Meet OpenAI o1 - preview now released!

Hi,

Further developments at OpenAI!

An early version of their new model, OpenAI o1, has now been released and its deep problem-solving capabilities will be of particular interest to anyone working in science, coding or mathematical fields.

What’s new?

The new models have been designed to tackle complex problems with a focus on improved reasoning.

A quick response can be incredibly useful BUT when you’re dealing with complicated problems and more technical work, the quality of the answer is far more important. 

So how does OpenAI o1 achieve this?

• Using a chain of thought when attempting to solve a problem

• Recognising and correcting its own mistakes

• Learning to break more complicated steps into smaller, simpler steps

• Trying different approaches if the current approach isn’t working

What are the benefits?

Because OpenAI o1 has been designed to spend more time thinking through problems before responding, it allows improved performance when handling complex challenges.

Great news for anyone using AI for difficult tasks which involve data sequencing, complex formulas and multi-step workflows.

When testing the o1 model against GPT-4o, it sounds as if OpenAI has made significant advances.

For instance, the new model outperformed GPT-4o significantly in an International Mathematics Olympiad (IMO) qualifier, scoring 83% compared with GPT-4o’s 13%. In coding tests, it reached the 89th percentile in Codeforces competitions. 

Though the current version lacks some of the user features we’ve got used to with GPT-4o, it has the potential to be a powerful tool in areas like quantum physics, genetics, healthcare research, and software development.

What else is new? 

Safety 

Safety is at the fore in this release with enhanced safety protocols which have allowed OpenAI o1 to outperform earlier models in resisting jailbreaks - a term used to describe users’ attempts to bypass safety rules.  

OpenAI is also working with US and UK AI Safety Institutes to ensure continued improvement and governance of these models.

OpenAI o1-mini

Almost like a kid brother to the o1 series, OpenAI o1-mini has been developed as a faster reasoning model which is well-suited to coding tasks. It offers a more cost-efficient option for developers (around 80% cheaper) and should also be made available to ChatGPT free users in due course. 

Looking ahead

It’s still early days and the released version is very much a preview but this is just the beginning for the OpenAI o1 series. (Even the naming convention suggests that OpenAI view this as a fresh start rather than a tag-on to an existing model)

It doesn’t yet rival all of GPT-4o’s features so we’re not relegating GPT-4o to the back of the cupboard just yet. But future updates should bring web browsing, file and image uploads as well as broader features to make OpenAI o1 even more versatile.

It’s available for immediate use in ChatGPT Plus and Team with access for Enterprise and Edu users over the next week. 

It will be interesting to see how the o1 series develops.

OK, speak soon!

Inge
Top Apps Founder

P.S. Check out the Top Apps AI courses here.