The OpenAI o1 model is truly astonishing. Before its release, what directions can we choose or what preparations can we make, besides spending money?
1 Like
Good question.
First, realize that while the o1 model is better at some tasks, it is often overkill for many. As such, one should learn what models generally produce acceptable results for various tasks, then use those models first with those tasks.
Second, just because a model is more expensive does not mean it is better than a cheaper model. For example, Sora is for video generation; one should not expect it to do calculus problems. Even slight changes in model names, such as GPT 4 and GPT-4o, might give different results. Do not generally assume what a model can do without reading the system card (GPT-4 System Card) and trying a few prompts.
Third, expect the models to degrade over time and then improve again. From a few years of using the various models and keeping a pulse of what users see reflected in this forum, it is what it is.
Fourth, just because it is new and exciting does it mean it is needed? Do not get caught up in being on the bleeding edge because you will miss something. I spend most of my time problem-solving and adding new tools to my so-called toolbox. If there is not a problem needing a new tool, then don’t go looking for it; in other words, don’t fix what is not broken. If there is a new tool and it looks like it can help you solve something better, then learn to use the tool, but do not turn a hammer into a screwdriver.
As with such questions, the list can continue, but this should put one in the proper frame of mind.
3 Likes
Thank you for your detailed and thoughtful response. Your points about selecting the right model for the task and avoiding unnecessary complexity resonate deeply with me.
I especially appreciate your fourth point about trying, discovering, and adding new tools to the toolbox. This perfectly aligns with how I see the pursuit of cutting-edge technology. Exploring new tools isn’t just about their novelty—it’s about broadening perspectives and unlocking creative solutions for real-world problems. The more tools we have, the more possibilities we can explore when tackling challenges.
With AI’s growing impact across industries, I find myself reflecting on how best to navigate my career as a programmer. In a world where technology evolves so rapidly, charting a clear path forward while staying adaptable has become crucial. How would you suggest approaching career development in such a dynamic, AI-driven landscape?
Given that model performance can fluctuate over time, how would you recommend staying updated on the latest trends? Are there specific communities, forums, or evaluation benchmarks you find helpful?
Additionally, beyond financial investment, are there particular technical skills, infrastructure setups, or best practices you’d recommend mastering to fully leverage powerful models like the o1 model?
Looking forward to hearing your insights!
1 Like
Humans seem to be perpetually driven and controlled by their imagination, indulging in the enchanting worlds it constructs while simultaneously striving to master it, lest they be entirely consumed by it.
I delight in the creativity and discovery it brings, yet I can’t help but feel weary — unsettled by its boundlessness.
There are so many variables in that one can not give any specific answer. One can not cover everything, for example
Let the real world problems guide you into what you need to learn, put in the time needed to learn it correctly even if that means months on just that.
Use a few different LLMs for the same tasks and setup so that the task can be switched to another LLM if one starts degrading or failing.
As for connecting with others in communities and forums there is no one specific one of note.
Instead of benchmarks take a look at system cards if they exist for a model, system cards often have more information useful for model selection.
The OpenAI news page references a few of them.
https://openai.com/search/?q=system+card
As hinted at in my first reply I really don’t use o1 models much, maybe once every few days. Most of my use with them was early on in trying to identify what tasks they are better at and what tasks they fail.