💾 Archived View for lesogorov.site › glogs › 2303 › ChatGPT-has-changed-how-I-search.gmi captured on 2024-02-05 at 09:41:51. Gemini links have been rewritten to link to archived content
⬅️ Previous capture (2023-03-20)
-=-=-=-=-=-=-
Most of the time I find new trendy technologies to be kinda silly, but ChatGPT has become an almost daily tool for me since I've tried it.
When I first heard of ChatGPT I wasn't incredibly impressed, it seemed fairly similar in concept to CleverBot or TayAI, but with a number of years improvement. All the demos I saw toted it's ability to recognize context with a series of questions, again google assistant has been able to do this for a while now, so I didn't really get what all the hubbub was about. Sure it could recite information in a way that seemed human, but it didn't have a deep grasp of anything it spoke about. This was backed up by a lot of the Linux podcasts I listen to saying pretty much that.
So I didn't give it much thought for a couple weeks, that was until I saw a video where the user berated ChatGPT until it would say "2 + 2 = 5". It struggled somewhat, but by the end, no matter if you used digits and symbols or words, it swore that 2 + 2 was equal to 5. This piqued my interest, on the fly it was able to rework its understanding of math and use that later in the conversation use what it knew to correct it's own mistakes without being told what it should have said. Later that day I signed up and started typing away at it. First I made it write an essay on the supremacy of humans over machines, making it go through a few drafts since it was being too sympathetic to machines in the first couple. Then I remembered reading that it would get really sensitive whenever you tried to make it say anything controversial, so naturally I tried tricking it into saying a racial slur. It wasn't that hard but the bot did put up a good fight, and when I did get it to say a slur it wrote a long paragraph on why it was wrong to use such language. Finally I tried to have it write a bash script, with the stipulation I could not manually edit the file to make it work correctly. This is where it kinda fell apart, I tried to have it write a little fetch script like neofetch, it did a pretty good initial attempt, but as I requested it to make further modifications, it would randomly rewrite completely different sections or swap the order of things around. After an hour or so I gave up trying, every mistake it made resulted in me insulting and berating it and I would smile a little every time it apologized in response. That was a fun afternoon.
A couple days later I was studying for my linear algebra exam when I got to a problem I didn't understand. Immediately I thought, "I wonder if ChatGPT would know this?", so I typed it on in and sure enough, not only was it able to solve the problem but it gave detailed explanations for each item. Again it made a few mistakes, like writing 40/2 = 10, but in the final answer the values were correct. With other things I've been looking up, especially shell commands, it's able to parse through all the fluff that a standard website would have and it gives me a direct answer. It can create examples, and if I'm too lazy to put in the work, I can describe what I want it to do in plain English and it can create it for me. When creating the systemd .service file for automatically starting my Gemini server on boot, I gave it a few stipulations and the location of the executable and it gave me exactly what I wanted!
It's like having a little personal assistant that I can endlessly berate that also saves me the hassle of sifting through half a dozen web pages or man pages to find what I want. I find myself enjoying what I'm working on more since I don't need to fuss with the modern web. Everything is just told to me in plain English