John Elder
Region: US
Friday 28 August 2026 15:19:10 GMT
72904
1855
533
487
Music
Download
Comments
Joseph Borbely :
as a software architect i couldn't agree more. I made this point just this morning
2026-09-04 04:18:10
0
Eugene :
I agree it’s not “intelligence” in the human sense. For some it’s entertainment; for others it’s a huge domain accelerator.
2026-09-03 13:39:09
0
@AmMoPy :
AI as it is now is just Automation, you give context it spits out its training data. Thats what really differentiate one model from another, training data. How it spits out this data is the I in AI, call it reasoning or whatever, its just layers or corrections before the final output. A simple example would be an ensemble in classic ML, just layers of models voting on the final output. Im a heavy user of AI and I love it and advocate for it, John is addressing the Hype wich is true.
2026-09-03 15:40:06
0
keelsoftinc :
I learnt django with ur course on YouTube, I think I’m gonna learn AI with some of your courses as well cos it’s easy to understand
2026-09-03 12:47:58
0
War Badger :
You started out correct if reductionist, then sorta showed a lack of understanding what model training is, how inference works, and the difference between short/medium/long horizon tensors in predicting coherence. Saying it puts tokens in a sequence is sorta like saying your internal stream of thought in your meat brain is just a word processor. I mean. Yes. It is. But your brain is doing a lot more than that. And you can argue the large bodies of cells attached to your prefrontal cortex are cheating, or that your fingers are unintelligent, but from a systems engineering standpoint the difference is mostly irrelevant. It’s a wildly nascent technology. The underlying capabilities are unreliable. Alignment is nearly witchcraft and different capability optimization in post training makes different models useful for different things and unsafe in others. Naked LLMs are probably a dead end in the same way vacuum tubes were a dead end. The way it’s being marketed as AGI is somewhere between negligent and malfeasance. But the number of cognitive tasks it can perform under a good manage with domain expertise to a centroid standard is astonishing and matches any plausible definition of Artificial Intelligence proposed by any philosopher who coined the words. Is it computing 2+2? No it’s retrieving the answer from a training data set which is very much a Wikipedia of sorts structured in a weird vector graph that makes it sorta unreliable for traditional search retrieval. So are you by the way, this is how you know what 2+2 is without using fingers. Can a SOTA model chain-of-thought sequence derive novel things when supervised by humans? Some of the most incredible recent scientific breakthroughs say yes. Is it human cognition? (What I think you’re trying to argue). No. It has nearly nothing in common with human cognition. And that is a very valid argument that you are making very poorly, I think.
2026-08-28 22:03:21
46
tim :
You’re wrong. It is actually reasoning.
2026-08-28 22:24:00
12
disin_23 :
It tries to act smart building excessive layers* and gates* choking up pipelines creating endless patches for debugging. You definitely have to inspect the blueprint before you start to build your home project
2026-09-02 16:29:20
0
President AI :
Machine Learning?
2026-09-03 08:21:13
0
coolest.dude.on.tiktok :
You couldn’t be more wrong. It’s amazing.
2026-08-29 04:46:08
17
POSTGO :
This hot take is a hot mess of nonsense.
2026-08-28 21:46:59
29
oppyeahyouknowme :
You have an outdated understanding of how LLMs work. You are describing a transformer from several years ago. Modern AIs use reasoning on top of a transformer.
2026-08-29 04:03:44
44
demondivawardrobeplayboy :
This is cognitive dissonance pay this man no mind
2026-08-29 06:14:46
8
justinlooker5 :
I've built out over 500k lines of code so far. it is like working with an ESL coding team. if you provide good acceptance criteria and put strong guards on it (and get hostile agents to iterate with it and improve) you can get fantastically good results. I have been a software engineering manager, and it is significantly less error prone than a normal dev. it is not perfect. it is not intelligent. but if you are doing the intellectual heavy lifting it can do the majority of the dev without issue.
2026-08-29 02:01:38
8
Daily Bandicoot :
It does a better job than most engineers I’ve hired and paid six figures.
2026-08-29 11:17:08
5
user9944592416880 :
This sounds smart, but it's so lazy. It's like saying steam can never power anything because it's just water.
2026-08-30 20:21:16
6
northruphumanfirst :
Agree brother. But it’s not gonna go away completely. The next gen will be Neuro symbolic and eventually there will be a system that uses these LLMs as a voice module. It is funny companies are finally interested in spending the money to build more advanced deterministic integrations. First time I saw open claw, I was furious because this room of hyped up ai builders was amazed at “AI”. Come on it is a traditional software wrapper making up for the crap AI produced. As for going public it is fraud. Retirement investment firms have to buy new stocks to comply with diversity. So they buy this junk then the company files bankruptcy as their leaders walk away with a massive payout and the workers who had retirement savings now are left holding the bag, or hot potato. We need to be talking about this more. I have been on this rant for a year. I pushed practical use content but I myself am starting to use AI less. I love thinking and designing for a good way to use this. What’s the rush to make us irrelevant? That’s the big question.
2026-08-28 19:45:57
6
Mikeeagle5000 :
I used to agree wholeheartedly with you, but after using it on a significant c++ project, I have become more and more impressed at how well it understands what I'm trying to do. It also produces error free code. You may say that it doesn't really understand, but it's doing a great job faking it!
2026-08-28 23:08:50
9
Bush Baby509 :
He is absolutely correct. Agentic LLMs, meaning models that call external tools or run code to fetch information, are still just pattern-matching language models using scaffolding. That's fundamentally different from AGI, which would mean general, autonomous reasoning across any domain without that scaffolding. So calling agentic tool use AGI is a real category error, a lot of serious researchers push back on that label for exactly this reason.
2026-08-30 23:24:37
5
CharlyV :
Only those of us who work extensively with LLMs know how false the “AI is already smarter than humans” narrative really is. I agree with you: they’re impressive tools with enormous potential, but they’re still like teenagers—capable of doing remarkable things, yet constantly needing guardrails to stop them from doing something stupid. And sometimes they do it anyway. Where I disagree is coding. Some companies may still need highly skilled senior engineers, but many junior and mid-level coding tasks are already being replaced, and that trend will accelerate. What LLMs still cannot do—and don’t seem close to doing—is what you mentioned: reason and interpret reality with the multidimensional depth of a human being. That still looks very far away.
2026-08-30 13:30:48
7
Jonatan Reiners :
An interesting though. Why do we think it's smart?
2026-08-30 18:50:19
0
John Haley :
No wonder I’m underwhelmed.
2026-08-30 15:28:22
0
DQ Blizzard of Oz 🎰 :
yea I was confused for a hot minute until I got my bearings.. its an LLM wrapped in Python.. I was like that's it??
2026-08-29 05:46:23
1
HarpsGCord :
It’s not
2026-08-29 15:41:41
0
Gumaryo :
Very wrong.
2026-08-30 13:22:49
0
To see more videos from user @johnelder.ai, please go to the Tikwm
homepage.
