OpenAI nextgen GPT
đź“ť Note:
This article was generated usingllama 3.1:8bmodel with help of Ollama;
Thumbnail was generated using Flux Schnell model with help of ComfyUI;
The Future of Large Language Models: Introducing OpenAI's "o1" Series 🚀
As a tech enthusiast, I'm excited to share with you the latest development in the world of Large Language Models (LLMs). Yesterday, OpenAI, the company behind ChatGPT, released a new series of "Thinking" models called "o1". In this article, we'll delve into what these models can do and why they're a game-changer for the future of LLMs.
The Limitations of Traditional LLMs 🤔
Traditional LLMs work by generating the most probable next token based on the input. This straightforward principle allows them to provide decent results, but it's not without its limitations. For instance, when faced with complex questions or tasks, these models often struggle to provide accurate answers.
One classic example is the question "How many R letters in the word 'strawberry'?" 🤔 Traditional LLMs would typically answer 2, which is incorrect. To solve this type of question, more advanced techniques such as a "Chain of Thoughts" are required.
Introducing OpenAI's "o1" Series 🔥
The "o1" series models address these limitations by introducing "reasoning" tokens that are truncated from the output. This allows them to iterate over the answer multiple times until they provide a well-thought-out response. The result is more accurate and contextually relevant answers.
Two distinct models are available: o1-preview and o1-mini. If you're using the API, these models can be accessed as part of your workflow.
Pricing đź’¸
While the "o1" series is a significant upgrade over traditional LLMs, it comes at a price. The cost for using these models is higher than their predecessors:
- $15.00 per 1 million input tokens and $60.00 per 1 million output tokens for
o1-preview - $3.00 and $12.00 for
o1-minimodel
However, OpenAI plans to release the "o1" series for free ChatGPT users in the near future.
What's Next? đź”®
The "o1" series is not a replacement for traditional LLMs but rather an addition that solves more complex tasks. In the world of Wiregate, we firmly believe that such models will eventually replace all other simple models.
What are your thoughts on the future of Large Language Models? Share your opinions in the comments below! đź’¬
Want to learn more about this topic? 🔍
- Check out our blog for more tech-related news and articles (https://blog.wiregate.io)
- Visit OpenAI's website for more information on the "o1" series:
Stay tuned for more updates on this topic! 🚀
Prompt đź“ť
```
My name is Niki and you are watching `Breaking Brains` from September 13, 2024.
Yesterday OpenAI - the company behind ChatGPT - released a new series of “Thinking” models called “o1”.
Typical Large Language Model (LLM) works using the straightforward principle of generating the most probable next token based on the input.
Given the prompt of `Continue the phrase in one word: I was riding a`.
The most reasonable answer would be `a bicycle`.
It’s much more probable than answers like `a bomb` or `a snail`.
By looping through the same `input` + already generated `output` the model gives you the most probable answer.
This method is very simple and gives you decent results but the model doesn’t think about the answer.
It would give you wrong answers to more complex questions.
The most popular and tedious question which majority of models give you the wrong answer is `How many R letters in the word "strawberry"?`
Typical answer is `2` which is wrong.
In order to solve this question a lot of techniques are used to instruct the model to think about it.
Like a “Chain of Thoughts”.
`o1` series models introduce “reasoning” tokens that are truncated from the output but actually solve this problem by iterating over the answer over and over until it gives a well thought answer.
Let’s look at some examples:
The strawberry question.
After a few seconds the model gives the right answer which is 3.
Second example is very interesting.
Standard models always give you the wrong answer.
There are 5 people in the room. A is watching TV, B is playing table tennis, D is sleeping, E is eating. What does C do?
The answer is obvious for the human being.
Because playing table tennis requires two people.
As you can see gpt-4o-mini gives you the wrong answer of not understanding the context of the problem.
Meanwhile `o1-mini` gives a correct answer.
Available in two distinct models if you are using API as `o1-preview` and `o1-mini`.
Pricing!
The o1 series is much more expensive than gpt-4o models.
Costing a whopping $15.00 per 1 million input tokens and $60.00 per 1 million output tokens for the `o1-preview` and $3.00 and $12.00 for `o1-mini` model.
The good thing is that OpenAI plans to release the `o1` series for free ChatGPT users in near future.
`o1` is not a replacement for gpt-4o but rather an addition to solve more complex tasks.
`o1` series is the most logical and reasonable next step in LLM’s evolution that would give more accurate answers to user prompts.
Here at Wiregate we do think that such models would replace all other simple models in the near future.
And what do you think about the future of Large Language Models?
Let us know in the comments below.
By the way - check out our blog - we do post a lot of interesting tech stuff there - link in the description.
Like and Subscribe to be in the tech news loop and I will see you in the next one.
```
Given above script for the video write a LinkedIn article at least 500 words.
Include a lot of emoji's in the text to make it more rich.
You must write in Markdown syntax.
You must use `##` for marking the paragraphs.
You must use brackets and parenthesis for the links.
Add following link to the video we recorded for people who don't want to read in the beginning of the article: (https://img.youtube.com/vi/JDc77spm84U/maxresdefault.jpg)](https://www.youtube.com/watch?v=JDc77spm84U)
Link to the blog: https://blog.wiregate.io
You must include following links in the article:
- https://openai.com/index/introducing-openai-o1-preview/
- https://openai.com/api/pricing/
- https://platform.openai.com/docs/guides/reasoning