OpenAI’s ChatGPT 4 Is Here. Is It Time to Forget ChatGPT?

ChatGPT 4 has officially arrived. The san-Francisco-based startup, OpenAI on Tuesday announced the official arrival of ChatGPT 4, which is clearly the most anticipated tool which has the superpower to describe images in the text given the storm ChatGPT brew late last year. The launch of ChatGPT and its aftermath was nothing less than legendary. Obviously, no one envisioned the chatbot would create tech hysteria with millions rooting for it.

ChatGPT did everything we wanted our sophisticated machines to ever do – write effortlessly, generate essays, screenplays, and even conversations. While it did blow us out of our minds, there was room for improvement as its powerful capabilities were based on an older generation of technology which was older than a year.

Recommended: E********* With ChatGPT – 10 Simple Ways To Get You Started

The Making of ChatGPT 4

Now coming to the star of the moment – ChatGPT 4, boasts cutting-edge technology which not just creates text but also has the ability to describe the images in response to any user’s input. In fact, OpenAI calls this a milestone of sorts in the field of deep learning.

ChatGPT 4 is a large multimodal model which accepts image and text input and offers text outputs. OpenAI says that even though ChatGPT 4 may be less capable than humans yet it exhibits human-level performance.

In the last 2 years, OpenAI earnestly worked and rebuilt their deep learning stack and co-designed a supercomputer for the workload with Azure from the scratch. They trained GPT-3.5 as the system’s initial “test run” a year ago. During this time, OpenAI identified several problems, rectified them, and strengthened the theoretical underpinnings. This resulted in ChatGPT 4’s training to be surprisingly stable, making it OpenAI’s first successful large model whose predictions were accurate.

ChatGPT 4 API Waitlist

Emphasizing API access, OpenAI, in its blog mentioned that it was ‘making GPT-4 available as an API for developers to build applications and services.’

Users may experience delays in access because Open AI processes requests for the 8K and 32K engines at varying speeds based on capacity. Under OpenAI’s Researcher Access Program, researchers examining the societal effects of AI or difficulties with AI alignment can also apply for discounted access.

Recommended: Power of AI: 5 Ways You Can Use These ChatGPT Alternatives to Fulfill your Business Goals

For broader use, OpenAI is working closely with one partner, to begin with, while we get the picture input capabilities ready. To encourage future model improvement, the brand is also making OpenAI Evals, the platform for automating the evaluation of AI model performance, publicly available.

ChatGPT 4 V/S ChatGPT 3.5 – the difference

While casually speaking, the difference between GPT-3.5 and GPT-4 can be negligible but it comes out when the nature and intricacy 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.’

We tested on a range of benchmarks, including recreating tests that were initially created for humans, to understand the differences between the two models.

For the AP free answer questions and Olympiads, we used the most recent publicly available tests, and for other tests, we purchased the 2022–2023 volumes of practise tests. No specific training for these exams. During training, the model noticed a minute problem, but OpenAI believes the results to be representative—refer to the technical report for details.

Most of the existing ML benchmarks are in English. And so, to assess its proficiency in other languages, OpenAI translated the MMLU benchmark— a set of 14,000 multiple-choice questions covering 57 categories, translated using Azure Translate into several languages (see Appendix). Out of the 24 languages tested, GPT-4 remarkably outperformed the English-language performance of GPT-3.5 and other LLMs like Chinchilla and PaLM. It also included low-resource languages such as Latvian, Welsh, and Swahili.

As far as steerability goes, users can specify the style and task that their AI should perform by describing those instructions in the “system” message, in contrast to ChatGPT, which had a predetermined verbosity, tone, and style. Under reasonable limits, system messages give API users the ability to drastically alter the user experience.

ChatGPT – 4: Limitations

Even though GPT–4 is way more advanced, yet it has similar imitations to ChatGPT. OpenAI in its blog very openly admitted that it still has some hallucinations with facts and one should be very careful in high-stakes contexts. OpenAI does say that yen hallucinations have ‘significantly reduced’ as compared to the previous models. GPT is 40% higher than the previous GPT of 3.5.

ChatGPT has made progress on external benchmarks like TruthfulQA, which has the skill to separate fact from an adversarially-selected set of incorrect statements. ChatGT is marginally better at this than GPT3.5 however it was noticed that after RLHF post-training, there was a considerable gap.

In general, GPT-4 does not learn from its experience and is unaware of events that have taken place after the bulk of its data is shut off (September 2021).

Sometimes, GPT-4 can also be confidently inaccurate in its predictions and neglect to double-check its work when it’s likely to make a mistake.

ChatGPT 4: Risks

The expanded capabilities of GPT-4, however, provide new risk exposures. OpenAI hired more than 50 experts from fields including AI alignment issues, cybersecurity, risk, trust and safety, and international security to test the model in an adversarial manner in order to identify the scope of these risks.

Recommended: How Generative AI is Transforming Audio Content

The brand’s ability to examine model behavior in high-risk situations that demand expert evaluation was expressly made possible by their discoveries. Feedback and information from these specialists helped to mitigate and enhance the model; for instance, we gathered more information to make GPT-4 better at turning down requests for instructions on how to manufacture hazardous substances.

In order to better understand and evaluate potential effects and to develop evaluations for potentially dangerous capabilities that may appear in future systems, we are working with outside researchers.

Conclusion

With the massive fan-following of ChatGPT and rapid advancements in artificial intelligence technology, ChatGPT – 4 is going to supersede ChatGPT with its super-efficient capabilities and powerful, state-of-the-art technology. Brands are already vouching for it to save time dramatically, design campaigns without any help, or maybe just effortlessly do the work of an entire team or reduce the number of hours by half. ChatGPT 4, needless to say, is set to revolutionize work as well as life.