OpenAI Releases its Powerful AI Upgrade: GPT-4

OpenAI released GPT-4 on March 14, 2023. The new tool is based on multimodal AI, indicating that the system will be able to operate on multiple inputs, like video, images and sound. OpenAI, the creator of the chatbot sensation ChatGPT, highlighted the use of its latest technology “multimodal,” which indicated that images, as well as text prompts, can be integrated to generate content.

Multimodal Large Language Models

The big takeaway from the announcement is the use of multimodal AI. Modality refers to the input type that a large language model deals in. It can encompass text, speech, images and video. The previous versions, GPT-3 and GPT-3.5 were able to operate in just one modality, text. But reports have highlighted that GPT-4 may be able to operate on four modalities, images, sound, text and video.

With the integration of multimodality, Microsoft(OpenAI) will make the models in GPT-4 more comprehensive. And the point of the breakthrough here is that this model will now transcend language with its ability to draw knowledge across different languages. While the new language model from OpenAI might not seem that dangerous, the worst risks are the ones a user cannot anticipate.

While OpenAI has broken new ground, as per the company, GPT-4 is more highly capable and accurate than the previously released ChatGPT. GPT-4 has performed astonishingly well on a lot of tests, including the Uniform Bar

Exam as well as the Biology Olympiad. GPT-4 was able to ace a number of examinations including the Advanced Placement exams.The added intelligence in GPT-4 enables it to respond more fluidly than its previous version, making it more comfortable with a wider range of tasks. It appears to be less unhinged than the original Bing.

Unlike Bing, GPT-4 flat-out refused to take the bait when a user tried to get it to talk about consciousness, or get it to provide instructions related to immoral activities. In addition to working with text, GPT-4 is also able to analyze the contents of images. OpenAI has not released this feature to the public yet over the concerns about it being misused.

Technology has advanced so far that it works in all languages. GPT-4 can function across all languages. Another interesting attribute is that Microsoft is now working on confidence metrics to ground its AI with facts, thereby making it more reliable. So, should we be excited about the release of the new GPT-4 or scared of it?

And the right answer is both. While on the positive side, GPT-4 is a powerful engine for creativity, it can prove to be a tool giving rise to a number of cyber threats and attacks.