AI again

I worked with Gemini 2.5 Pro over the past year, and the experience was mixed. It often helped me generate code (software), but at times its reasoning was opaque, and the results frequently didn’t run correctly until after many iterations. Yesterday the Gemini 3 Pro Preview was released, and I’ve been testing it for two days now. My first impression is that the solutions are noticeably better and that it makes fewer faulty inferences.

At some point, computers will be giving themselves development tasks and checking each other’s work to ensure everything runs smoothly. We’ll be standing on the outside, able only to watch in amazement. It’s a chilling thought, but it no longer gives me nightmares. Still, it remains thrilling — with a bitter aftertaste.

---

I'm down the ComfyUI rabbit hole, it's absolutely amazing what you can do even with open source. Workin' on sales pitch to Santa for a $6K computer upgrade.

---

You want run your own LLM? I have no phantasy what could make a computer so expensive, beside a high end graphics card for heavy compute tasks.

---

Just doing AI graphics/video and you're right, the video card RTX5090-32GB is $3K alone. It requires big case & 1000w power supply so gotta build whole system around it. Living in US with the tariffs & 'chip shortage' probably not helping.

Tried one of those online subscription services that gives 'generation credits' per month & burned up monthly allotment in couple days.

---

:(
program AfficheHallucination;

{$APPTYPE CONSOLE}

uses
SysUtils;

var
i: Integer;
begin
for i := 1 to 100 do
Writeln('hallucination!');
end.

---