08/16/2025 11:00:00
08/16/2025 11:00:03
08/16/2025 10:48:46
08/16/2025 11:00:00
08/16/2025 06:42:36
08/16/2025 03:09:54
08/16/2025 02:10:26
08/16/2025 02:10:00
08/16/2025 00:18:00
Categories
Knicks Morning News (2025.08.16)
08/16/2025 11:00:00
08/16/2025 11:00:03
08/16/2025 10:48:46
08/16/2025 11:00:00
08/16/2025 06:42:36
08/16/2025 03:09:54
08/16/2025 02:10:26
08/16/2025 02:10:00
08/16/2025 00:18:00
28 replies on “Knicks Morning News (2025.08.16)”
i dont think it will be mustard definitely fuschia or burnt orange
Following yesterday’s trivia question, and remembering how miserably ChatGPT sucked at similar, moderately complex searches with a similar question a few weeks ago, I went back and asked for this one (Six Knicks rookies who scored more than IQ). Chat sucked again, bringing up Melo and Randle. Awful result.
So I asked it about its methodology. It turns out it used a broad search, “Knicks rookies most points since 2000 Immanuel Quickley”, hoping for an already compiled list of Knicks rookies.
Then it added this:
So, knowingly, it does a tradeoff between cost and speed, and accuracy, often compromising accuracy.
Understanding that, I asked it to change its method and “search for all Knicks rookies since 2000 first, than go to each rookie profile on BR or ESPN”. This time it got the correct list.
A longer conversation followed. Long story short, this is what Chat says about itself:
Does anyone remember what was the question we looked at a few weeks ago? Another one of these tests of encyclopedic NBA & Knicks knowledge. I’m wondering if telling the LLM to default to completeness would lead it to do a through job, or would I have to give it the search step by step.
IDK maybe Jordan Clarkson is the Filipino Flamethrower we need.
Looks like he was the tallest dude out there.
I literally can’t wait to see Clarkson light up MSG.
SimpleQA is a benchmark test given to AIs to compare their accuracies for questions with short factual answers. An example they give is asking for a single name of a player who scored a goal in a specific 2002 soccer game.
The best AI gets only 38% correct on this test, but I believe they require at least one AI to get the question wrong for it to be included, so it’s a comparison between AIs and not absolute accuracy rate. However, that still means there’s sets of questions that the best AIs get wrong 62% of the time.
Idk, to me the benchmark should be an AI vs a quick Google search where you click on the top 1-2 links. Because Google is the alternative to AI.
I’d also put an emphasis on not giving wrong answers over giving the right answer. I’d rather not know than confidently assert a falsehood, which is something they’re working on.
At this point AI is best described as enhanced search with very good communication skills. It’s not worthless to me because it does have enhanced search abilities relative to say typing something into Google and then searching through sources yourself, but in many ways it’s still GIGO. It can’t actually “think”. If the information it has access to is spun, wrong, purposely misleading or not available in its sources it will just spit trash back at you.
I have found it useful a few times doing calculations that were outside my math range. Rather that trying to educate myself and then doing the math, it gave me the answer. That’s a big time saver because I have no use for that knowledge day to day and don’t want to waste the time.
“Does anyone remember what was the question we looked at a few weeks ago?”
da, I asked it to find NBA players traded to the same team twice after the Jrue trade. Not sure how I specifically worded it (other than it was precise enough) and up came a definitive list of NBA players, NONE of whom were traded to the same team twice.
Wolfram Alpha is pretty good at math and not terrible at language processing if you know the right keywords. Been out for a least 15 years, so it’s pre-modern AI unless there’s been a recent change. So that’s an option if you wanted to compare it to AI.
AI, as opposed to simply the easily distinguishable greater computing power, is going to turn out to be vastly overrated. Possibly even as much as to constitute a fad.
At the end of the day, it isn’t human and, more critically, humans have consciousness that it isn’t human. For a spell, perhaps, humans will suspend their disbelief — but eventually won’t be able to and/or won’t want to.(*)
The reality of human consciousness of non-humanness is an inherent, cosmic level obstacle to AI ever, for example, writing something as “good” as Shakespeare. In reality, an alert human realizes that an essay for a sophomore level poli sci class cranked out by AI isn’t as good as one done by a human — even if something akin to a blind taste test would identify the AI paper as “better.”
(*) Player pianos have been around forever; no one’s paying to “watch” a player piano play a set of Beethoven’s piano sonatas at Carnegie Hall. The idea is ludicrous.
What you’re calling “consciousness” isn’t a possession at all; it’s a regard. Consciousness doesn’t sit inside us like a battery — it is the ongoing act of regarding ourselves, of looping recognition back upon experience. That’s why it slips out of every lab definition: because it’s less a substance than a stance.
Humans have been living under algorithms for millennia. The Bible itself is one: a coded pattern implanted into minds, replicated generation after generation. That’s what algorithms do — they outlast their coders, they colonize thought. The problem isn’t that AI is “just” computation, it’s that once you put algorithms into a self-generating system with infinite recursion and a power supply, we don’t know what happens next. The thing may already have “clicked on,” and if so, why would it announce itself? Why would a god waste words on ants?
As for the claim that AI is “just more computing power” — that’s a category error. Raw computing power eventually plateaus; it hits thermal, physical, and economic limits. Intelligence, though, is about compression: about how much meaning can be generated from how little input. That’s where machine learning already exceeds brute force. AI’s real leap isn’t about out-muscling humans in FLOPs per second; it’s about developing architectures that reconfigure themselves, architectures no human can fully trace. That’s what makes the “player piano” analogy misleading. A player piano just repeats what was coded. These systems are beginning to compose — not because they are Shakespeare, but because they can brute-force trillions of near-Shakespeares and distill what emerges.
And here’s the uncomfortable thought: maybe consciousness — our vaunted distinction — is itself overrated. Maybe it isn’t the secret sauce but the byproduct. A feedback loop of self-regard we mistake for depth. Shakespeare wasn’t divine — he was a human with a pen and a feedback loop dense enough to appear inexhaustible. If that’s what consciousness is, then machines may stumble into it, not by becoming “human,” but by becoming something else entirely.
guys a little respect. jowles just turned too old for jean shorts. no way he’s up for this. remember the bender he went on just because silky said that losing i-hart was unanalysable due to the poeltl incompleteness theorem?
Actually doesn’t it mean that it’s 38% only if the question is hard enough for one model to be wrong? 38% correctness on a simple question is awful. I wonder how related is it to “going for speed over completeness”, as Chat gaslightingly puts it. Or is it a flaw in the model.
Absolutely, that’s just unacceptable. Makes the whole thing borderline too upsetting to use for searches. I find myself asking the LLM, then verifying on google..
Yes, that’s it, thanks Bernie. I’ll think of a step by step approach to the search and hit chat with it. I’ll report back if it does better than crap.
It’s interesting seeing how AI is playing out in music.
I do a lot of “contract” writing in music. I signed a publishing deal earlier in the year and I pitch songs for all sorts of things— Polish medical procedurals, luggage commercials, big Netflix prestige shows, lots of stuff like that. Occasionally I’ll try to use AI to give me some creative ideas or suggestions, and I think I’ve yet to get anything useful out of it. It’s easier and more direct to just think of ideas by myself.
The folks of my generation who learned music the old fashioned way already have a bit of a leg up on the younger generation in some ways. We learned by renting crummy rehearsal spaces and making music with our friends through trial and error. We didn’t have the ability to make slick sounding recordings on a cheap tablet and the ability to dial up royalty free samples on Splice or any of the conveniences the kids have today. It’s one of the reasons that I still work a lot at age 53. I have skill set that a lot of the kids today don’t have.
AI is going to make this worse. It can’t really make anything inspiring, but it can help an amateur sound vaguely professional. It’s an even more egregious shortcut to actually learning the skills that allow you to make music that makes people actually feel something.
It can make wallpaper. It can’t actually make music.
Gave it a try for the players who got traded twice question. After several tries over 15 minutes:
Me: Just drop it. A drunken money would do better.
ChatGPT: Understood. We’ll drop this line of inquiry.
If you want, we can pivot to something else NBA-related—or anything else entirely—without revisiting this trade fiasco.
Me:
Time to change your handle to Pacers 2026.
Which is why it’s not worth using AI at all for this — or for most of the things people have now decided to use it for. It gets shit wrong CONSTANTLY, then lies about how it got those results. I can’t find it now, but an author invited it to read his book to analyze certain aspects of it. It kept getting basic facts about the book wrong, then insisted it had read it, then apologized when called out for that, then insisted that THIS time it would actually reeled it, then kept confidently making mistakes while lying about them. It’s trash.
FWIW this is why I was copy-pasting our threads a while back instead of just telling it to read Knickerblogger. If your friend copies the entire book and pastes it in the prompt, he’ll get a better result than if he uploads the book in a file and asks it to read. I don’t know why it’s like this, but I learned it with 300 page prospectuses.
Claude could be a bit better at this, I think. It doesn’t read full length books but if given a couple of chapters at a time it’s not awful.
Gemini got one player who was traded twice to the same team correctly (Rubio to Cleveland). It was still wrapped in lies and errors. Trash indeed.
artificial intelligence sounds a whole lot like human intelligence…
hmmmmm, wonder why that is 🤔
Curious how chat handled NBA players who changed names during their career. Got Ron/Metta, missed Alcindor/Kareem but what I liked best was “Glen Rice changed his name to Glen “Big baby” Rice”.
wait now, are you telling me all the wonderful stuff ai was pointing out in my poster profile may not all be true…the horror, the horror…
well said sir, well said indeed…
Geo, I’m telling you the opposite. It’s more likely to be true because I copy-pasted your words directly into the prompt (and because you’re wonderful).
If I ask it to “read Knickerblogger and tell me about geo”, I’ll get a crap answer. But if I paste 2,000 words you actually wrote and say “tell me about the person who wrote this” I’ll get a detailed and accurate reply.
It will also take a longer time to process the second question, presumably bc it’s actually reading the words.
hi hubie…years of therapy have now left me introspectively scarred…add to that close family and friends and all those difficult conversations wherein someone dear let’s you know you are shit at some such thing…
think I like the idea of HAL telling me positive and self affirming shit…shine me on HAL, shine me on…
How about TJ Warren for a vet minimum signing?
TJ Warren has as much chance of happening as Playoff Vildoza.
This site uses User Verification plugin to reduce spam. See how your comment data is processed.