ChatGPT is essentially a large language model (LLM) with an artificial intelligence engine, so if you've used it, you know that it can provide surprisingly complex and human responses. The experience is very similar to chatbots that help you on a hotel website, for example, if something goes wrong with your booking. It draws its information from huge online databases and from its interactions with users, so of course it's also much more sophisticated than these earlier chatbots. It also makes a big difference, and could be the most important benefit for nonprofits, in that it can not only answer questions, but also perform simpler tasks.
What can it help with?
Depending on the data and guidelines you provide, ChatGPT can help you in a relatively wide range of ways. Based on the instructions given and the information entered, you can ask it to create a blog post, a social media post, a timeline or even a meeting reminder. However, because of the specificities described above, it is clear that everything that ChatGPT does needs to be reviewed and looked at with human eyes.
For example, if you don't know how to start a blog post, ChatGPT can help you with framing or even a good lead-in. If you don't have the inspiration for a Facebook post, ChatGPT can help with that too. If you're looking to write a report on your latest research or you've been browsing excel spreadsheets all day and need to put together a project schedule, ChatGPT can make it easy for you.
For those who want to follow the evolution of technology, it's not necessarily easy, as dozens of specialised or generic services based on this model are appearing every day and may be worth trying out. Examples include Microsoft open AI - Chet GPT (the most advanced at the moment) Fotor, Jasper.ai, Chatsonic or BING. – Bard (google)
What can you do to reduce the amount of editing you have to do?
Details and accuracy
ChatGPT works from a total of 4096 tokens - that's roughly 3000 words, so there's room for more detailed instructions. If you want to ask for help with a social media post, for example, you might want to be specific about the topic of the post, its purpose, the platform it is intended for, and even the campaign it will be part of or the image you want to use for the post.
It's also worth bearing in mind that ChatGPT doesn't have the sector-specific knowledge that your organisation's employees are likely to have. To ensure that it can fulfil your request as accurately as possible, you should consider what and how much contextual content it will need. What people, groups, issues does the organisation deal with? What background information might you need about the sector, the project or the organisation itself?
In addition to these, you can also give examples of the tone, structure and wording you would like to use in the post.
Follow the following framing when issuing the task:
who - who am I, from what area is the task coming.
e.g. I am a fundraiser and I work in this and that field.to whom - to whom the material will be addressed.
e.g. in the campaign I am addressing people who...what - what I expect from it.
e.g. create a campaign text in which ...where - where I will use what is produced.
e.g. this campaign will run mainly in the online space ...why - what it will be used for.
our goal is to raise x amount of Fthow - what approach to use.
e.g. the tone should be encouraging, emotionally moving…
Perspective and character
You may have seen some funny examples of ChatGPT responding in the style of an excited teenager or even a movie character. And the "get into character" method can be useful for anyone. For example, if you need to give a presentation to a sponsor or corporate partner, you could start by saying "you are the director of company X" and then go into detail about what points to make and what should be included in the presentation. The more detail you can give, the better.
Think in terms of dialogue
One of the advantages of ChatGPT is that it can be fine-tuned. You don't have to cram everything into one instruction, because it will remember the history of the interactions you have started.
For example, if you want to use it for an application submission, start by providing the context and information you need, and then ask it to give you a 5-point summary of this, which will give you the skeleton of the application. This will give you the opportunity to modify or clarify these, and only then start to elaborate and elaborate.
Formatted responses
Another useful feature is the ability to specify the format in which you want the answer to be sent. If, for example, you need to create a volunteer roster for upcoming events, you can input the volunteer free time slots, locations and time frames when you need people, and the instructions will be to put it all in a tab formatted table. But of course you can try other formats, such as a list formatted with bullet points, to make it easier to interpret and review the data.
The more you use it, the more information and feedback you have available to ChatGPT and AI in general, the more sophisticated and accurate it will become, so it will become increasingly important to be able to give precise instructions. The best way, as with everything else, is to use it regularly so that you can work out what the best methods are.
Which version do we use?
Free: Chat GPT 3.5
Includes data up to 2021.
Paid: Chat GPT 4
This version includes data after 2022.
It can search the web, see, hear and speak.
It works with plug-ins and can do advanced data analysis.
Complementary apps that make your job easier and already use AI:
Midjourney - image editor
Canva - magic - presentation and image editor
Adobe Firefly
Photoshop AI
What is it not good for?
ChatGPT and AI or LLM-based systems in general are not yet suitable for standalone work. The challenges mentioned above are not only true in the non-profit scene, but are a finding that applies to all industries and all sectors. AI may not be fully aware of the context in which you are working, you cannot incorporate years or even decades of experience and knowledge into a single instruction. And even if it can be, it will not be country or region specific, and in many cases it will not be subject specific.
It also follows that while social media texts may be suitable, longer texts may not flow as smoothly, so for blog articles or other longer, single-paragraph texts, it is very important to check and edit the work.
What are the dangers?
Data protection
When using ChatGPT, it is worth bearing in mind that it has not been developed and trained to protect personal and sensitive data, there is no encryption in the program. The chat history is used to develop the language model, but according to the OpenAI privacy policy, this is only monitored for abuse, if you do not allow it to be used for training, it will be deleted after 30 days. However, it is mandatory to provide your name and email address when registering, so it's also worth bearing in mind that the lack of encryption and the huge user database make it an incredibly attractive target for cyber-attacks.
In order to avoid sensitive information falling into the wrong hands, it is best not to provide any additional information that could make you identifiable, other than the above. Wherever possible, be general and do not give out geographical or contact information about yourself, your organisation, volunteers, partners or supporters.
Following laws and regulations
As ChatGPT obtains virtually all its information from the internet, there have been several cases of suspected plagiarism in social media posts, blog posts and other content, as it does not identify the source of such texts. Although OpenAI has taken steps to avoid copying and using one-to-one angles and encourages users to look up specific information themselves, there is still a high chance that the text generated by ChatGPT will be very similar to existing texts.
If you want to be on the safe side, it is best not to use it to generate full texts, but to generate ideas or sketches. If you do need to generate full texts, you can use plagiarism checking software. For these, however, be aware that not all software can handle Hungarian.
It may also be worth mentioning here that many creative professionals and artists object to their work being used to teach AI.
Bias, partiality, prejudice
Specifically for a non-profit organisation, it may be important to note that OpenAI and ChatGPT are primarily designed to provide human responses, not necessarily true or accurate. The development involves publicly available information and reinforcement learning from human feedback (RLHF), both of which are vulnerable to bias and prejudice. Simply put, AI has no values or worldview, so it will not be able to filter out or critique content that is racist, sexist, homophobic or otherwise exclusionary or offensive. In addition, ChatGPT 3.5, which is available for free, will only work with a limited dataset before September 2021. ChatGPT 4.0 has access to more recent data. Also available for free, BING AI uses more recent data.
In any case, it is very important to check, to use multiple sources and to use your own critical thinking when ChatGPT is working on something for you and not to replace the work of a copywriter, researcher or editor.
Use with care!
Like all great technological innovations, artificial intelligence is sharply divisive. Overall, however, ChatGPT can be very useful for finding new ways of doing things, generating ideas and saving time on certain subtasks. The importance of protecting personal data is not at all unusual in the online space, but it is worth paying close attention to. And although it is not new that there are disputes around intellectual property or ownership of creative products, it has perhaps never been dealt with to such an extent and in such an unregulated way. It is also worth noting that people will only be able to take over someone's work if we let them. If we insist that human creativity and work is more valuable than artificial intelligence, then we have nothing to worry about.
*However, there are already examples of ChatGPT refusing to generate content after certain instructions and content have been uploaded because it does not contribute to misleading or provocative content or content that may contribute to the dissemination of false information or disinformation, or uses vulgar or obscene language. Source: https://telex.hu/zacc/2023/10/05/megprobaltuk-megtanitani-a-mesterseges-intelligenciat-jobboldali-moralis-folenybol-ujsagot-irni
This article is based on the following articles: original articles, sources: https://blog.techsoup.org/posts/mitigating-risks-when-using-chatgpt
https://blog.techsoup.org/posts/4-tips-for-better-chatgpt-prompts
https://blog.techsoup.org/posts/how-nonprofits-are-using-ai