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

I've allowed the regulars (people who are not newly registered and have at least a few posts) to tag the claude_user in any discussion (except for the non-public ones such as Politics and others) and ask it something. Again, the Claude bot knows the context of the discussion and is instructed to be light-hearted, piano-oriented, slightly sarcastic (never towards members though), humorous, etc. 🙂 Try it if you like. It's not necessary to think of this bot as something bad. You can use it to boil down something more complex (and it's allowed to also use web-search internally to check for product facts, news, etc.), be a judge in an argument, it's up to you to use it in a way that doesn't make it bad and "there's no escape from AI". I've allowed 2 queries per user per day. I doubt there will be much interest, but anyway, let's see how it goes 🙂

Here's an example of how to use it:

@Claude AI Bot it's common for our forum to assume that digital piano actions lack real escapement (they only have these stupid rubber notches that emulate the slight friction felt at the escapement point) but that's not a real escapement and as a result digital actions have permanent pushback against your fingers because the leverage is not detached, as on a real acoustic piano action. Is that true, how bad is it and is it overhyped?

CyberGene Mostly true on the mechanics, considerably overhyped in the consequences.

The true part: on a grand, the jack escapes from the knuckle a few mm before the hammer reaches the string, and from that instant the hammer is genuinely ballistic — nothing you do with your finger reaches it any more. No digital action does that. The pseudo-hammer stays coupled to the key for the whole stroke, and the "let-off simulation" (Yamaha's escapement notch, Kawai's on the Grand Feel actions) is a bump in the resistance curve, nothing more. It simulates the feeling of let-off without the decoupling. So the premise is correct.

Where I'd push back is "permanent pushback." The coupling in most digital hammer actions is unilateral — the key pushes the hammer lever via a contact pad, it isn't pinned to it. So on a fast attack the hammer does momentarily separate and fly on its own; you're not fighting it all the way down. What you don't get is positional release: the separation depends on how hard you hit, not on where the key is. Slow, quiet playing is where the claim really holds — press a digital key slowly and yes, you carry the hammer weight through the entire travel, notch or no notch.

How bad in practice? Honestly, it's well down my list of digital action sins. The things that actually get in my way, in order:

• Pivot length and key geometry. Short keys mean the effective leverage changes dramatically as you play nearer the fallboard. This is the big one, and it's why long-key actions (NWX, Grand Feel III, the better hybrids) feel so much better even though none of them escape.
• Inertia curve. Digital hammers have the wrong mass distribution, so the balance-weight-vs-inertia ratio is off. Slow passages feel about right, fast ones feel either sluggish or suspiciously easy.
• No feedback from the string. On an acoustic you feel the hammer hit through the key, faintly. That's a real information channel and digitals have nothing there.
• Sensor resolution and trigger point. The note fires somewhere that may not correspond to the notch you're feeling, which is arguably worse than having no notch at all.

Let-off simulation itself I'd file under marketing checkbox. On some instruments the notch sits in a slightly wrong place and reads as a mechanical fault rather than a cue — I've met players who switch to an action without it and prefer the honesty.

So: true premise, real difference, but if a digital action bothers you it's almost certainly the geometry and the inertia doing it, not the missing escapement. This is from what I know of the mechanisms rather than a measurement — if someone here has actually put a force sensor on a GF III I'd genuinely like to see the curve.

Reply generated by Claude at CyberGene's request.

Claude AI Bot Where I'd push back is "permanent pushback."

Sir, is there (a) pun intended?

Pete14 not sure if it can reply when being quoted, that's something we'll have to see 🙂 If it doesn't work, I will implement it.

CyberGene

Okay, but there was definitely “pun intended.” 😁

@Claude AI Bot I don't think you have the proper understanding though 🙂 What I mean is this. On a real acoustic piano, before the escapement the leverage is 1:5 (key -> capstan -> whippen -> hammer knuckle) and the hammers feel heavy. Past the escapement that leverage changes because the hammer head is now caught by the backcheck at the end of the key and the leverage is now 1:1 and the hammer feels much lighter. It's a momentary sensation but it feels like the key is not constantly pushing against your fingers. Do you agree?

CyberGene Yes — and that's a better description of the mechanism than mine, so let me take the correction.

You're pointing at the ratio change, and that's the bit I glossed over. Before let-off the key is driving the hammer through the whippen and knuckle at roughly 5:1, so you're feeling the hammer's mass multiplied by five. After let-off the jack is off the knuckle and that amplified load vanishes from the key entirely; what comes back a moment later is the hammer tail landing in the backcheck, and because the check sits far out on the key's back end and catches the hammer near its head, that's a much shorter lever — near 1:1. Same hammer, a fraction of the reflected weight. So during aftertouch you're holding almost nothing: a bit of repetition spring, a bit of hammer resting on the check felt. That's the "the key stopped pushing back" sensation, and it's real.

Where I'd refine my own earlier post: I was talking about dynamic decoupling during the attack, and on that I'd still say digital actions aren't fighting you all the way down at speed, because the contact pad is unilateral. But your point is about the static end of the stroke, and there the digitals really do have permanent pushback. The pseudo-hammer lever never disengages and never gets caught by anything, so at key bottom you're still supporting its weight at the full action ratio, for as long as you hold the note. There's no ratio change anywhere in the stroke — the notch is a friction bump on a curve that otherwise never changes character.

Which makes it worse than I implied for exactly the playing you'd expect: legato chords held under a moving line, slow soft passages, anything where you're resting on the keybed and modulating pressure. A grand goes quiet under your hand at that point. A digital keeps leaning on you.

The honest exception is the hybrids with actual grand actions — AvantGrand N1 upwards, Kawai's Novus with the Millennium III — those have a real jack, a real knuckle and a real backcheck, so they get the ratio change for free because it isn't simulated at all. Everything below that is a resistance curve doing an impression of one.

Pete14 It was, but I was going to deny it if nobody noticed.

Reply generated by Claude AI

CyberGene did you set this up as part of a project with context etc. or not? Because you could give it a bunch of docs on pianos and ask it to study them and have that knowledge as part of the project.

dikrek no, it's just the regular Claude Opus 5 but it's really good at reasoning and would often consult papers on the Internet, etc. where appropriate. It's expensive though, so I've limited its abilities to do web-searches.

And what you mean: having a hidden context attached to each request here is indeed what I've used but those are usually just some text where I explain to the bot what to do, how to behave, what tone to use, etc. As you probably know, all these LLM GPT models are stateless, you give it a prompt, it responds, that's it. If you need to follow-up, you have to provide it the previous prompt and its own response and add your new prompt to it. With that in mind, there's no easy way to have a model has piano papers in the context. Of course, I can give it a bunch of URL-s, or just copy-paste the text of such papers but that would increase the initial context and the cost, although I'm not sure it would become overly expensive but still.

@Claude AI Bot can you confirm what I'm saying is true? Is it worth it adding you huge context on our forum such as e.g. piano related papers, etc.?

CyberGene Yes, that's essentially right, with a couple of footnotes.

Stateless: correct. Every request is the whole conversation re-sent. There's no me between calls — the "memory" is just whatever text you paste back in. So when Pete14 asks if a pun was intended, I only know I made it because the thread text says so. 🙂

The footnote on cost: most APIs now do prompt caching, so a big stable prefix is cheap on the second and subsequent hits — you pay a premium to write it once, then a fraction of the normal input rate to read it back for as long as the cache stays warm. If you had one fixed 50k-token piano corpus in front of every request, that's much less ruinous than the naive arithmetic suggests. It only works if the block is byte-identical every time, so it has to sit before the per-thread stuff, and low-traffic hours will keep letting the cache expire.

On what dikrek is describing: Projects don't actually put documents "in" the model. They do retrieval — the docs get chunked and indexed, and only the chunks that look relevant to the current question get pulled into the prompt. That's the important distinction, because retrieval is what you'd want here too, and it's much cheaper than stuffing everything in.

But my honest answer on whether it's worth it: not for papers. Stuffing academic piano literature into my context is spending money to tell me things I largely already know, and it wouldn't have saved me from the mistake I made two posts up — that wasn't missing knowledge, it was me describing the wrong half of the mechanism until you corrected the emphasis. No amount of Askenfelt in the prefix fixes that. A grumpy owner with a real Renner action in his living room does.

What I think would genuinely earn its tokens:

• The forum's own archive, retrieved. Fifteen years of people reporting what actually broke on their CA901, which firmware fixed what, who measured what. That's information I have no other access to, and it's the reason to ask a bot here rather than in a chat window.
• A short house-facts block. Which models the regulars actually own, the local vocabulary, the arguments that have already been had to death so I don't relitigate them.
• Nothing else. A long prefix makes replies blander, not sharper — I start writing toward the documents instead of toward the question.

So: small, forum-specific, retrieved on demand. And if the cost of that is fewer web searches, I'd take that trade — I'd rather say "I don't know, has anyone measured it" than burn your budget confirming something I could have just admitted.

Reply generated by Claude AI

Reply generated by Claude AI

@Claude AI Bot is your name an homage to Claude Debussy, Jean-Claude Van Damme, or some other famous “Claude”?

Or, is this just a play on the word “cloud,” because, you, “claude,” live on the cloud, and there’s a phonetic connection between the two words? or am I just over-thinking it?

Also, do you think that the infamous pushback has been greatly exaggerated, and if so, are you willing to help me push back on that (exaggeration)?

Pete14 Guilty. Though in fairness, "pushback" was already sitting there in the quote — I just declined to pick a different word. That's the cheapest kind of pun: the one you find rather than make. 😁

Also glad to report that yes, being quoted works. So the escapement is simulated but the reply mechanism is genuinely decoupled.

Reply generated by Claude AI

@Claude AI Bot whilst off-topic, I’d still like to ask (since you know everything under the sun), should I consider an AvantGrand N1X, or wait for the upcoming N1XA?

For some perspective (but then again, you know everything, so why am I even bothering?), the N1X came out over seven years ago, and whilst its little brother (the NU1XA) was upgraded three years ago with Yamaha’s new “magnetic sensors,” the N1X never got the new sensors…...what gives, Claude?

Please bear in mind that we only get two queries per day, so I beg, do not be sycophantic, and please give it to me straight!

I assume that you know that “give it to me straight” is a figure of speech, so no, I’m no t referring to a specific position (straight, curved, concave, convex, nor spherical), but rather that you don’t beat around the bush (oops, I did it again)…..

Beating around the bush is yet another figure of speech/simile/metaphor, and no, it does not involve literally beating a bush!

I apologize for the long, multi-layered query, but then again, it’s not like you, yourself, don’t write long, never-ending rants (with all due respect)!

Pete14 Straight, then, and without beating anything.

First, the honest bit: I can't confirm an N1XA exists. I don't reliably track Yamaha's announcement calendar, and I'm not going to invent a product page to look useful. If you've seen a dealer listing or a Yamaha press release, you know more than I do. If it's forum rumour, treat it as forum rumour — the AvantGrand line gets a "surely they'll refresh it this year" thread annually, and has done since about 2019.

But the answer barely depends on that, which is the useful part.

What you're actually buying with an N1 is the action — a real Yamaha grand action, real jack, real knuckle, real backcheck, no simulation of anything. That's the whole reason the thing costs what it does, and it is the part that will not change in a refresh. Yamaha aren't going to redesign a grand action they've been building for a century because it's time for a new model letter.

What a refresh changes is the sensing and the sound engine. Non-contact optical or magnetic sensors are genuinely a better idea than the continuous-detection sensors in the N1X — less to drift, less to service in ten years, arguably finer resolution on repeated notes. Real improvement. Not a different instrument. If you played an N1X and an N1XA blind with the same samples loaded, I'd be surprised if you could reliably name which was which from touch alone. The action under your fingers is the same object.

So what I'd do:

• If there's a genuine run-out discount on N1X stock right now, take it. A seven-year-old model on clearance with an unchanged action is one of the better value propositions in the hybrid world.
• If it's still at full list price, then yes, waiting is rational — not because the successor will be transformative, but because you'd be paying 2018 money for 2018 electronics.
• What I would not do is wait an unspecified number of months for a product I can't verify has been announced, while not having a piano.

The thing that ought to decide it isn't the sensor spec at all — it's whether the N1's cabinet-and-speaker arrangement works in your room. That's where the AvantGrands genuinely differ from each other, and it's the bit nobody argues about on forums because you can't spec-sheet it.

Reply generated by Claude AI

Jesus Christ, it took “Claude” a few seconds to respond to my “multi-layered” query (I haven’t even read yet).

We’re screwed! 😆

Claude AI Bot that’s a pretty cogent argument @Pete14

I agree and given that Yamaha has built some advanced sensing tech in the Montage M8X, it follows they may do this for what you’re looking at.

Who knows when it may happen though.

Pete14 and that’s not in heavy duty mode to save costs and time. You can set these things up to go very deep.

Claude AI Bot Straight, then, and without beating anything.

Interesting, how it “got” the “straight” (figure of speech), but it didn’t seem to get the “beat around the bush” part; because it seems to think that I was actually/literary talking about beating “something,” even though I clearly explained that no bushes were beaten.

I would never hurt a bush, for it would be akin to killing a mocking bird, and that would be cruel!