re: I use AI like grok/chatptg/gemini etc often and its bad how often its flat out wrong.

I’ve tried using it to solve a problem several times and it told me the steps. I said “that’s what I did, it didn’t work” Then it would respond, “oh it’s because of xyz try this”. Which was completely wrong and then the remedy was repeated. I said, “no it’s not because of xyz and you told me the same steps again and it doesn’t fricking work.” And it just can’t figure out that I’m saying their suggestions don’t work. More often than not lately I’ve just gotten frustrated. So many times I’ve had to correct it for basic facts and it’s like “oh you’re right! My mistake”

quote:

---

am less bullish on it now than I was when I first started using it months ago. Seems to be getting worse.

---

Because these things are not true artificial intelligence yet, but basically more efficient googles that scour the web to find what it thinks you’re looking for and predicts what the next sequence of 0’s and 1’s will be

The real question is why do you still use something often that doesn’t work?

quote:

---

As an example I googled how old am I and put my date of birth in and Gemini was off by 3 years. This is the most basic of math questions.

---

This probably stems from using a large language model to do math.

LLM’s aren’t the end-all, be-all of AI development. They aren’t good at math because they don’t actually calculate anything.. they try to solve using natural language which is very hit-or-miss. Gemini is actually better at math than most from what I understand but.. yeh.

The real promise of LLM’s is that they can serve as a human-machine interface between the user and a plethora of software tools (both AI and non-AI). For a math problem, that might mean the LLM takes the input, reformats it (internally) for output into other tools, feeds the information to the other tools, collects the outputs, summarizes the final “answer,” and then reports it back to the user. Those other tools could be a simple direct math calculation, or they could be other AI models specializing in more detailed analysis or even capable of writing and executing their own python script to solve the problem.

We are just beginning to scratch the surface of this kind of integration. GPT is multimodal (capable of handling text, images, and audio) since mid-2024. We are starting to see integration at the end-user software level via Copilot and other tools. It’s still clunky. But in (almost?) all cases, the LLM is the top-level interface because it excels where the rest fail - handling inputs from humans.

When you consider that most people had never used an LLM at all before 2ish years ago, and that ChatGPT launched less than 3 years ago, it really has been advancing incredibly fast. Hell.. GPT only gained access to real-time search results about a year ago.

I tried using it to provide team records over a certain period. It kept getting things wrong leaving out years,

quote:

---

Garbage in Garbage out.

---

A fourth grade teacher (not in IT) taught us that. I remember, it was taped to the ceiling of the classroom right over our heads. That was 1984.

I have alot of problems with my chatgpt. I'll have to prompt it over and and again to get what I'm asking

I'm in a grad school program where I have to read a ton of academic articles written 80-100 years ago. I trudge through them, make my own notes...and then ask ChatGPT to summarize the article in layman's terms in 600 words or less. If the article isn't publicly available, I will upload the PDF.

In this very specific example, it does a good job converting old academic language into something readily understandable.

quote:

---

This. With good prompts and direction for sourcing or providing some foundational facts, the premium versions do well to turn bland info into something more palatable and actionable.

---

Additionally, I make 1 conversation for each class. This helps in organizing things, but as the class goes on, it picks up on the things I ask and starts to connect the dots between all the articles I have uploaded or asked about.

We are going to become so reliant on AI, like our phones. But it's going to make people even more dumb. People will lose brain muscle because what would have been used for critical thinking, will stop being used since most will let AI do the heavy lifting.