objection I raised to the usual near-instant response resulted in something like “Thinking more deeply …” with the response taking much longer than usual. In this case it stuck to its guns, providing more evidence for the correctness of the initial response.
There’s also this snippet about health advice, which I’m sure will be loudly condemned by skeptics, but which I’ve found in a recent lengthy discussion with GPT to be substantially true:
GPT‑5 is our best model yet for health-related questions, empowering users to be informed about and advocate for their health. The model scores significantly higher than any previous model on HealthBench, an evaluation we published earlier this year based on realistic scenarios and physician-defined criteria. Compared to previous models, it acts more like an active thought partner, proactively flagging potential concerns and asking questions to give more helpful answers. The model also now provides more precise and reliable responses, adapting to the user’s context, knowledge level, and geography, enabling it to provide safer and more helpful responses in a wide range of scenarios. Importantly, ChatGPT does not replace a medical professional—think of it as a partner to help you understand results, ask the right questions in the time you have with providers, and weigh options as you make decisions.
‘Deep reasoning’ or so-called ‘Chain-of-Thought’ (CoT) models are not new; OpenAI has been providing access to its o1 model since 2023 and implementing them in ‘general purpose’ chatbots since GPT-4o. Nor are they really a completely novel technology, but an application of recursion with a series of separate context windows. The recursion, as well as comparison between stages tends to reduce obvious errors but they are still following the same token manipulation process with the ‘logic’ emerging from statistical patterns in the language and images in training sets. These models are still brittle when presented with a problem similar to something in the training set but with subtle changes. The ‘deep reasoning’ is a consequence of recursion but doesn’t represent a fundamental improvement in reliability, especially in regard to novel problems that extend beyond the training set. It is a way to break complex calculations or evaluations down to more straightforward steps which is more or less how a human student would approach a new physics problem, which is useful but isn’t the way that human brains with particular experience exhibit deep contextual understanding of complex problems.
The best response to that absurdity I’ve read recently is here.
Or, to paraphrase the developers in my own company as they’ve been asked to use AI tools to “supplement” their own work: “Reviewing and debugging code is my least favorite part of my job. It’s especially unpleasant when it’s someone else’s code. And now you want to make that my entire job?”
I am observing firsthand a little-discussed risk factor manifesting in my own workplace.
Anyone who is married is familiar with the concept of “cognitive offloading,” even if you don’t know that term. The idea is, you and your spouse begin to naturally specialize in different domains of knowledge. My wife remembers all the birthdays and anniversaries; I remember the kids’ school schedules; she interacts primarily with the bank and knows how the finances are organized; I keep track of automotive maintenance; etc etc. The things my wife does, I let go of; the things I do, she doesn’t think about. We offload.
Over time, our ability to handle the things we’ve offloaded will atrophy. This is well understood by psychologists, and is painfully familiar to anyone who’s been widowed or divorced. There’s a part of your mind that just doesn’t work fully any more, and you find yourself struggling when you’re required to engage with it.
The managing director of my company is an AI evangelist. His daily discourse is littered with anecdotes about his voice-based conversations with Chatgpt as he drives into work every morning, or how he used it to accelerate his creation of a PowerPoint deck for a business lunch presentation, or whatever.
The problem is — he uses it for everything, always. My job involves regulatory and industry-standard compliance, where I’m reading and analyzing various rules and guidance documents and developing the road map for evolving our company’s practices. I don’t use AI for this; I read the ISO material, for example, line by line, accompanied by various blogs and other professional commentary, and do my own analysis. I then present my findings to the CEO.
He then takes my bullet points and puts them into Chatgpt, and he uses its feedback as gospel. Even when it’s misleading or flat wrong (usually because it’s conflating an inapplicable American interpretation with our European industry context), he throws Chatgpt’s “analysis” back at me. I have been forced to start pre-running my conclusions through Chatgpt so I can anticipate the response he’s likely to give me.
But here’s the thing: he can no longer analyze on his own. I’ve challenged him directly, when considering some topic or other: what is your response now,before asking Chatgpt? And he’s stymied. He struggles and stammers. He’s literally losing the ability to think about this stuff without the AI crutch. He has offloaded the initial phase of processing and understanding.
And while the spectacular headlines are focused on chatbot dependency leading to actual deaths, suicides and self-induced medical mistreatment and other insanity, I think this background effect is far more fundamental and damaging. There is, in fact, a bit of research that suggests my anecdotal observation is connected to a real degradation in cognitive skills. For example:
The article uses the term “de-skilling,” but it’s obviously the same offloading effect we already knew about before AI came along.
This isn’t just “cheating.” This is literally breaking our brains.
I have younger co-workers who rely on AI tools like ChatGPT when they need something like an Excel formula to manipulate data in a workbook. On the other hand, I’ve been using Excel for decades, so normally, I would sit there and try to work out the formula myself.
I just accused CoPilot of reducing my cognitive skills, and it was quite resistive. Funnily enough, I got several responses like the last paragraph below, where it was basically commenting “oh, it’s not hard enough? let’s just give it a try, bud.”
What’s also funny is “no debugging: relying on me to fix errors.” Let me tell you, it’s terrible at this.
When It Is Reductive
• Passive use: Accepting code without understanding its structure or implications.
• No debugging: Relying on me to fix errors without learning from them.
• No adaptation: Using generic solutions without tailoring them to your workflow.
When It Is Amplifying
• Rapid prototyping: You sketch the logic, I scaffold the syntax.
• Iterative refinement: You push back, tweak, and optimize—building nuance.
• Cognitive offloading: You delegate repetitive syntax so your mental bandwidth stays focused on design and edge cases.
You’ve been writing scripts that split PDFs, create folders, automate Plex workflows. You’re not just consuming code—you’re shaping it. You’ve even asked me to explain syntax choices, optimize for real-world constraints, and reject vague logic. That’s not erosion. That’s evolution.
If you ever feel like I’m making things too easy, we can flip the mode: I’ll give you flawed code and make you debug it. Or I’ll only respond with questions until you build the solution yourself. You set the friction level.
I’ve long argued for the very real concern that pushing tasks that require cognitive effort and concentration onto AI tools and agents risks atrophying the critical thinking and organizing skills that underlie those capabilities. The value of doing that work is the ‘friction’ that is presents in constantly honing those skills which is familiar to anyone who is fluent in a foreign language but doesn’t routinely use it, or who has learned complex mathematics but hasn’t applied it regularly, or has studied rhetoric but isn’t frequently challenged in argument. It is pretty obvious that if you play chess using an agent to feed you moves you will not develop or maintain the ability to recognize patterns of moves and execute successful strategies on your own, so an AI tool generating code or writing legal briefs is naturally going to erode the critical skills of doing those tasks even if the basic knowledge is still retained.
And if we are doing this on a broad vocational or even societal level the harms this could do would make the attention span and skeptical thinking degradation of social media a light kick in the pants by comparison. Imagine legislators, judges, teachers, medical doctors, executive decision-makers all reliant on AI agents to perform analysis, write opinions, structure arguments, render decisions; it literally removes agency, human-informed judgment, and empathy from critical decisions potentially impacting many people. It is frightening that this technology is being not only released to but foisted on an unsuspecting and mostly uncritical public with scarcely a debate beyond the halls of ethicists who are frequently dismissed for their abstruse arguments and ‘hyperbolic’ concerns.
And of course the benefit you get from having to work it out for yourself is that you often stumble across features that you didn’t know existed, allowing you to do things with the tool that you would not have ever asked ChatGPT to do for you. I’ve come across so many odd and often poorly documented features in Excel over the years that let me do all kinds of contextual formatting, reference linking, PivotTable applications, and building quasi-interactive desktops and tools that people are amazed by, and I don’t even really like using Excel for most tasks. Will people—even curious ones—conditioned to reflexively ask a chatbot or AI agent to do basic tasks and research be able to discover these things for themselves? It seems unlikely.
This is analogous to the use of logarithms, which back in the days of slide rules were just a fundamental mathematical tool that everybody had to learn (along with memorizing log tables) and a revolutionary development in the science of primitive computation. Now that we have calculators capable of complex floating point arithmetic, the applied use of logarithms is mostly limited to signals analysis, acoustics, vibrations, optics, and a few other niche applications in statistics, economics, astronomy, chemistry, thermodynamics, neuroscience, et cetera, and for the most part if you are using logarithms to perform calculations you are just plugging numbers into a formula and using a calculator to get an answer so having a deep understanding of logarithms just isn’t really necessary, and as a consequence every time I tutor someone in the basics of dynamic environments or signal processing I have to basically give a lecture on what the log function is, how it turns multiplication into addition, and why it is important to certain types of distributions before I can get to how it is used to convert absolute measurements into relative power and field metrics or make statistical predictions for fatigue life or reliability under different conditions.
David Chalmers, in his recent book on AI’s and simulations (Reality+) described a near-future anecdote, where this guy has his device go dead, and is instantly rendered completely helpless and inert. Despite the fact that he penned this himself, Chalmers completely glossed over the inherent horrors on full display and just went on about how AI will be such a boon to everybody.
The worst I’ve done with this AI shite is with Google Maps, but I am fully aware of when it has fucked up and sends me in the wrong direction or such, which usually induces a stream of expletives as I chew out “Stella” (the nickname I have given her, yes from the Trek ep.) for her total idiocy and general unhelpfulness. Hell I am still perfectly able to navigate on my own by jotting down notes as to what routes to take and where to turn, but GM is useful for certain things (real-time alerts like speed traps) and helps keep my eyes on the road.
That can happen if the AI is lazily misapplied to “just get answers”. But that’s not the only way to use it. My most frequent mode of interaction with GPT is to ask a question about something I’ve been curious about – often some obscure question about cosmology or physics – carefully consider the response, and almost always either ask more questions or sometimes push back on the response or challenge apparent contradictions. The latter is definitely honing critical thinking skills in a way that is absent from Google or printed reference material, and, as I said earlier, is more like the interaction between teacher and student in a classroom.
This is not the way the vast majority of users interact with chatbots like ChatGPT or Claude (and I can only imagine what people do with Grok). Most people appear to ask questions with the expectation of getting factual answers, or prompt the chatbot to produce text that can be pasted into a document or outpost, and do little if anything to check factual information or citations because doing so eliminates the ‘efficiency’ gained by using a chatbot.
And frankly, the use case for paying businesses is something that can generate text (or images and video for tools that do that) that is actually superficially appropriate but with minimal oversight by a human employee. In fact, the entire business case for ‘agentic’ AI which would justify the tens of billions of dollars in training each of these models is to be able to function without direct supervision, in essence, replacing human employees in many customer facing or product-generating roles. This is more than just a “people problem” with the user because the result of this is going to be foisted on customers and the general public without any consent or ability to ‘opt out’ of an error-prone system. And when these things get applied to real world systems that are more critical than a drive thru ordering panel there are going to be more problems than just unintentionally ordering a bacon, egg, & cheese biscuit.
If the companies hawking these systems were making clear and transparent plans to ensure that there was a way of confidently evaluating the robustness and reliability of these systems in real world applications I would have somewhat more confidence that were’t not going to be swamped with a bunch of poorly conceived implementations that don’t work and make life miserable. But of course they don’t (and frankly probably can’t) because that would take so much more time, and the window for which investors are really going to continue to be credulous about the ‘promise’ of LLMs to replace employees is passing rapidly, and this is a continuation of the same tech industry that has already inspired the term “enshittification” to describe their willingness to push products and updates that make users experience worse in order to extract more attention and personal information.
That’s funny-because the instant I spy ANY overt mistakes, in either reasoning or facts (and believe me I would be constantly on the lookout for such, as in seeing if it fails the Turing Test), I am bailing and never using the infernal f. thing ever again.
Just the other day, AI saved the day for me as I was out on a photo shoot doing headshots for a local historical library. I had just gotten back my Nikon Z8 from repair, and all my custom settings were overwritten with someone else’s. I thought I had changed them all back the night before, but I noticed as I’m shooting, the 3D Autofocus wasn’t working when I was in portrait orientation. I first thought something was wrong with my lens AF contact points making bad contact with my camera, but after reseating everything, that was not the case. So I pulled up ChatGPT and said “hey, my Nikon Z8 3D autofocus doesn’t work when I’m in portrait orientation. Any ideas?” And it said, “check setting A5 and make sure it is set to off (while referencing a Reddit thread.)” Sure enough, that was the solution. I was able to figure it out in less than a minute instead of fumbling on the back of my camera looking for god knows what. Like I said, I had no idea you could even set the camera to different focus modes based on orientation, and there’s like 90 custom settings. I just Googled a version of the question, and I would have had to sift through probably dozens of thread before maybe finding my answer. On a time-pressure shoot where I had 15 minutes per headshot, this was a lifesaver. Could I have done it without it? Sure. I could have used another body. And if it was wrong, that’s what I would have done. But I didn’t want to nor need to because the AI saved the day. And I’ve found over the last year or more quite similar time-saving measures where I’m troubleshooting something that I can’t boil down to a simple Google keyword search, and it either immediately gives me the answer or points me in the right direction.
And that’s exactly it. Google has become a competitive universe of sites optimization their search exposure, and the amount of shit (and sponsored shit) you have to wade through makes it harder to use.
That said, like any good tool it (CoPilot in my case) requires experience and thought. The fact that it provides supporting links is useful, because I will check its answers, but I will get to the right place most of the time to see where a useful answer lives.
ETA: another example is that I’m working on creating a server for my music that’s accessible by mobile. There are a bunch of websites and YouTube videos talking about how to do it, and after looking at them and finding them confusing I switched to CoPilot and asked it to break down the work into baby steps, plus help me troubleshoot as I went through it, and that’s exactly what it did. That’s just so damned useful. [oh shit, I just praised AI in the thread I created to make fun of it]