ChatGPT advice to build a new PC and smartphone

FabiFr

1. Desktop PC Build (< 500 EUR) - Focused on Disabling Temporal Dithering

Selection Criteria:

- CPU without integrated GPU -> forces use of dedicated GPU for driver control. - AMD GPU with open-source driver support (temporal dithering control).

- Budget under 500 EUR.

- Monitor without PWM or DC dimming.

Recommended Build:

- CPU: AMD Ryzen 5 5500 (~100 EUR)

- Motherboard: MSI B550M PRO-VDH (~80 EUR)

- RAM: 16 GB DDR4 3200 MHz (~50 EUR)

- GPU: AMD Radeon RX 6400 (~150 EUR)

- Storage: 500 GB NVMe SSD (~40 EUR)

- PSU: Be Quiet! System Power 9 500W (~50 EUR) - Case: Fractal Design Focus G Mini (~50 EUR) Total: ~470 EUR

Recommended Monitor:

- BenQ GW2480T (~150 EUR)

- Flicker-Free (no PWM), Low Blue Light, IPS, ergonomic stand

Why This Setup?

AMD open-source drivers (amdgpu) + CoreCtrl allow you to disable temporal dithering. Dedicated GPU avoids locked integrated GPU behaviors.

Balanced cost, upgradability, and visual stability.

2. Fairphone 4 - Best OS Choice for Disabling Temporal Dithering

Why Fairphone 4?

- Modular, repairable, IPS LCD (no OLED PWM). - Unlockable bootloader -> install custom ROMs. - Community ROMs: LineageOS, /e/OS.

Recommended OS Options:

- LineageOS: Open-source, privacy-focused, rootable.

- /e/OS: Google-free Android, privacy tools, user-friendly. - Stock OS: Ethical Android, fewer controls.

Justification:

LineageOS and /e/OS allow root access, minimal visual effects, possible kernel customization. Google-heavy ROMs often add unavoidable dynamic effects.

languidicity

The monitor uses FRC according to displayspecifications.com, AMD GPU's have been shown elsewhere on this forum to use hardwired dithering that cannot be disabled using software and there is no assurance that the Fairphone 4 doesn't use hardware or software dithering.

Oh, and 'AI' is a misnomer.

JTL

languidicity Thank you for illustrating the point that asking ChatGPT (and other "AI" agents) for advice in the direction is almost certainly leading a path astray, sometimes without realizing it.

randomboolean

AI has been a huge help for me as a research partner. I mostly use it in DeepResearch mode across 3–4 different tools to search this forum and others. It helps me find threads, understand complex topics, and explore different angles much faster.

I’ve used it a lot to find true 6-bit and 8-bit panels – though that’s still tricky. You usually need to find the datasheet for each panel and check it yourself. More often than not, “8-bit” panels turn out to be 6-bit + FRC. But with AI, I was able to find a list of possible 8-bit panels in record time, which would’ve taken ages manually.

If you use AI properly, it can speed up your research a lot. But in the end, you still need to test things yourself. There are no guarantees, and it always comes down to trial and error.

Its conclusions are hit or miss – maybe 50/50. It’s great for coming up with new ideas to explore, but I wouldn’t trust its recommendations blindly.

Compared to how I used to research this stuff a couple of years ago, using AI now feels like a night-and-day improvement. I can find more options to test, and it even links topics across different forums or brings up ideas that haven’t been mentioned here yet.

JTL

randomboolean AI has been a huge help for me as a research partner. I mostly use it in DeepResearch mode across 3–4 different tools to search this forum and others. It helps me find threads, understand complex topics, and explore different angles much faster.

I’ve used it a lot to find true 6-bit and 8-bit panels – though that’s still tricky. You usually need to find the datasheet for each panel and check it yourself. More often than not, “8-bit” panels turn out to be 6-bit + FRC. But with AI, I was able to find a list of possible 8-bit panels in record time, which would’ve taken ages manually.

If you use AI properly, it can speed up your research a lot. But in the end, you still need to test things yourself. There are no guarantees, and it always comes down to trial and error.

I wasn't saying it was always a disservice and it certainly can give ideas on where to research next. My criticism was more to the fact certain people treat these AI agents as outright source of truth with no external validation of their outputs.

randomboolean

JTL Agreed! I’ve noticed the same even outside the eye strain space. But with this topic, everything is dialled up to the extreme. Tiny details make a massive difference, and unfortunately AI still tends to hallucinate or gloss over them, even with solid sources. I don’t mind walking the last mile manually though – especially if it saves a few hours of digging.