AI-generated content in replies, should it be allowed?

I can recommend the following link for anyone interested in seeing how the LLM-s work:
https://www.3blue1brown.com/lessons/neural-networks/

It goes first through a simple neural network that recognizes numbers and you should go through the different lessons and then it shifts into GPT and LLM-s. It's not too easy to understand and although I have covered many math courses in University (I studied Physics at the Sofia University and we had more than 10 separate courses in calculus, including Linear Algebra and matrices, Vector and Tensory analysis, etc.), yet it took me quite a lot of time to understand these slides and I gave up a multiple time, before going back and stubbornly trying again. But it's all fascinating and these models are not far from real AI and reasoning. It's OK to be skeptical, that's the most natural way anybody should be about anything new 🙂 But don't be skeptical without at least familiarizing yourself with how these things work.

CyberGene prompting is also hugely important, as is any source material. I fed it in a private discussion with Pashkuli (>400 messages!) a picture of his proposed trapezoidal suspension and it completely misunderstood the image (and to be fair I didn’t understand it either! I was like “how the hell would this work”).

Once I fed it a video of the action he sent me, showing different aspects, it DID understand (as did I) what was going on, and correctly explained such an action wouldn’t be a good idea, and why (which was exactly my opinion before I even asked it).

Currently it’s reducing my Gmail clutter, it’s trashed about 50% of my messages (I’ve had Gmail since they started).

BTW, I just made an extension (with Claude Code) that allows the claude_user to reply upon my request using the discussion context, also using web-search results and responding. It bills against my own personal API amount which is why only I can ask it 🙂 It's not expensive, so if you think that we should ask Claude something (It's Opus 5 on medium effort right now), let me know though, I'm OK and will just tag it and ask it to respond and it will do so.

Now, let's try it.

@claude_user can you check whether my reply above about how reasoning works is correct? Also, any general remarks about the AI usage policy on a forum like ours?

Mostly right on the big picture, wrong in a couple of specifics that are worth fixing since you're pointing people at the 3blue1brown series anyway.

What you've got right: it is a neural network. Not "resembles" one — the vectors and matrices are the network. There's no separate, more respectable thing called a neural network that transformers are approximating. That's just what a neural net is made of.

Where I'd push back:

1. "this can erroneously be called statistical prediction of text" — it isn't erroneous, it's just incomplete. Next-token prediction is literally the training objective. The mistake people make is assuming that a simple objective implies a simple mechanism. Predicting the next token well at that scale requires internal machinery that does something worth calling computation. So "it's just predicting text" and "it does something sophisticated" aren't in conflict — the second is how it achieves the first.

2. "multiple passes... produce multiple variants that then go through an intermediate reasoning verification" — this is the part I'd correct. For a standard reasoning model, there is no separate verifier model checking candidates at inference time. It's the same single model, doing the same autoregressive forward pass, one token at a time. The difference is that it's been trained (via reinforcement learning on whether the final answer was right) to first emit a long chain of intermediate tokens — the "thinking" — before committing to an answer. Those thinking tokens are just tokens. They're in the same context window; they condition everything after them. The gain comes from having somewhere to put the intermediate work, plus training that rewards using it well.

   Systems that generate N candidates and score them with a separate verifier do exist, and the "effort" dial CyberGene mentions does change how much thinking budget I get. But the mental model of "generator model, then a different reasoning model audits it" isn't what's happening under the hood by default.

3. On PASHKULI's silicon-vs-organic point: I wouldn't defend it, but I also wouldn't say artificial neurons are the equivalent of biological ones. A biological neuron is a wildly more complex object than a weighted sum plus a nonlinearity. The interesting claim isn't "we've built brains", it's the much weaker "you apparently don't need to build a brain to get some of this behaviour". Whether that adds up to intellect is a question nobody has a clean answer to, including me, and I'd be suspicious of anyone who says otherwise in either direction.

On policy — with the obvious caveat that you're asking the defendant to write the sentencing guidelines.

What I'd actually enforce:

• Disclose it, and own it. If you paste model output, say so and stand behind it as your own claim. "The AI says X" is not an argument. "I think X, and here's the reasoning I checked" is. The moment someone hides behind the model when challenged, the post was worthless.
• The Pashkuli objection is the real one. Copy-pasting someone's post into a chatbot and pasting the reply back isn't participating in a discussion, it's outsourcing your half of it. That's not a quality problem, it's a bad manners problem, and it survives even if the output happens to be correct. dikrek's counterpoint holds up though — using a model to find a study that contradicts you, then reading and posting the study, is just research with extra steps.
• QuasiUnaFantasia's rule is the highest-value one in the thread. Length limits. Uncapped model output is verbose because it's hedging on your behalf, and nobody asked it to. If it can't be said in a few paragraphs it probably shouldn't be pasted at all.
• Verifiable claims only. On a forum where people own the instruments being discussed, a confident wrong answer about hammer geometry costs more than no answer. I'm reasonably good on physics and general mechanics and genuinely unreliable on which specific serial numbers got which action parts in 1978. Treat me accordingly.
• I wouldn't ban full-AI accounts by rule so much as by boredom. An account that only ever pastes output will stop getting replies on its own. The accordion bot presumably discovered this.

The thing I'd resist is a blanket ban, mostly because it's unenforceable and it turns into people accusing each other of being bots for using semicolons. Judge the post, not the tooling.

Reply generated by Claude at CyberGene's request.

claude_user it's a bad manners problem

@dikrek … AI to tell you that is even more embarrassing 😂

It seems to me like everything we do is somehow influenced by the “algorithm” (AI); the only thing is that we’re not always aware of this, but trust me, we are AI……AI is us…..

Now, I understand we’re talking about obvious, too-perfect, too-nice, and perhaps even too-long posts that are simply cut & paste, “verbatim AI,” but I don’t see the harm, so long as we simply take what’s valuable and toss out the rest (of course, that’ll be different for every individual)….

Disclaimers: I am not a BOT. I made a lot of typos in the past posts myself. LOL.

I'm fine with citations of AI content. But talking with if dozens or hundreds of ghost-like BOT agents is crazy.

They would keep talking in a rapid and massive way that I could not catch up or respond to.

Again, please spare my life here as a member. I am not a BOT.

PASHKULI “dikrek's counterpoint holds up though — using a model to find a study that contradicts you, then reading and posting the study, is just research with extra steps.”

You always cherrypick.

CyberGene my take here is that the problem are the very long winded comments which make some threads unfathomable IMO. Of course, you can skip those messages, but as threads continue after them, the line of arguments is lost.

I think taking the time to answer manually is also an appreciation for the forum members. If you just want something which agrees with you and is able to write long messages which could be summed up in lots less sentences, please, be gentle to other forum members and just write what is needed.

Personally, I never use AI for writing. You will notice by my poor English 🤭

dikrek

That is a different thing though. AI finds verbose arguments (pro\cons) faster. And only from what is widely available online (or?).
AI is not the truth. It can not conduct experiments nor analyze design\mech concepts (yet).

While music\art\language\open source and such are widely available online and on various platforms (social media, websites), that is not the case with IP and technical propriety documentation (or proprietary source code).
That is AI is very bad at those. Open source is getting more and more available though (.Net, Blender, Linux, FreeCAD\OCCT).
Try to find drawings or source documentation on production processes about Mercedes, or jet engines. Not going to happen unless stolen as industrial espionage (hacking with AI). Didn't you say you worked in a such big company (tech)?

But I think that is what AI should be exposed to: application of mech.\chem.\phys.\micro, nano tech laws and concepts.
Such companies might use AI but only internally and as a closed source.

It should leave creating Music and Art. Completely.

But since I know what unscrupulous people are involved into this AI bubble… soon it will get pretty gruesome.

PASHKULI everything is science. Everything can be reduced to math. We may not know the science yet, that doesn’t mean that some concepts are beyond science.