Recent content by Syntactic

I spent a long time trying to chase down a Canadian source, and the chain ended with a couple of people at ID Group, who were apparently the regional distributors for melamine products made by Polymer Technologies Inc., which may or may not be unbranded BASF Basotect. They were supposed to get...

I do not know how to tell you in any other words that you cannot use LLMs to do this kind of work. 20 * (sqrt(5.1 / 0.31) - 1) is 61.12, not 34.93.

GPT is trying to play improv and has been trained on enough data to seem like it knows what it's talking about. And sometimes it does! But it's a roll of the dice. It is fundamentally unreliable. And it is very, very, very bad at math.

Just for kicks, I went back to your original post in the other thread and asked GPT-4 the same question you posed. Your GPT-produced answer was: I asked GPT twice. First answer: Second answer: Three attempts at your question, and nearly completely different answers to each part of it...

I linked to them above. They were in your other thread, in that post you didn't reply to.

But it isn't good at calculations. It's tricking you into thinking it is. Proof is just a few posts above. This is from OpenAI themselves (their emphasis):

Do you think there is something special about speaker calculations that makes them more accurate than basic math calculations? What would you consider to be a reasonable proof of ChatGPT failing? Be specific and verifiable.

It's not even good at math! This post will be #82. The square root of 828282 to ten decimal places is 910.0999945060. Let's ask GPT-4. This is direct from the (paid) OpenAI API playground. It is not reliable. OpenAI will tell you it is not reliable. The evidence shows it is not reliable. It...

Are you just pretending that the complete and thorough debunking of ChatGPT's 'analysis' in your ChatGPT to calculate DIY speakers thread didn't happen?

My dude. GPT-4 hallucinates like crazy. Look. You're an engineer now:

If you can get a good Bluetooth connection in your home, AptX is great. It's effectively lossless. LDAC is also an excellent codec but it uses more battery and has higher latency. I would go wifi over SBC in a heartbeat, but if you have AptX or any of its newer variants then you're good.

Does anyone know why I would see dramatic comb filtering with the Neumann MA-1 room correction applied? No changes to monitors or mic between these two measurements, just toggling the correction checkbox on and off.

Listening to black metal and noise on the KH150s has been nothing short of revelatory, especially after calibration. The bass extension makes them excellent for EDM but I was not prepared for how much better they were for distortion-heavy audio. I'm coming from a pair of Dynaudio BM6A MKIIs...

In wall speakers covered by acoustically transparent art/photo prints held in picture frames.

If you want a subjective tool to tell you things that may or may not be true because it is trying to figure out the answer you want to hear, GPT is great. If you want a tool to assist in a "Science Review" then look elsewhere. The idea of using a LLM tool to generate R code to easily...