ChatGPT Taught me MATHS

[ddiamond84]

I’ve had Maths since I started getting into Eurorack and I haven’t understood it since getting into Eurorack. I watched the recommended videos and read that recommended guide that people share.

Tonight I asked ChatGPT to explain it to me in really simple terms, and it honestly did a great job and I finally get it.

I’m not saying this is a good thing, and those resources are fantastic. I just could not seem to grasp it using those, where ChatGPT really gave me simple useful analogies and patch examples.

I find AI really scary and exciting- I keep learning terrifying new problems with it, and think of incredible uses for it. I dunno, I just thought it was interesting and unexpected that it was so helpful for me finally understanding MATHS.

Edit: it’d be more accurate for me to say, people taught AI how to teach me to use MATHS. It really does take human labor and effort for it to work at all.

[kurrwa]

so show us at least!

[ddiamond84]

What I learned or what it came up with?

[Spaceman Jacques]

One thing I will always say whenever AI is brought up: Please take it with a grain of salt and not as some infallible gospel. This applies far beyond modular synthesis. AI can get things VERY wrong sometimes and it can be very dangerous because people often assume it's 100% correct all the time, likely because it's good at phrasing things in a way that sounds...competent. You won't notice until you start asking it things you already know about. But sure, it can be a great tool.

[ddiamond84]

Totally! So far it's been pretty easy for me to see when it's obviously wrong, but I always double check it for this reason! It's a good warning!

[d_squid]

One thing I will always say whenever AI is brought up: Please take it with a grain of salt and not as some infallible gospel. This applies far beyond modular synthesis. AI can get things VERY wrong sometimes and it can be very dangerous because people often assume it's 100% correct all the time, likely because it's good at phrasing things in a way that sounds...competent. You won't notice until you start asking it things you already know about. But sure, it can be a great tool.

I asked it to help me rewire one of my electric guitars and it was just straight up wrong about where to wire the caps, but you would never have guessed by how confident the instructions sounded. What you say is very true.

[beat4less]

What would be interesting is to tell what ChatGPT taught you that you didn’t find elsewhere

[etckla]

Maths is pretty well documented on the web, so a model will provide pretty reliable results. Without further context, most models will do a decent job of explaining something that is well documented.

And they will fabricate stuff that is not well documented, and - as people have noted - they will remain brazenly confident.

There are also some types of task they are obviously andsurprisingly awful at. Have you ever asked ChatGPT to plan a room layout? Awful. I think many problems reliant on understanding of spatial or physical details are beyond the current crop of LLMs (or at least their training data)

But ask it to code you a Python webapp to do transient extraction and it'll do it.

What I find more generally reliable as a workflow is when there's a complex module with a big manual or multiple manuals. I will load all the documentation into a ChatGPT Project, or Notebook LM and can then ask it questions and it tends to respond pretty reliably. I've done this with:
- Arbhar
- Assimil8or
- Various bits of 90s rack gear
- Teletype - I fed it manuals, github repos and a bunch of forum threads I'd exported
- OAM Uncertainty - I fed it the hardware spec so it'll code circuitpython for pretty much any idea I can think of

[friendlybleep]

Here's a nice couple of paragraphs that summarise the current situation perfectly.

And here's the text in the image for people who might have sight problems;

June 19, 2025 at 12:21 PM

Chatbots - LLMs - do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they're "right" it's because correct things are often written down, so those patterns are frequent. That's all.

When a chatbot gets something wrong, it's not because it made an error. It's because on that roll of the dice, it happened to string together a group of words that, when read by a human, represents something false. But it was working entirely as designed. It was supposed to make a sentence & it did.

[bakamonster]

NotebookLM is pretty damn good. Uploaded all my modular manuals to one project and now have a searchable knowledge base of anything in my system. The audio podcast it created from those sources was pretty funny though.

[dubonaire]

Here is something else to clarify the situation. Not all chatbots are LLMs. If there is one thing that characterizes public discourse on AI it’s ignorance. Notwithstanding that, humans don’t axiomatically know facts either. When humans are “right” it’s generally because they reflect the consensus, which happens to be the dominant pattern, or paradigm, which history tells us is more often than not superseded.

[etckla]

Here's a nice couple of paragraphs that summarise the current situation perfectly.




2025-11-16 09.59.11_bsky.app_cc5475805213.jpg


And here's the text in the image for people who might have sight problems;

June 19, 2025 at 12:21 PM

Chatbots - LLMs - do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they're "right" it's because correct things are often written down, so those patterns are frequent. That's all.

When a chatbot gets something wrong, it's not because it made an error. It's because on that roll of the dice, it happened to string together a group of words that, when read by a human, represents something false. But it was working entirely as designed. It was supposed to make a sentence & it did.

To take this thread even further from the Eurorack topic, I dont think it's fair to say that these systems aren't being designed to create accurate results. It's true there is a lot of "probabilistic black box" stuff going with the models, but also design and engineering work on creating systems that improve the truthiness and usefulness of the results.

Ultimately there are people that leverage the tech effectively, others that feel frustrated by the limitations of the tech and feel it's all a con, and others that decide they never want to go near them for whatever ethical/artistic/human reasons.

Anyway none of this is Eurorack specific so apologies for derailing.

[kurrwa]

What I learned or what it came up with?

either or?
i dont get the point of this thread otherwise

[Electro Something]

Loopop’s incredible tutorial video taught me Maths.

[khyber]

This is what I get back when I ask ChatGPT (whatever the free model is): "explain the eurorack module maths by make noise to me".

Not super enlightening, a chatbot can only repeat what it's read in the dataset, not rationalize it or break it down in new terms.

**Make Noise Maths** is one of the most famous and versatile Eurorack modules. People often call it a *Swiss-army knife* because it performs many different jobs depending on how you patch it. Here’s a clear, musical explanation of what it does and how to think about it.

---

# What Maths *is*, in plain language

Maths is a **dual function generator**. Each of the two main channels (Channel 1 and Channel 4) can act as:

* **AD envelope generator**
* **LFO (cycling envelope)**
* **Slew limiter / portamento**
* **Envelope follower**
* **Envelope shaper**
* **Sub-audio modulation source**
* **Audio oscillator (in a pinch)**

Then the middle section mixes, inverts, and logic-combines signals from the sides—this is where the “Maths” happens.
---
# The panel broken down

### **1) Channels 1 & 4 — the big ones**

These are nearly identical, and each has:

* **Rise**: Attack / glide up / slew up time
* **Fall**: Decay / glide down / slew down time
* **Both**: Link rise + fall (when the switch is in the middle)
* **Cycle button**: Turns the channel into a looping envelope → LFO
* **Signal input**: For slewing or processing external audio/CV
* **Trigger input**: Fires an AD envelope
* **Unipolar output (positive)**
* **Bipolar output (positive/negative)**

**What they “feel like”:**
Think of each channel as a very flexible envelope that can also function like a hand-drawn motion curve for any signal you feed it.
---
### **2) Channels 2 & 3 — attenuators/offsets/Mixers**

These middle channels look simpler but are extremely useful. Each one can act as:

* **Attenuator** (shrinks a signal)
* **Attenuverter** (shrinks *and* inverts a signal)
* **DC offset generator** (if no input is plugged in)
* **Mixer input** (added to the SUM output)

These feed into the mixer section.
---
### **3) The "Maths" section (middle outputs)**

These produce composite signals:

* **Σ (SUM)** = channels 1 + 2 + 3 + 4
* **OR** (actually “analog OR”) = max of channels 1 & 4
* **INV** = inverted SUM

People often forget how powerful these are. SUM, OR, and INV let you perform real signal math: combinations, logic-like behavior, weird curves, etc.
---
# 🎛 Common Uses — the “real-life” cheatsheet

### **1. Envelope Generator**

Patch: *Trig → Ch1 Trig In*, then take *Ch1 Out → VCA CV*
Adjust rise/fall for attack/decay shapes.

### **2. LFO**

Press **Cycle** on Ch1 or Ch4.
Rise/Fall set the waveform shape and speed.

### **3. Slew Limiter (Portamento)**

Patch: *Pitch CV → Ch1 Input*, then *Ch1 Out → Oscillator CV*
Adjust rise/fall for glide time.

### **4. Envelope Follower**

Patch audio into Ch1 Input
Take envelope from Ch1 Out
Rise/Fall determine smoothing.

### **5. Trigger delay**

Trigger → Ch1 Trig
Take *EOR* or *EOF* as the delayed gate
Adjust rise/fall to change the delay.

### **6. Mixing, offsetting, inverting**

Channels 2 & 3 are perfect for:

* Fixing signals that are “too strong”
* Turning envelopes into ducking curves
* Mixing LFOs into complex modulation

### **7. Complex modulation curves**

Send Ch1 and Ch4 into SUM or OR.
You can create:

* Random-like modulation without randomness
* Bumpy envelopes
* Staggered motion curves
---
# A mental model that makes Maths easy

Think of Maths as **two “motion generators”** on the left and right, and a **math processor** in the middle.

* Left & right make shapes.
* Middle combines shapes.

That’s really it.
---
# Want specific patch examples?

If you tell me what kind of music or system you're using (drone, techno, ambient, generative, west-coast, etc.), I can give you **custom Maths patches** that fit your style.

For fun I then asked: "give me a techno patch example":

Here’s a **tight, driving techno patch** built around **Maths**. It uses Maths to create a punchy modulation engine for your kick, bassline, or percussion—no exotic modules required.

I’ll keep it clear and practical so you can patch it immediately.
---
# **Techno Patch: “Punchy Pumping Mod Engine”**

Maths is used for **sidechain-style ducking**, **percussion accents**, and **LFO groove**.
---
# **Goal**

* Strong, tight kick (external module)
* Bassline or stab that “ducks” around the kick
* Groovy LFO shaping filter movement
* Subtle accent modulation for hats/percussion
---
# **PATCHING STEPS**
---
## **1. Kick drum drives the whole patch**

**Kick trigger → Maths Ch1 TRIG**

* Set *Rise*: almost fully CCW
* Set *Fall*: around 10–11 o’clock

**Result:** a punchy decay envelope locked to the kick.

Use **Ch1 EOS (End Of Cycle)** as a sync pulse for sequenced elements or clocks if needed.
---
## **2. Make the bass or stab “duck” like a sidechain**

Use the envelope you just made on **Ch1**.

**Ch1 Output → VCA CV IN (for your bassline)**

* Ensure your bass goes into that VCA’s audio input.

Adjust the **Fall** to dial in the “pump” amount.

**Rise = how fast it snaps back**
**Fall = how deep/long the duck is**

This is the classic techno throb.
---
## **3. Groovy filter movement with Channel 4 LFO**

**Press CYCLE** on **Ch4** to turn it into an LFO.

* *Rise*: ~1 o’clock
* *Fall*: ~2–3 o’clock
→ This gives a skewed triangle-like wobble.

**Ch4 Output → Filter cutoff CV**

Now increase or decrease Rise/Fall to shift the groove between forward/lagging feel.
---
## **4. Add shuffle-style motion using the Maths SUM**

Take advantage of channels 2 & 3.

### Step A:

Patch your sequencer’s **accent output → Maths Ch2 IN**
Set Ch2 knob to attenuvert until the accents give spikes in cutoff.

### Step B:

Ch4 (LFO) is already in the SUM as well.

### Step C:

**SUM output → Filter cutoff CV input (instead of Ch4 alone)**

Now the filter is modulated by:

* the Ch4 LFO
* micro-accents from Ch2
* whatever else you add

This creates **micro-timing feel**—similar to Roland shuffle.
---
## **5. Add bite to hats with the OR logic**

Take advantage of Maths’ analog logic.

**Connect:**

* Ch1 (kick envelope) → OR input
* Ch4 (LFO) → OR input

Take the **OR output → Hat VCA CV**

This gives your hats:

* subtle pumping from the kick
* fluctuating sharpness from the LFO

Perfect for rolling techno hats.
---
# 🎛 **What you should hear**

* Kick stays solid and central
* Bass ducks cleanly on each beat
* Filter breathes with groove accent modulation
* Hats get dynamic sharpness that moves with the mix

This creates that **rolling Berghain-style techno** modulation grid.
---
# Want it tuned to your rig?

Tell me what drum modules, oscillators, or filters you have, and I’ll build a **patch tailored to your exact case**.

[tomchurchill]

This is what I get back when I ask ChatGPT (whatever the free model is): "explain the eurorack module maths by make noise to me".

(snip)

And straight away there's a mistake, in the Channels 1 & 4 feature list:

* **Unipolar output (positive)**
* **Bipolar output (positive/negative)**

Maths doesn't have a dedicated bipolar output on channels 1 & 4, only a unipolar output (the Unity Signal Output).

(NB: Serge DUSG does have a bipolar output as well, which is maybe where ChatGPT got the idea that Maths would be likely to have it too...)

Honesty, there are so many great resources for learning Maths online, I don't know why anyone would resort to this stuff.

[beat4less]

I suspect this was a one night thing

[lfad]

I love ChatGPT and it can be a great resource for eurorack patches and ideas and information. I've had fun numerous times gabbing with it about eurorack modules, brands, and patches.

I definitely agree not to take anything it says as gospel though, it does get stuff wrong. But in general I find people who are super down on it (often with a weird undercurrent of hostility) kind of silly -- it's an impressive and fun tool for all kinds of purposes if you give it a chance. Just take it with a grain of salt and fact/double-check important stuff is all.

[kurrwa]

But in general I find people who are super down on it (often with a weird undercurrent of hostility) kind of silly -

this will age well ^^

[lfad]

this will age well ^^

Whatever that means.

[kurrwa]

this will age well ^^

Whatever that means.

"silly" really isnt the word i'd use

[Biom]

Honesty, there are so many great resources for learning Maths online, I don't know why anyone would resort to this stuff.

Laziness, perhaps? As a factor, that drives technological advance too often these days.

The problem with ChatGPT, as I see it, is that you always need to check if the results aren't wrong, which means studying things yourself beforehead. Thus working with ChatGPT sometimes becomes the double work, mostly absurd, as it is in our case with Maths.
The more we know about things we're asking, the less ecstatic our ChatGPT impression is, I believe.

[Uncle Meatball]

But in general I find people who are super down on it (often with a weird undercurrent of hostility) kind of silly -

this will age well ^^