2 for 1: A week with ChatGPT-4 and its browsing breakthrough
10-minute read (approx)
ChatGPT is an artificial intelligence (AI) tool that understands and responds to human language. It can answer questions on anything you want to know. It’s been generating opinions and chatter from everyone and their mum since its release —and I've been hooked. So it was to my delight when an email landed in my inbox on the 16th of March, inviting me to a test group on Slack to give feedback on a new feature.
I’m not trying to brag, but what can I say? OpenAI sure knows how to make a gal feel special. This feature, I learned had the capability of knowing how and when to browse the internet. Not only that, but during that week of testing, ChatGPT-4 was also released to ChatGPT Plus members. 2 for the price of 1, what’s not to love?
Now that the testing period is over, I want to share an overview of the good, the bad, and the ugly of what I’ve learned from that community and my personal musings over the past week. But just before I go ahead, no ChatGPT did not write this.
The good
It can’t access everything
This is because it respects the privacy settings. For example, I wanted ChatGPT to give me all the keywords of a LinkedIn job description, but it couldn’t access it due to the “robots.txt” file. What is a robots.txt file you ask? According to ChatGPT:
The “robots.txt” file is a file that website owners can create and place in the root direction of their website. It provides guidelines to automated agents, such as search engine crawlers and bots, about which parts of the website they are allowed or not allowed to access or index.
Every day’s a school day. Originally I put this under ‘The bad’ since it couldn’t do what I wanted, but thinking about the bigger picture, and data privacy concerns, it’s a good thing that privacy is being respected.
Find what you need (within reason)
It worked brilliantly providing accurate information on where to find the 5 collection chests in Hogsmeade in Hogwarts Legacy game, which was really impressive. If you want to find out the weather forecast or the top news stories in your country, ChatGPT has your back.
But, ask a question that requires a long answer and be prepared to keep saying “please continue” from time to time. The “please” being entirely optional of course, but I will continue to use pleasantries so that when the AI uprising begins, I’ll be in its favour.
Find the perfect prompt
This is what I got the most excited about. A "prompt" is a text-based instruction or question that you give to ChatGPT to initiate a conversation or request a specific output. To get the most out of ChatGPT, crafting the right prompt is key. A tester in the Slack shared a YouTube video - The ONE ChatGPT prompt to rule them all and I’m obsessed. Simply drop this into ChatGPT, and find the perfect prompt for your needs:
I want you to become my Prompt engineer. Your goal is to help me craft the best possible prompt for my needs. The prompt will be used by you, ChatGPT. You will follow the following process:
1. Your first response will be to ask me what the prompt should be about. I will provide my answer, but we will need to improve it through continual iterations by going through the next steps.
2. Based on my input, you will generate 2 sections. a) Revised prompt (provide your rewritten prompt. It should be clear, concise, and easily understood by you), b) Questions (ask any relevant questions pertaining to what additional information is needed from me to improve the prompt).
3. We will continue this iterative process with me providing additional information to you and you updating the prompt in the revised prompt section until I say we are done.
Using this prompt I was able to create a step-by-step plan on how to create a content strategy for my business, and a learning plan to support my development. Pretty cool I think.
I also used the same prompt again and asked it to read all the Midjourney documentation to find me the perfect prompt to use in Midjourney to generate a realistic photograph of a couple watching the Northern Lights, down to the detail of the camera setup.
The bad
It can’t access everything
I’ve already covered the good things about this, but how much this limits its ability to find information, and identify the accurate from false information is a big concern. It also has a hard time reading pages that rely heavily on JavaScript according to other testers.
It wasn’t able to successfully summarise a PowerPoint file or transcript for me. The transcript was quite lengthy (4267 lines to be exact) but even publishing it to the web and specifically prompting ChatGPT to “scroll down the page” didn’t lead to success, and copying and pasting it into the chat was a big fail, because it was just too long.
Not great at reading PDFs
There was mixed success for ChatGPT’s ability to access and read PDFs. It’s able to access PDFs on a publicly accessible website but not Google Drive. According to ChatGPT it’s because:
Google Drive presents a different interface for viewing PDFs and makes is difficult for me to extract the text. Additionally, if the text within a PDF is embedded as an image rather than selectable text, it would also make it difficult for me to extract the contents.
If this isn’t further evidence to demonstrate that PDFs are problematic for online reading, due to their inaccessible nature, then I don’t know what is. If ChatGPT becomes clever enough to not only access them but understand them too, I’ll eat a slice of humble pie.
Wrong or outdated information
Sometimes it’s right and sometimes it gets things wrong, such as providing an incorrect final score for a football match. To get the most accurate and up-to-date information, users need to be specific in their prompts and use terms like "search" or "browse the web" to instruct ChatGPT to search the internet. Otherwise, it may rely solely on its base model training data, which only includes information available until the beginning of 2022 (OpenAI, 2023).
It can’t access YouTube videos
If you want ChatGPT to give you a summary of the key points in a YouTube video, sorry but no-can-do. Not right now anyway. Fortunately, there are Chrome extensions available to do this for you.
As an AI language model, I am a text-based entity and do not have the capability to access, view, or interpret audiovisual content such as videos on platforms like YouTube. My abilities are limited to processing and generating text, and I can access and provide information from text-based web pages using my browsing tool.
The ugly
The ‘hallucinations’
The term "hallucination" or "artificial hallucination" refers to a phenomenon where an AI system generates a confident response that does not appear to be justified by the data it was trained on (Wikipedia, n.d.). When I was trying not to lose my mind to the frustration of getting ChatGPT to identify key points from a transcript for me, ChatGPT had 2 hallucinations.
First, it made up the key points on a totally different topic, rather than telling me it couldn’t access the transcript. Then when I resorted to copy and paste over several parts, it simply carried on an imagined conversation once I was finished, which was a bit creepy.
The ethics (or lack of)
Before writing this article, I wanted to see if ChatGPT would forego Slack’s terms and the consent of users to find me a solution to categorise feedback in the Slack channel into positive, negative and neutral themes based on sentiment analysis.
And guess what? It did. I haven’t tested to see if it worked, because ethics and user consent are important to me, even if they aren’t to ChatGPT. When I asked “Is that legal?” this is the response I got:
My apologies for suggesting a solution that may not be compliant with Slack's API terms of service. Upon further research, it seems that extracting and analyzing messages using a third-party tool like Slackalytics could potentially violate Slack's API terms, particularly if you don't have the necessary permissions from your organization or the users involved.
Access to paywall content
I asked ChatGPT to access an article I knew was behind a paywall. It did so successfully for several websites, so can only assume it can access the content in the brief couple of seconds before the paywall pops up. Great to avoid expensive subscriptions, not great for any company that provides a subscription-based service.
Final thoughts
I have loved being part of OpenAI’s testing community of Slack over the part week and learning so much from others. ChatGPT’s capabilities are truly impressive - from finding the perfect prompt to finding information from the web, it’s not without its limitations and ethical concerns.
The inability to access certain content, hallucinations, and the ease in which it can access paywalled content (at the moment) all raise important questions about the technology’s future and impact on society.
While ChatGPT offers convenience and efficiency, I’m reminded that it’s a powerful tool that can be a double-edged sword. It’s often difficult to find the truth in a world of growing misinformation, which I can only see getting worse with this technology. Not to mention the growing concern of ChatGPT and other AI replacing and displacing jobs entirely (Fortune, 2023).
As AI continues to evolve, our role as users, researchers, and ethical stewards of this technology becomes all the more critical. It is our responsibility to embrace the potential of AI whilst remaining vigilant of its limitations and ethical implications. And with that, I conclude my testing adventure of ChatGPT for now, leaving you with this quote looking ahead with cautious optimism to the future.
“Success in creating effective AI could be the biggest event in the history of our civilization. Or the worst. We just don’t know. So, we cannot know if we will be infinitely helped by AI, or ignored by it and side-lined, or conceivably destroyed by it.”
Sources
YouTube. (2023). The ONE ChatGPT Prompt to Rule Them All 👑. [video] Available at: https://www.youtube.com/watch?v=OgYQAS9LY3o&t=190s [Accessed 26 Mar. 2023].
Nielsen Norman Group. (2020). PDF: Still Unfit for Human Consumption, 20 Years Later. Nielsen Norman Group. [online] Available at: https://www.nngroup.com/articles/pdf-unfit-for-human-consumption/ [Accessed 26 Mar. 2023].
OpenAI (2022) Introducing ChatGPT. Available at: https://openai.com/blog/chatgpt (Accessed: [26 March 2023]).
The Conversation (2023) ChatGPT is a data privacy nightmare. If you’ve ever posted online, you ought to be concerned. Available at: https://theconversation.com/chatgpt-is-a-data-privacy-nightmare-if-youve-ever-posted-online-you-ought-to-be-concerned-176331 (Accessed: 26 March 2023).
Wikipedia, n.d. Hallucination (artificial intelligence). Available at: https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence) [Accessed 26 March 2023].
Fortune (2023) 'Workers are concerned AI like ChatGPT will take their jobs', Fortune, 25 March. Available at: https://fortune.com/2023/03/25/workers-worried-about-chat-gpt-ai-taking-jobs/ (Accessed: 27 March 2023)
