How did Chat GPT4 pass all those exams? So linked is a copy of the discussion I had with ChatGPT and what was shocking was the mistakes - it got a simple 3.16^7 wrong. it was spewing out the wrong values.
So as you can see I'm using ChatGPT and GPT4 to help me write the new books. Particularly asking it to experiment with rewriting the formula in words (like using bullets in a different structure or parenthesis), as well as checking if it understood my formula.
The thing is, I don't have anyone to read my writings to tell me if they understood it (other than Nicco and he's working on improving our Art throughput as we integrate Procreate in our workflow) - if my math formula is wrong. I used to have mentors for that, but they were kicked out of the company for being argumentative (lolz). I am an IT project manager relying on soft EQ skills to sense motive and align goals - that's a good chunk of my work - doing mathematical Analysis is unfortunately not part of my work and when it does become part of it - No one can correct me because no one has a clue. Its frustrating when your love for science makes you the WEIRDO in the large social circle at work - and its in your hobby and friends, you can find other weirdos who can correct you (and tell you stuff you don't know).
So when ChatGPT came along and GPT4, despite being NERFED by MS to make their BING search engine more powerful (By having access to the web, unlike GPT4 that can only answer what information you put in it) it reduced my Excuses and by mental BLOCKs. my mental blocks being - I don't have the mental bandwidth to CHALLENGE my own math and go through where I can get it wrong.
We had the 9 coins puzzle in Pathfinder beginner box set, and with a generous clue from the GM we were able to solve it, but I love being able to solve this puzzle. Typically I have 4 other friends who can check my math, and I expect it only in games - otherwise i expect to pay for that service (typically our Indian supplier is really good helping us with these kinds of logic puzzles in implementation).
So I'm writing Hard Scifi Stuff and I'm a WIS spec'ed gamer focused on diplomacy and sense motive because people make up an organization and convincing and demonstrating benefits from a particular course of action. I dream about having a good INT and compensating using note and documentation, and recordings of myself relearning everything over and over again.
So you can imagine my frustration when MS nerfed the 20usd/mo GPT4 (allowing only 25 responses every 3 hours) and killed the open AI nature of it (a preview if they ever got their hands on Free Software Foundation IP, which allows the IP owner to revert the IP to closed source). Its weaker than the BING GPT4 and I'm powerless to complain about it. (Standfords Llama and Alpaca are my hope now; as if Migrating to nextcloud was already a pain).
It doesn't inspire confidence that MS lobotomized GPT4 so I'm not working with the AI that should be able to test if my explanation is correct. I'm not working with an AI that can pass all those College level tests. I have Grammarly correcting my grammar, and GPT4 was supposed to help me if someone CAN follow my logic and tell me if its wrong. With IT and high uncertainty jobs everything is A theory that needs to be tested and we always have to push the envelope - make this run cheaper, cut costs here and there, etc... and there are the tons of complacent giant corporations that hope you don't figure out the hungrier scrappier competitor is giving better value - as your clients keep expecting more for less.
So the use of GPT4 is limited to the 2000-word copy-paste cap, and anything else it Hallucinates. So it is mostly helping with Problem Statements, at work Writing procedures, and converting minutes to Actionable Statements. For Writing - testing the problem statement and giving me Leads, and opportunities to check regarding my understanding. I make it rewrite stuff, well this version of GPT4, is probably different, It used to be able to do Nuance but not anymore. so the technical mechanics of writing can be tested with Chatgpt - spitting out what it thinks is a more coherent explanation of my thoughts. When I write I have missing punctuation, articles, and words I thought I wrote but didn't (as I imagine my mental FPS is winding down as aging is telling my body to lower my metabolism and lower the process cycles my thoughts operate in). A dumber ChatGPT means I have to compensate - some ideas I cannot rely on it. (If I'm going to correct the AI I'm going to correct an AI that will be open source)
Growing up my problem with math was I took a lot of patience from my teachers as they explained the logic of the math problem. Now with GPT4 being in khan academy and in chatgpt I can ask it to make more examples and different demonstrations of some key concepts... IF I can trust it.
No comments:
Post a Comment