GPT-3.5 vs GPT-4: average user trying to see the difference

New versions of LLM models keep improving at an accelerating speed. GPT-3.5 is like a schoolkid who has grown into GPT-4 — an adult with a higher education. Tech notes state more and more efficiency, billions of parameters, superior reasoning and concluding abilities.

Midjourney compares GPT-3.5 and GPT-4

So can an average user like me really see the difference between “young” and “adult” versions?

Out of curiosity, I decided to ask the LLM to find the best title for my future blog. One thing led to another, and here I am, writing my first blog post!

The first step was to come up with the most banal title and then hand it over to LLM to make it smart and pretty:

“Please come up with similar phases to “Through the eyes of a manager”. Use authentic English language, add play on words if possible.”

That’s what I got in response:

GPT-3.5

Sure! Here are some similar phrases that incorporate a play on words and maintain an authentic tone:

“Through the lens of a leader” “From the vantage point of a supervisor” “In the shoes of a team captain” “Viewing the world through a manager’s magnifying glass” “From the cockpit of a corporate pilot” “Navigating the landscape as a project architect” “In the perspective of a pathway planner” “Through the looking glass of a department director” “From the helm of a team ship” “With a manager’s binoculars”

Feel free to let me know if you need more options or variations!

GPT-4

Here are several alternative phrases similar to “Through the eyes of a manager,” incorporating authentic English and some playful wordplay:

“From the Manager’s Vantage” “In the Manager’s Mirror” “A Manager’s Mindset” “Behind the Manager’s Desk” “Through the Manager’s Lens” “The Manager’s Perspective” “At the Helm: A Manager’s View” “The Manager’s Take” “Under the Managerial Microscope” “Through a Managerial Prism” “In the Driver’s Seat: A Manager’s View” “From the Captain’s Chair” “Through the Office Window” “A Manager’s Eye View” “From the Top Down: A Manager’s Look”

These phrases provide varied angles, from direct perspectives to metaphors that evoke the experience and insights of a manager.

So is there a difference in the “intellectual level” for an average user at first glance? A short and quick answer: yes, there is a huuuge difference! Frankly speaking, I didn’t expect it to be so obvious.

GPT-3.5 is nowhere near GPT-4 in the suggested responses. It feels like automatically grouped expressions from different professional fields (GPT-3.5) vs way more carefully selected options (GPT-4).

GPT-3.5: why so many random roles (captain, manager, pilot, architect)? What is this “From the vantage point of a supervisor” or “Through the looking glass of a department director”? Who could possibly consider this a creative manipulation of language?

In case of GPT-4 the options sound more mature and to the point (“The Manager’s Take”, “Behind the Manager’s Desk”). Of course, they are not all perfect and also provide clichés but it feels more precise and eloquent.

The concluding sentence is also worth mentioning. It shows a difference in “education level”. GPT-4 has a hint of academic sophistication by providing a clear and precise summary:

“These phrases provide varied angles, from direct perspectives to metaphors that evoke the experience and insights of a manager.”

Let’s compare it with GPT-3.5 which does not provide any summary, only a call to action:

“Feel free to let me know if you need more options or variations!”

Although it is a challenging task to find a creative title, I was lucky with an option from the GPT-4 list. “The Manager’s Take” blog is up and running, which is a success!

To conclude: yes, both GPT-3.5 and GPT-4 are valid and viable. I can clearly see how GPT-3.5 can be helpful with answering questions, translating and, for example, generating summaries for long texts. It is also free which is another advantage. Still, for any writing activities I would definitely go with GPT-4 for depth, eloquence and linguistic nuances.

P.S. It’s hard to stop comparing. One other response, this time from Llama 3.1 8B. It’s like approaching completely different personality:

Here are some similar phases with a play on words:

Through the lens of leadership: This phrase suggests looking at things through a managerial perspective, just like using a camera lens. From behind the bench: In sports, the manager sits in the dugout (or “behind” home plate). This phrase implies being in charge and overseeing everything that happens. In command of the scene: This one positions you as the conductor of a symphony, orchestrating every element to create a harmonious outcome. On top of operations: You’re literally sitting at the pinnacle, with an all-encompassing view of what’s happening on the ground. At the helm: This nautical phrase suggests steering a ship through treacherous waters (or calm seas). As manager, you guide your team to chart its course. In charge of the board: Like a chess player studying their pieces on the board, this phrase highlights your strategic oversight and decision-making authority.

Feel free to pick the one that fits your style!

Llama decided to provide clarification for every expression. I would say that Llama can compete with GPT-4. Although GPT-4 is trained on close to 1 trillion parameters vs 8 billions for Llama, at some point this difference is not as noticeable because such questions obviously do not require full capabilities.