OpenAI released a major ChatGPT update in March, bringing multiple upgrades to the generative AI system. Making this tool more powerful and capable of handling a wider variety of tasks.
ChatGPT was already impressive as a translation tool ahead of the update so how much has it improved? In this article, we run through the key differences between ChatGPT 3.5 and ChatGPT 4 for translation.
OpenAI has refused to confirm technical details about the ChatGPT 4 upgrade. Rumour has it that the newer version increased parameters to 100 trillion, up from the 175 billion believed to be used by the previous version.
In practical terms, early testing has revealed several improvements with ChatGPT 4:
- Processing power: ChatGPT 4 can handle prompts of up to 25,000 words, from 3,000 with version 3.5.
- More human-like: The new system is better at understanding human language and generating human-like responses.
- Contextual understanding: The new system is more capable at detecting contextual patterns like emotions in content.
- Information synthesis: The latest version of ChatGPT is better at generating answers that combine information from multiple sources – something version 3.5 struggled with.
- Logical coherence: ChatGPT 4’s responses have a stronger logical coherence than its predecessor version.
- Information accuracy: OpenAI has said the latest version of ChatGPT produces 40% more accurate results and is 82% less likely to respond to requests for disallowed content.
- Image prompts: You can use images and graphics as prompts with ChatGPT 4.
The upgrade is truly impressive, even if it can’t live up to the hype generated by mainstream media. OpenAI’s CEO Sam Altman even said before the update rolled out that people were “begging for disappointment”.
His comments quieted rumours of 100 trillion parameters, but they also touch on a more pervasive issue surrounding generative AI technology.
There’s overall media sensationalism about the progress being made here, including the capabilities and immediate threats tools like ChatGPT could pose. For example, the real vulnerabilities of the technology are often highlighted (e.g. misinformation, security threats, lacking regulation etc.) over its positives.
It is difficult to answer this with concrete numbers. One problem is that ChatGPT 4 is only available to paying subscribers while version 3.5 is still accessible to free account holders. So we’re getting a lot of mixed data and feedback that’s muddying the waters.
When using ChatGPT 4 and comparing it with the previous version, it is clear that it is a better system.
Above all, the new version makes fewer mistakes like misinterpreting the source content. ChatGPT 4’s improved comprehension skills help it to also identify things like creative language and emotion more effectively – and impressively it can incorporate this into its translations.
Most of these improvements appear to be happening on the natural language processing of the source language. In other words, it is doing a better job of understanding prompts and this improves the input quality for translations.
We can’t see any evidence of clear improvement on the translation end itself, but the interpretational stage is clearly better.
Of course, the interpretation of the source material is just as important as the translation stage itself. It is also the “easier” area to make improvements – and, ultimately, it does improve the quality of translation.
Even still, ChatGPT 4 is beating dedicated translation tools like Google Translate at certain tasks.
It is important to consider all of the variables. Without knowing what data ChatGPT has access to, it is difficult to make comparisons. For example, ChatGPT almost certainly has access to translated versions of many books – so in theory and practice, it should outperform a tool like Google Translate in several instances.
ChatGPT 4 has all of the same weaknesses as its predecessor and any comparable AI system. These problems aren’t going to disappear for now. OpenAI has managed to reduce the prevalence of some of these issues; others will require more significant technological advances to address.
ChatGPT 4 is just as confident as its predecessor when it generates some inaccurate information. Yet, this isn’t unique to ChatGPT and it is one of the biggest risks involved with this type of technology. The upgrade has reduced the prevalence of misinformation which is a positive development.
We also have to remember that ChatGPT is not a translation tool after all. It is an OpenAI system that is capable of performing a variety of automatic translation tasks.
If you need help implementing the latest language tools and AI solutions in the market, including generative AI solutions for translating content, our team is here to advise and help you find the right package for your projects. Get in touch with us by completing the form on our contact page and we’ll get right back to you.