Is OpenAI's latest o1 model that good?
Вставка
- Опубліковано 28 вер 2024
- ►Full article and references: www.louisbouch...
► Our new book Building LLMs for Production: amzn.to/4bqYU9b
►Twitter: / whats_ai
►My Newsletter (My AI updates and news clearly explained): louisbouchard....
►Join Our AI Discord: / discord
Extra Ressources:
OpenAI release blog: openai.com/ind...
OpenAI release blog 2: openai.com/ind...
OpenAI system card: openai.com/ind...
Nathan Lambert’s great article on it: www.interconne...
David Shapiro fun livestream testing it: • OpenAI Strawberry Live...
How to start in AI/ML - A Complete Guide:
►www.louisbouch...
Become a member of the UA-cam community, support my work and get a cool Discord role :
/ @whatsai
#gpt4o #o1 #openai
I'm underwhelmed by o1 announcement. You can get a pretty similar quality (and increased latency) with an agent that reflects on the answers and just uses GPT 4o. It seems they reached a plateau for raw GPT outputs and are switching focus to AI applications and agents.
This can't be true , all the ai hyped happened in 2 years, so you are simply saying that 2 years of improvements was all for ai and humanity can t increase the output quality more. Is it sound meaningfull ?
Exactly. It's hilarious people fall into this GPT-4 vs o1 comparison crap. What's next, comparing LLMs with text TF-IDF+SVM classifier?
My experience too. I thought my 4o output was better.
FYI I asked o1 for an answer to the trolley problem and it explained that it would pull the lever to 1 person in detail
Let's dive into the essential takeaways:
1. Explore OpenAI's o1 for tasks requiring deep analysis and complex problem-solving, such as writing code or solving intricate puzzles.
2. Utilize o1 for projects involving translation, cross-cultural communication, or working with diverse datasets, as it excels in multilingual tasks.
3. Choose o1 when accuracy and depth of understanding are paramount, especially for tasks where speed is not the primary concern.
4. Opt for faster models like GPT-4 mini for quick and straightforward tasks where speed is more important than intricate reasoning.
5. Learn more about OpenAI's o1 and its capabilities by exploring the provided resources: OpenAI's blog post, system card, David Shapiro's livestream, and Nathan Lambert's article.
UA-cam needs more heros like you ! 🤩
Can it solve the "Octopus in a Spacesuit" problem, something most humans can easily visualise, correctly, but which AIs struggle with? The animal must be entirely enclosed yet be able to move and see from the location where the eyes are. Sounds easy, and for the mind's eye it is! To be clear this is an image generation task.
Give me the question I’ll ask it
Agree with what u said about trade-off. However, it seems CoT built in during alignment has more benefits compared to CoT during inference
Agreed!