OpenAI’s GPT-4 is rather proficient in Afrikaans

  • OpenAI’s GPT-4 is a large multimodal model that can accept both image and text input.
  • While the model can only respond with text, it’s said to be more accurate especially when handed more nuanced instructions.
  • The model is also incredibly proficient across a range of languages including Afrikaans.

As generative text platforms powered by artificial intelligence continue to grow, those making use of OpenAI’s tech are about to get a leap forward.

That’s because the firm just launched its GPT-4 model. This large multimodal model is able to accept image and text inputs and generate responses. For example, in a post on its website, the model is fed an image of eggs, flour, milk, and butter along with the prompt, “What can I make with these ingredients”. The model responds with suggestions of various baked goods.

However, the model may not be immediately better in average interactions compared to GPT-3.5 (the tech that powers ChatGPT) OpenAI notes in a research blog post.

“In a casual conversation, the distinction between GPT-3.5 and GPT-4 can be subtle. The difference comes out when the complexity of the task reaches a sufficient threshold—GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5,” the firm writes.

Testing the model in a number of benchmarks however, yielded improved results for GPT-4. In simulated exams, the model scored in the 88th percentile in the LSAT exams compared to the 40th percentile GPT-3.5 scored.

However, to go a step further OpenAI wanted to benchmark GPT-4 in other languages. To to this it translated the MMLU benchmark into 26 languages. The benchmark is a suite of 14 000 multiple choice problems spanning 57 subjects.

One of those languages was Afrikaans where GPT-4 scored 84.1 percent in Afrikaans compared to 85.5 percent in English. Clearly, there are more than 44 people speaking the language, Charlize.

More interesting is that GPT-4 scored 78.5 percent in accuracy when using Swahili which OpenAI notes is a “low-resource” language alongside Latvian and Welsh.

The research OpenAI has presented alongside this release is fascinating and we highly recommend visiting the link above even if you’ve only had a passing interest in ChatGPT.

The new model is currently available as part of ChatGPT Plus which costs $20 (~R362) per month. It’s also available as an API for developers to build services with.

Further to this, Microsoft also announced that the new Bing is running on GPT-4.

“We are happy to confirm that the new Bing is running on GPT-4, which we’ve customized for search. If you’ve used the new Bing preview at any time in the last five weeks, you’ve already experienced an early version of this powerful model. As OpenAI makes updates to GPT-4 and beyond, Bing benefits from those improvements. Along with our own updates based on community feedback, you can be assured that you have the most comprehensive copilot features available,” Microsoft’s corporate vice president of Modern Life, Search and Devices, Yusuf Mehdi wrote in a blog post.

Windows 11 users who have access to the preview can now also access the new AI powered Bing from the search bar.

OpenAI does note that GPT-4 is still not perfect and is still limited by social biases, hallucinations and adversarial prompts. That having been said, the firm does claim that the tech is 82 percent less likely to respond to requests for “disallowed content” and 40 percent more likely to produce a factual response.

Now we wait to see how folks use – and abuse – this latest update.

[Image – BoliviaInteligente on Unsplash]


About Author


Related News