GPT-4 and the Quest for Human-Like Intelligence in AI
The Service Desk Institute (SDI)
Inspiring service desks and support teams to be brilliant
We have been introduced to the powerful ChatGPT less than four months ago, and the whole world has again been struck by a new ‘shiny thing’ released by OpenAI.
Of course, we're talking about OpenAI's most advanced system GPT-4.
As a part of OpenAI's ongoing commitment to providing deep learning to a broader audience, they have developed a new large multimodal model. Although this model isn't quite up to human standards just yet, it has shown human-like performance on a number of industry and academic benchmarks.
But what new features does GPT-4 bring to the table compared to ChatGPT??
When comparing ChatGPT and GPT-4, it's essential to keep in mind that they aren't competing against one another. GPT-4 is an upgraded version of GPT 3.5, which ChatGPT uses.
Before we dive into GPT-4, let’s remind ourselves of some impressive ChatGPT results.
After all, it has been threatening to change how we do our jobs since its release.?
ChatGPT is a large language model (LLM) or a subset of artificial intelligence that has been trained on massive amounts of text data to create human-like answers to a conversation or other natural language inputs. One of the most known LLMs today is exactly GPT-3, which stands for Generative Pretrained Transformer 3 or known as ChatGPT.
Only five days after its release, ChatGPT had already attracted the attention of about 1 million users!?
And if that’s not impressive enough, it reached the milestone of 100 million active users in January this year.
Based on the GPT-3 (Generative Pre-trained Transformer) architecture, Chat GPT is pre-trained on a massive dataset of text, tested in real-world scenarios, and fine-tuned to perform tasks such as:
So, no surprise that loads of people are using ChatGPT and other similar tools today.?
And we can already notice different types of users. Everyone, from curious users like hobbyists, and students to serious users like professionals and organisations, have found value in exploring the possibilities offered by this tool.?
There is also an increasing group of people and organisations seeing this more as a business opportunity.??
In fact, ChatGPT's underlying technology is already being used by companies including Meta, Canva, and Shopify in their customer support chatbots.
?
Here Are Some Interesting Facts About GPT-4
Not available for public testing
One of the first thoughts that no doubt crossed our minds when we heard about this new release was, well, how do we try the new model?
This new model has not yet been available for public testing. However, if you'd still like to try it out, you can do so if you subscribe to a paid version or ChatGPT Plus and join the GPT-4 API waitlist.
GPT-4 can understand visual inputs
?The first task GPT-4 can do, which GPT-3.5 never really got to do, is to use a different input type to produce answers. As OpenAI's cofounder and lead researcher Sam Altman explains, the GPT-4 is a "multimodal" model that can take text and images as inputs, allowing users to ask questions based on visuals and produce the text outputs.
Describing the contents of images and charts, recognising their context, and even creating a functional website from the drawn picture of the website — the possibilities are enormous.
And, of course, if you ask the right questions, you’ll get better results.?GPT-4's abilities can undeniably be described?as impressive!
?
Longer Context?
The context length has increased! So, what does this mean? Other than the difference of being able to process visual inputs, it can also produce and handle around 25000 words of text. This is much larger than ChatGPT, which can handle about 3000 words.
So, if you compare GPT-4 to ChatGPT, you will notice that you can put more text into the prompt and get more extended output too. This allows you to use long-form content, extended conversations, or document analysis.
More creative than ChatGPT
It's also more creative than the previous versions. With GPT-4, you have more power to solve complex problems and edit and iterate over the writing or technical tasks more accurately.
Another difference between the Chat GPT and GPT-4 is the reasoning capabilities. This means it can better understand the context of the problems you're trying to solve and provide a more accurate answer or solution.
领英推荐
GPT-4 is less likely to make mistakes, which was a big worry for many people who used Chat GPT.
Based on the Open AI internal evaluations, GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce accurate responses than PT 3.5.
GPT-4 has a longer memory
If we look at the GPT-3.5 and the old ChatGPT, they have a limit of 4,096 "tokens," or 8,000 words, which is approximately four to five book pages. GPT-4's maximum token count is 32,768, around 64,000 words or 50 pages.
That’s adequate for a play or short fiction!
Limited number of messages
Since its release, this new model has had the limitations of 100 messages every 4 hours or 600 messages in 24 hours. But since it’s still in the testing phases, we assume this will change soon.
GPT-4 is more multilingual
The initial testing of language capabilities is promising, but it is still incomplete and needs to fully embrace multilingual capabilities. However, GPT-4 takes a step in this direction by showing that it can answer thousands of multiple-choice questions in 26 languages with a high level of accuracy.
?Highlighted metrics?
GPT-4 has been tested on a different set of benchmarks, including the ones designed for humans. One of the things that millions of people will probably first notice in the promotional materials published by OpenAI is that GPT-4 scores in the top 10% in a simulated bar exam.??
Whereas GPT 3.5 scored in the bottom 10%.?
?
Limitations of model information
?Another interesting thing about the new release of GPT-4 is the limitation of model information. OpenAI's Technical Report says that there are some limits on the information about GPT-4 architecture that can be shared, such as the size of the model, the number of parameters, hardware or the training method.
Reason? According to OpenAI, there are two main reasons behind this – the competitive landscape and the concern about the safety implications of large-scale models. This means they’re not going to tell us how they train this model.?
What Does GPT-4 Mean for Customer Service?
We recently asked you what your thoughts on this new release are. And what industries do you think will be most impacted by GPT-4?
Most of you recognised the significant effect that GPT-4 would have on customer service – and we couldn’t agree more!
Besides being better at comprehension and understanding, GPT-4 is also better as a language model. Because of the functionality of GPT-4 to provide faster and more accurate responses, as well as to assess circumstances to provide the best answer or solutions swiftly, customer service has the potential to improve with its use significantly.
Moreover, GPT-4 can generate in-depth consumer profiles from prior interactions with automated systems, which could result in more personalised service.
We have been introduced to Bing and Microsoft 365 Copilot products that are already powered by GPT-4. And businesses are now able to take advantage of the same underlying advanced models to build their own applications leveraging Azure OpenAI Service.
These solutions include making the whole customer experience better, summarising long-form material, and helping to create software. And many companies like Epic Healthcare, Coursera, and Coca-Cola are already making plans on how to utilise this for their business.
?
We’d love to know your thoughts about GPT-4? How do you think most users and companies are going to use this new piece of technology?
We hope you all enjoyed reading this!
?? If yes, hit “Subscribe”,?and stay tuned for more of the same coming next month!?
Your input is crucial to shaping the future of the industry!
Please take a few minutes to complete our survey - and be a part of the conversation!
Share your valuable insights and opinions on health and wellbeing, employee experience, people skills, industry trends, and technology trends.