Testing Chat GPT-4o’s Capabilities — Part 1

Lakshmi Prakash
Design and Development
5 min readMay 28, 2024

Open AI has launched GPT-4o (as they say, “o” for “omni”), and everyone is excited! As always, people, especially the technology lovers and AI enthusiasts are keen to see what this new, improvised AI model can do for us.

One, the model has significantly better capabilities when it comes to non-English languages! That is interesting.

When using GPT-4o, ChatGPT Free users will now have access to features such as:

  1. Image Recognition Accuracy and Providing Exactly The Information That Was Asked For:

For this, I tried uploading photographs of some plants and asked the model to identify them and give me good growing conditions and what conditions could be harmful for these plants. And GPT-4o did a good job!

2. Understanding non-English Languages and Responding in The Same Language:

I tried asking a simple question in my local language (I got the Tamil script from AI translation, yes): “My back hurts because I spend most of my time working on a computer sitting on a chair. Can you suggest some remedies to help me?” in Tamil, and while the response took some time, yes, there was a delay, but the answer was almost accurate!

How good is Chat GPT-4o?

3. Searching and Reading Content from A Browser:

For this, I asked the GPT-4o model to give me my horoscope for the day (because this is not something you can pre-train a model with, plus different sites say different things). Here is what I noticed: the model said it searched 6 websites, gave me an answer and cited two websites for reference.

While one website’s (Horoscope.com’s) answer was accurately read and rewritten as the answer, another website’s answer (Cafe Astrology) did not match the given answer at all! Either the cited references are wrong or the information read is wrong.

4. Reading Data and Drawing Graphs:

According to the information provided about the capabilities of GPT-4o, the model is capable of analyzing data, yet repeated attempts failed when a dataset was uploaded and I asked it to analyze it for me —

Does this mean that the model is incapable of reading data and analyzing it, or is this an NLU problem, or is this a problem caused by the dataset being quite large, or is there something else?

5. Image Generation:

This was deeply disappointing!

Conclusions:

Based on my first experience of Chat GPT-4o in the first week, this is what I would conclude:

  • The image recognition is good, quite reliable!
  • The understanding and communication in Tamil, a non-English language, is quite impressive.
  • The web-/browser-reading capacity can be unreliable — better to do your fact-checks yourself!

Predictions on How Human GPT-4o Can Be and Which Jobs Could Be Replaced by this AI model:

  • Specialists and experts who use their knowledge and expertise based on years and years of practice and learning, especially in fields where the “knowledge” is mostly constant and universal and can be easily memorized (which machines can do easily) and does not require much understanding, like consultants in the field of gardening, may be at risk of losing their jobs in the coming few years.
  • With many different NLP models learning non-English languages better and better with newer upgrades, the jobs of translators in the fields of writing and publications could be taken over and/or easily replaced by AI after a few years. At this rate, no human language can be so complicated after 5–10 years that an AI cannot understand and translate. Businesses that make money out of translation as their primary service could also be at risk of closing down.

For more information, you can check these links:

https://openai.com/index/hello-gpt-4o/

https://openai.com/index/gpt-4o-and-more-tools-to-chatgpt-free/

https://www.theverge.com/2024/5/13/24155493/openai-gpt-4o-launching-free-for-all-chatgpt-users

--

--

Lakshmi Prakash
Design and Development

A conversation designer and writer interested in technology, mental health, gender equality, behavioral sciences, and more.