“I think it is finally time to anthropomorphize the heck out of it”: some brief field notes on conversing with an unreasonably effective yet clearly massively subTuring software machine…
Your twist at the end is very good. For me the path for how to use AI is still not clear. Researching topics I know can be useful. If I know the topic I can assess the accuracy of the summaries. If I don't I can believe the bullshit and don't stand a chance.
Yesterday I decided that it was time to undertake a serious new project with Chatbot, and it was the very first time I'd tried to something with it. Made a new free ID, did a brief dialogue, and broke off to do something else. Logged in again, or tried to, and got a total and persistent failure.
Of course, when I first went to the site, it had asked my permission to load the brand-new v5. In spite of my software development experience, I blandly agreed. (Hey, new software could be pretty flaky in the 1980s-90s! But not today!!) Could not get anything but the newby signup page. Wasted much time working with the highly intelligent AI trying to get back to my previous interrupted dialogue; of course, it all failed in the obvious way: giving me the newby screen, never offering the option to log on to the ID I had established. I finally gave up and swore off the goddam thing.
Clearly the thing had never been through a Beta test at all! If such a thing had happened at Autodesk while I was there, heads would have rolled by nightfall, and I can be sure of that, because I'd have personally lopped them off. (Unless John Walker (q.v.) had happened to be messing around in a break from developing some new feature and had stepped in this.)
That's interesting. I used the app on my phone and asked GPT-4 if I needed to install a new version to get GPT-5. In its typical polite, flattering language it told me I didn't need to do anything. I was already accessing GPT-5. To test it, I asked it some "off the wall" questions, like why did the ancient Hebrews have a personal grudge against snakes, as opposed to other dangerous animals that were common in the Middle East at that time. I got a knowledgeable and intelligent essay on the subject that even referenced Indiana Jones. I can't help but be impressed.
Looking at that top map, it was appallingly bad. This is important, as teh thrust is Agentic AI to act as an intelligent assistant. We don't need incompetent assistants doing anything, from calling a person to booking a restaurant, to booking a vacation. Does the AI know your preferences? Does it know when to pull you into a loop when choices are available? Does it know which platforms are safe to book with a credit card, can it determine scams or red flags, on, for example, Airbnb places? Does it know that southern Europe may now be too hot for many people and that a European vacation should opt for cooler, more northerly destinations?
I wouldn't trust such an agent even in an emergency. If you were a gazillionaire Tech Bro, would you trust such an unreliable agent? The money wasted would be nothing to the extreme irritation of a failed vacation - the equivalent of the Fyre Festival debacle.
Now, just maybe, ChatGPT5 has internalized expert models with scripts that do the sort of narrow tasks that Apple is aiming for. But I doubt that they are even close to prime time. One needs to train an assistant who learns how best to do things to your satisfaction. No agentic assistant is going to do this without making mistakes, sometimes seriously. Furthermore, as ChatGPT is cloud-based, you have no real privacy, whatever OpenAI claims. Everything can change. as we are seeing in the newly fascist USA, with big tech companies falling over themselves to acquiesce to Trump's [illegal] demands.
OTOH, elsewhere, I was impressed with the city simulation software that ChatGPT5 built for Ethan Moellick. Which looked far superior to any vibe coding I have seen so far. On a very different blog, a commenter noted that Musk's Grok4 created a mathematical treatment of a [non-trivial] physics problem that was correct. So there are useful applications.
In the meantime, I don't expect much robustness from agentic AI until we get closer to AGI, or many users can start to train the models, and your personal one[s] can quickly learn what to do without committing mistakes, perhaps in simulation mode, which stops short of completing the task with user confirmation.
And yet people use this “eccentric” tool for important tasks, without the knowledge to know when it is simply making stuff up. Sort of like hiring a President who claims he reduced prices by 1500%. What could be wrong with that?
Re: anthropomorphize the heck out of it. I need to treat the machine as though it were a somewhat eccentric roommate: a companion inclined to fixate on abstruse topics, possessed of unsettling literalism, vulnerable to the occasional non-sequitur, yet blessed with inexhaustible patience and a boundless appetite for our questions.
Aviation movies often focus on the experiences of pilots and the challenges of flying, showcasing both military and civilian aviation. Popular examples include "Top Gun," "Sully," and "The Aviator," which highlight thrilling aerial action and real-life aviation stories.
Top Gun reminded me of the movie made by current Tomcat pilots back in the day when the Tomcat was still in service and the pilots were not retired. "Pop Gun". I found it via Ward Carroll.
If you want to hear what military pilots really sound like go to Gonky and Mover.
Anthropomorphizing is only useful if it provides at least a subset model of behavior. Just saying the target is insane, autistic or chaotic is accepting that anthropomorphizing is pointless. You can talk about light as a particle or a wave, but to do quantum physics you have to accept light as something else.
Your twist at the end is very good. For me the path for how to use AI is still not clear. Researching topics I know can be useful. If I know the topic I can assess the accuracy of the summaries. If I don't I can believe the bullshit and don't stand a chance.
Yikes! Adam crossed with Perikles of the Athenai sounds like Ethan Mollick.
Yesterday I decided that it was time to undertake a serious new project with Chatbot, and it was the very first time I'd tried to something with it. Made a new free ID, did a brief dialogue, and broke off to do something else. Logged in again, or tried to, and got a total and persistent failure.
Of course, when I first went to the site, it had asked my permission to load the brand-new v5. In spite of my software development experience, I blandly agreed. (Hey, new software could be pretty flaky in the 1980s-90s! But not today!!) Could not get anything but the newby signup page. Wasted much time working with the highly intelligent AI trying to get back to my previous interrupted dialogue; of course, it all failed in the obvious way: giving me the newby screen, never offering the option to log on to the ID I had established. I finally gave up and swore off the goddam thing.
Clearly the thing had never been through a Beta test at all! If such a thing had happened at Autodesk while I was there, heads would have rolled by nightfall, and I can be sure of that, because I'd have personally lopped them off. (Unless John Walker (q.v.) had happened to be messing around in a break from developing some new feature and had stepped in this.)
That's interesting. I used the app on my phone and asked GPT-4 if I needed to install a new version to get GPT-5. In its typical polite, flattering language it told me I didn't need to do anything. I was already accessing GPT-5. To test it, I asked it some "off the wall" questions, like why did the ancient Hebrews have a personal grudge against snakes, as opposed to other dangerous animals that were common in the Middle East at that time. I got a knowledgeable and intelligent essay on the subject that even referenced Indiana Jones. I can't help but be impressed.
Beautiful place, btw. What keeps cities from being more like that, vs [new apartment buildings on El Camino Real]?
The demographic segregation is jostling though.
Looking at that top map, it was appallingly bad. This is important, as teh thrust is Agentic AI to act as an intelligent assistant. We don't need incompetent assistants doing anything, from calling a person to booking a restaurant, to booking a vacation. Does the AI know your preferences? Does it know when to pull you into a loop when choices are available? Does it know which platforms are safe to book with a credit card, can it determine scams or red flags, on, for example, Airbnb places? Does it know that southern Europe may now be too hot for many people and that a European vacation should opt for cooler, more northerly destinations?
I wouldn't trust such an agent even in an emergency. If you were a gazillionaire Tech Bro, would you trust such an unreliable agent? The money wasted would be nothing to the extreme irritation of a failed vacation - the equivalent of the Fyre Festival debacle.
Now, just maybe, ChatGPT5 has internalized expert models with scripts that do the sort of narrow tasks that Apple is aiming for. But I doubt that they are even close to prime time. One needs to train an assistant who learns how best to do things to your satisfaction. No agentic assistant is going to do this without making mistakes, sometimes seriously. Furthermore, as ChatGPT is cloud-based, you have no real privacy, whatever OpenAI claims. Everything can change. as we are seeing in the newly fascist USA, with big tech companies falling over themselves to acquiesce to Trump's [illegal] demands.
OTOH, elsewhere, I was impressed with the city simulation software that ChatGPT5 built for Ethan Moellick. Which looked far superior to any vibe coding I have seen so far. On a very different blog, a commenter noted that Musk's Grok4 created a mathematical treatment of a [non-trivial] physics problem that was correct. So there are useful applications.
In the meantime, I don't expect much robustness from agentic AI until we get closer to AGI, or many users can start to train the models, and your personal one[s] can quickly learn what to do without committing mistakes, perhaps in simulation mode, which stops short of completing the task with user confirmation.
And yet people use this “eccentric” tool for important tasks, without the knowledge to know when it is simply making stuff up. Sort of like hiring a President who claims he reduced prices by 1500%. What could be wrong with that?
A look inside the cult.
https://www.bloodinthemachine.com/p/ai-disagreements
Re: anthropomorphize the heck out of it. I need to treat the machine as though it were a somewhat eccentric roommate: a companion inclined to fixate on abstruse topics, possessed of unsettling literalism, vulnerable to the occasional non-sequitur, yet blessed with inexhaustible patience and a boundless appetite for our questions.
MMMM?
Sounds like a high functioning autistic.
I'm sure some of the humans who design AI systems are high functioning autistics.
I stumbled upon the following:
Aviation movies often focus on the experiences of pilots and the challenges of flying, showcasing both military and civilian aviation. Popular examples include "Top Gun," "Sully," and "The Aviator," which highlight thrilling aerial action and real-life aviation stories.
aviex.goflexair.com avi-8.com
More
Auto-generated based on listed sources.
Why bother with AI? I could write this in 10 minutes max and it wouldn't need to be checked for craziness. Allen Kamp
Top Gun reminded me of the movie made by current Tomcat pilots back in the day when the Tomcat was still in service and the pilots were not retired. "Pop Gun". I found it via Ward Carroll.
If you want to hear what military pilots really sound like go to Gonky and Mover.
Another interesting opinion on AI: https://freddiedeboer.substack.com/p/llm-hallucination-seems-like-a-very?utm_source=post-email-title&publication_id=295937&post_id=170809877&utm_campaign=email-post-title&isFreemail=true&r=15flyq&triedRedirect=true&utm_medium=email
Thx much... -B.
> JH: 'Another interesting opinion on AI: https://freddiedeboer.substack.com/p/llm-hallucination-seems-like-a-very?utm_source=post-email-title&publication_id=295937&post_id=170809877&utm_campaign=email-post-title&isFreemail=true&r=15flyq&triedRedirect=true&utm_medium=email
Anthropomorphizing is only useful if it provides at least a subset model of behavior. Just saying the target is insane, autistic or chaotic is accepting that anthropomorphizing is pointless. You can talk about light as a particle or a wave, but to do quantum physics you have to accept light as something else.