This GPT has a PhD: Hear Me Out!

AI just got smarter, and I mean really smart. OpenAI’s new o1 series, which was released in September 2024, brings a new level of reasoning to the table. This model was designed to slow down and think before it responds, making it a first-of-its-kind model when it comes to handling tough problems in fields like math, coding, and science. It’s even being compared to a PhD student because it can tackle incredibly complex tasks with ease. But, unless you’re subscribed to ChatGPT Plus or Team, you won’t be able to experience this impressive jump in AI tech just yet.

So, what makes this model special? I’ll tell you a little story. After hearing all the hype about the model’s reasoning capabilities, I decided to test it out myself. I asked a simple question: “How many R’s are in ‘Strawberry’?” I had done this with other models before, but they often tripped up on such simple tasks. However, o1 nailed it on the first try, 3 R’s—without hesitation. This was the first AI model I’ve used that got it right the first time. That’s when I knew OpenAI wasn’t kidding about the o1’s problem-solving skills, oh and it does more than just find the number of R’s in strawberry lol.

It’s Smarter

The key feature that sets o1 apart from earlier models like GPT-4o is its ability to think longer before responding. Unlike GPT-4o, which prioritizes fluency and speed, o1 has been trained to slow down and evaluate problems carefully. This approach is essential for complex tasks that require deep reasoning, such as solving high-level math equations, debugging code, or even understanding advanced chemistry problems. OpenAI claims this model’s problem-solving abilities mirror those of PhD students, especially in disciplines like physics and biology, so yeah, I think you see where the title is coming from.

For example, in the International Mathematics Olympiad (IMO) qualifying exams, GPT-4o managed to solve only 13% of the problems. In contrast, the o1 model correctly answered an impressive 83% of the same problems. Which is way more than my best marks in school, probably. So this speaks volumes about its performance in challenging technical tasks.

However, there’s a bit of a trade-off. The model takes longer to generate responses because it’s reasoning through the task. This won’t be a problem if you’re tackling complex challenges, but if you need something quick and less precise, GPT-4o might still be your go to.

Real-World Use Cases

The OpenAI o1 series shines in STEM (Science, Technology, Engineering, Math) fields. If you’re a developer, data scientist, or engineer, you’ll love its ability to reason through intricate problems. OpenAI reported that o1 reached the 89th percentile in coding contests on Codeforces, a competitive programming platform. Imagine the possibilities if you’re stuck on a difficult algorithm or need to debug a large chunk of code, o1 can help you sort through it with its powerful reasoning capabilities.

Beyond coding, its performance in chemistry and biology means it could assist researchers in analyzing complex datasets or even devising new experiment strategies. It’s designed to be a partner for those in technical roles who need more than just casual conversations or superficial responses from their AI.

That said, it’s worth mentioning that GPT-4o might still have the edge when it comes to creative writing or more general tasks. The o1 model sacrifices some writing fluidity in favour of technical proficiency. So, depending on what you need, one model may be more suited to you than the other. This also implies that this model wasn’t made for everyone, unlike GPT 4o.

Want more AI insights like this? Don’t forget to subscribe to the Tino Talks Tech newsletter or allow notifications so you never miss out!