Whether you’re a business leader, work in what’s been dubbed an “exposed job, ” or develop new digital solutions yourself, ChatGPT may have taken up much of your brain space (and news feed) recently.
After all, can you name a new technology of recent years that’s captivated the world quite like this groundbreaking chatbot? New mobile apps, prompt engineering tips and “Experts-in-AI” memes are everywhere,as individuals and organisations begin to untangle what this large language model (LLM) means for their future.
So, does that make ChatGPT the most valuable tool for our future businesses? Or are their potential problems with ChatGPT to contend with?
AN OPEN MIND ABOUT OPENAI
Being an IT and digital transformation specialist, Node4 is naturally curious about ChatGPT (which is developed by OpenAI) and how it could shape our client’s and our own future. We view innovations – no matter how potentially disruptive – with an open mind, while remaining pragmatic about any pitfalls.
For example, does ChatGPT represent an opportunity to improve services and, therefore, the customer experience? Our partner Microsoft certainly thinks so. In fact, Microsoft is already enhancing Azure developer services with OpenAI integrations that are stretching imaginations and capabilities – imagine being able to literallytalk to your database in natural language and get answers from it!
On the flip side, ChatGPT does have its problems. That’s because it also has the power to empower nefarious actors even further. Think, deep fake creation, writing malware or performing automated phishing attacks.
So, does that make ChatGPT and LLMs a force for good, a dangerous pandora’s box, or something in-between? Probably the latter, but the reality is that it’s still early doors and we can’t say for sure how relevant any issues or downsides with ChatGPT will be.
NAVIGATING NEW GROUND
For many, myself included, ChatGPT is like nothing we have seen before. The model and its capabilities are absolutely mind-bending, and wildly popular as a result.
Just look at the numbers. ChatGPT attracted over 100 million users within two months, And that user count is continuing to rise now that GPT 4 is here – which has an order of magnitude more capability to the previous model, GPT 3.5.
Instead of casual users, we’re now seeing individuals and organisations benefiting from industry specific LLM’s to develop medical papers, write case law or design apps for specific purposes – some controversial, like completing homework! (Which some may say is a problem with ChatGPT).
WHERE CHATGPT IS HEADING
By this point, most of us are familiar with ChatGPT’s common use cases: content creation, translation and writing code, to name a few. You may also be acquainted with OpenAI’s sister model, DALL·E 2, which can create original images with incredible efficacy based on natural language prompts.
But in a recent TED Talk, OpenAI co-founder Greg Brockman stunned watchers by demonstrating the true extent of ChatGPT’s capability. In particular, Brockman showcased the results of several unreleased plugins. He showed ChatGPT:
- Generating recipes with images of finished dishes
- Drafting and publishing tweets without leaving the chatbot
- Interpreting data-intensive spreadsheets
- Fact-checking its work in line with user intent
After the talk, the head of TED, Chris Anderson, joined Brockman to discuss the problems with ChatGPT and the risks of releasing such a powerful tool. Brockman’s stance was to release the machine before it’s super powerful and let feedback drive incremental improvement.
Brockman says he wants to establish a feedback loop whereby the machine learns from user input and intent. His goal is for ChatGPT to gradually develop the capability to apply knowledge in novel situations, believing the outcome to be a better world for everyone.
Closing off his talk, Brockman said that artificial general intelligence (the principle behind ChatGPT) is poised to transform every aspect of how we use computers and urged us all to get literate in AGI technology. This was presumably to ensure everyone is ready for a future job market which could be radically different.
WINNERS AND LOSERS?
OpenAI themselves published a paper in March 2023 showing the impact on the labour market of LLMs and the potential pitfalls with ChatGPT’s latest release, GPT-4. The occupations with the highest exposure include mathematicians, tax preparers, writers, web designers, accountants, journalists and legal secretaries, while bloated software vendors are also at risk.
In a similar vein, many commentators have suggested that the future of startups will fundamentally change. Market entry will be accelerated as a result of expedited development and access to (artificial) skills. Imagine being able to launch a minimum viable product for ten times less cost?
So, does that mean there are ? Well, ChatGPT has seemingly established some new technology battlegrounds, Search being one. Google, which, as you’ll know, has dominated online search since the 1990s, may soon find itself needing to change strategy.
Anecdotal stories suggest that when ChaptGPT was released, it triggered Google’s board to press the panic button. But why was the risk perceived as being so high?
Well, for many search queries, the speed and conciseness of ChatGPT responses do away with endless scrolling and therefore have significant user appeal. As such, Google quickly responded and released its own model, Bard.
And if you follow AI (or any other) news, you’ll know that Bard’s launch didn’t exactly go to plan. Instead, it served more as a reminder that these models are still in their infancy, with much room for improvement and to be used pragmatically.
Even though Google tripped up, it would be unwise to discount them. For years, they’ve been the AI pioneer, leading the way with over 9,000 AI related papers released between 2020-2022 (with Microsoft a close second at 8,000 papers). If, and when, Google decides to go to market with the AI innovations they’ve kept under lock and key until now, 2023 may well be their year.
TREATING AI WITH CARE
As fascinating and potential-packed as ChatGPT and new LLMs are, we must be wary of their pitfalls and the specific problems with ChatGPT. Leveraging any evolving technology calls for expert evaluation, vigilance and a strategic mindset. The same applies to here, which has the following downsides at present:
Response Hallucinations: Another issue raised by ChatGPT is that it is prone to “hallucinations” – when outputs may sound plausible, but are factually incorrect or irrelevant in context. Hallucinations occur because ChatGPT is not sentient and only has training data for reference. By lacking real-world understanding and evaluation skills, it can therefore produce wrong responses by simply filling in the gaps itself.
Ethical Bias Concerns: LLMs are trained on largely unknown datasets from specific points in time. If the humans behind these datasets (and indeed, ChatGPT’s development) have conscious or unconscious biases, it goes without saying that response outputs may be flawed.
Qualified Expert Insight: By design, LLMs do not understand context. They have no concept of the meaning of what is being typed out, other than a likelihood percentage that a particular word follows another word. Whether it’s staying on top of the latest developments, tapping into customer mindsets or understanding technical nuance, sometimes human knowledge is irreplaceable.
Dependency on Prompts: With ChatGPT, “you get out what you put in” has reached an entirely new level of meaning. Currently, the response is contingent on how well you prompt. That includes the prompts and the user’s subject knowledge (i.e., a doctor may struggle to prompt about application reconfiguration or a developer about diagnostic probabilities).
However, the latter point may only be considered a pitfall for a little longer. Enter: AutoGPT. This exciting development essentially uses virtual agents to complete tasks on your behalf, including build the prompts necessary without user interaction. That solves at least one of the potential problems with ChatGPT!
How it works can be summarised as follows:
- It uses GPT-4 to carry out complex, multi-step procedures by creating its own prompts and feeding them back to itself, creating a loop.
- It breaks a larger task into smaller sub-tasks and spins off independent instances to work on them, while the original instance acts as a project manager to coordinate and compile the finished result.
- Browses the internet and includes information it finds in its calculations and output.
- Uses the results generated to create new prompts, chaining operations together to complete complex tasks.
- It can help with tasks that are too complex for traditional AI models, such as researching new technologies and companies, or planning detailed projects.
Some examples of AutoGPT in action including creating applications in minutes, devising a startup plan for £100 or performing detailed market research. Amazingly, users can watch virtual agents communicating in real time – using online content to solve problems and even correcting one another’s code!
That said, AutoGPT is just a few weeks old but in twelve months’ time, we expect to see real disruption. Some commentators have even predicted that AutoGPT one day could become the next big film producer, with verbal prompts able to manifest a movie (visual effects, video output etc).
ChatGPT and its related technologies have sparked a tremendous amount of interest amongst businesses and promise to help us achieve extraordinary things even faster. From assisting with day-to-day tasks to being the genus of an incredible breakthrough, the days of human and artificial intelligence collaboration are well and truly upon us.
To keep up to date with the latest in AI, follow the Node4 blog. Or, to understand how artificial intelligence and automation can enhance your IT infrastructure and processes, arrange a no-obligation chat with a Node4 consultant by clicking below.