OpenAI unleashes ChatGPT-4 with impressive new features
OpenAI is starting to roll out the improved ChatGPT version 4, but this is currently only available for the Plus subscribers. That said, even those who benefit from the Plus subscription seem to be having problems accessing all the features, as there are some bugs that still need to be ironed out. Now, right off the gate, OpenAI claims that version 4 produces not only more useful responses, but it is also safer. It is not really clear how this is safer exactly since Bing with its awkward hallucinations is also based on ChatGPT-4.
Besides those shortcomings, the new version has new cool features that help solve problems more efficiently:
- Chat answers can include as much as 25,000 words instead of 3,000 and the AI can also analyze up to 25,000 words and reorganize everything in a more concise form.
- Users can now upload pictures and ask the AI to provide advice or produce certain solutions based on the image. There is an example with a picture showing some food ingredients and the AI is asked to list some recipes that include those ingredients. In another example, somebody uploaded a picture of some html code written on a napkin and the AI managed to output that code to a fully functional website.
- ChatGPT appears to have Internet access now, so it is not limited to whatever data it was trained on.
- Increased reasoning capabilities.
Moreover, OpenAI lists some real-world use cases that benefit from the new version:
- Duolingo allows users to learn a foreign language faster and more efficiently.
- Be My Eyes - a phone app that analyzes food labels and gives more information about ingredients, healthy habits etc.
- Streamline user experience and combat fraud on Stripe.
- Financial wealth management at Morgan Stanley.
- Smart tutoring at Khan Academy.
- Language preservation for certain countries like Iceland.