Tyler Weitzman, co-founder, president and head of AI at Speechify.
The latest version of OpenAI’s language model system, GPT-4, was officially launched on March 13, 2023 with a paid subscription giving users access to the Chat GPT-4 tool. At the time of writing, full access to the model’s capabilities remains limited and the free version of ChatGPT still uses the GPT-3.5 model.
Overall, GPT-4 appears to be more functional, responsive, and secure than GPT-3 or GPT-3.5. Yet with BingChat from Microsoft using the GPT-4 language model and the company dealing with many complaints and criticisms about some of Bing Chat’s strange responses, it’s fair to say that these limitations temper any expectation that GPT-4 will be an out- of-the-box “revolution”. ”
CEO of Open AI, Sam Altman, even said in an interview that some people will be disappointed with the release of the GPT-4, because there will be nothing amazing in it. Overall, though, I think the significant reduction in limits shows a promising trajectory for the technology, and its capabilities across multiple business domains will both benefit and disrupt many industries.
As a co-founder and head of AI at my company, I have closely followed the development of ChatGPT. Here’s what I see the recently released GPT-4 has to offer those who want to be at the forefront of their industry.
Influenceable real-world outcomes: use in business
According to OpenAI, GPT-4 “passes a simulated counter exam with a score around the top 10% of test takers; in contrast, the score of GPT-3.5 was around the bottom 10%.”
On Twitter, I’ve seen many developers testing GPT-4 with requests to code old-school arcade games. While GPT-3 often provided buggy code that had to be requeried, in many cases GPT-4 provides perfectly working code to start.
While some consider it potentially concerning for programmers, my company is considering allowing candidates to use it in job interviews as they are expected to use it on the job.
However, there are legitimate concerns and some big tech companies have banned their use for engineering for fear that their private company code will end up in the hands of OpenAI to train future models. Even with this in mind, the tool has plenty going for it.
Unlike GPT-3.5, the latest model accepts images as input in addition to text instructions. For example, users can enter a handcrafted sketch into the AI chatbot and the sketch will be converted into a functional web page.
Businesses can also use the image processing feature to:
• Improve the customer shopping experience through personalized visual recommendations and searches.
• Improve customer service with better chatbot engagement.
• Optimize their content and quickly flag inappropriate images.
• Create captions and otherwise improve accessibility.
Longer word processing
Large language models such as GPT have a restrictive context window. This context window is rather restrictive, as it means that GPT cannot be easily used to generate something like an entire novel at once.
The new GPT-4 model has a long-form mode that provides a context window of 32,000 pennies (52 pages of text). That’s more than an order of magnitude larger than the previous GPT-3 API that only offered 2,049 tokens (three pages of text).
For example, you can enter the URL of a website in GPT-4 and ask it to analyze the text and create engaging long content. Or you can give it a 30-page lawsuit and ask it to be analyzed.
Companies can also use GPT-4 for financial analysis, cost-effective medical diagnosis, identifying vulnerabilities in cybersecurity systems, and as a way to analyze business plans.
GPT-4 specifically improved on being able to track the “system” message, which you can use to ask the model to behave differently. This allows you to ask GPT to take on a role, such as a software developer, to improve the performance of the model.
OpenAI claims that GPT-4 is safer and more responsive than previous models. During the company’s testing, it was “60% less likely to make things up.”
However, there are still some limitations. Like its predecessors, GPT-4 can still confidently present “hallucinated” facts and make multiple errors of reasoning. This is concerning because users can learn to trust that the model is accurate most of the time.
To remedy this, I recommend that companies establish robust systems to review and verify information contained in GPT-4 generated content before it is published or passed on.
Another limitation is the lack of knowledge of events after September 2021. This deprives users of the most recent data.
To make effective use of the latest update, it is important for business leaders to recognize the prospect of harmful advice, erroneous lines of code and inaccurate information.
How can companies use GPT-4 to thrive in an AI-powered world?
The first thing I think companies should do is develop a team with extensive AI knowledge to optimize the use of the tool. To excel in this AI-driven world, your company can:
• Stay informed. As a company, keep an eye on the latest improvements with GPT-4. Keep experimenting with new features to learn how to get more accurate answers and integrate them into your business to improve your overall performance.
• Prioritize users. User experience is the top priority of any customer-oriented company. So make sure your AI chatbot has a simple, easy-to-use interface that provides useful information to customers. By using customer feedback you can optimize the chatbot’s responses.
• Review your work. GPT-4 can produce accurate answers based on your directions. With increased math skills, it can get data sheets and draw conclusions. Let it review the code and analyze documents to evaluate if there are ways to optimize your final product.
gotechbusiness.com Business Council is the leading growth and networking organization for entrepreneurs and leaders. Am I eligible?