what’s so exciting about o1 and why are people going crazy with o1 when GPT 4 was already doing the same job?
In: Technology
From ChatGPT themselves
“We’ve developed a new series of AI models designed to spend more time thinking before they respond. They can reason through complex tasks and solve harder problems than previous models in science, coding, and math.”
Source: https://openai.com/index/introducing-openai-o1-preview/
The main difference between chatGPT-4 and chatGPT o1 is in how they think through questions.
chatGPT-4 just reads your input and gives you an answer straight up based on that. Think of asking someone a complicated math problem but only giving them 5 seconds to answer. They will make a guess, but they won’t really go through the problem and most answers are wrong.
o1 uses something called chain of thought reasoning. Instead of jumping to the final answer, it kind of “talks to itself” before and breaks the problem into smaller chunks, step by step, before giving the final answer. So, if you asked the same math problem it would go through the problem and ask questions to itself like “what kind of problem is this”, “What could be possible approaches to solve this”, “Which of the approaches seems most promising”, “What is the first step towards the solution” etc.. And only from these intermediate outputs it generates a final answer.
This makes it way superior for tough problems like math, logic, programming or puzzles but also means it’s slower and needs more resources to think things through. o1 also lacks a few features from chatgpt-4 like the ability to process images and audio.
And personally I also find that chatgpt-4 is a bit more eloquent, producing text that is nicer to read when it is supposed to just summarize something etc.
It’s not doing the same job. I use ChatGPT to write scripts. In fact, I’m using it right now. ChatGPT 4 can write some pretty decent code, but it will almost never work out of the box, unless it’s very simple. And the troubleshooting sometimes doesn’t get you anywhere. ChatGPT 4o is smarter, and can do better work. The scripts sometimes work right away, and with enough effort, can often be fixed if they don’t. The only exception being scripts that are too complicated — and let’s face it, the sky’s the limit when you’re only thinking of a project, and you don’t have to write it out yourself. So it gets a lot of problems it stumbles on. Now we have ChatGPT o1-mini and ChatGPT o1-preview. Mini does a great job, and usually gets it right. But if it doesn’t, you have to turn to o1-preview. That’s the one that really takes its time to think about things, and very often its code works first time. And we’re talking complicated code I couldn’t hope to understand. Reams of it. o1-mini wrote me a media player the other day. A webpage that populates a playlist based on a server folder of audio files, and allows you [to click] on the list and play the file. There are also the normal media player controls. And it works. How can you beat that for twenty bucks a month?
Latest Answers