One year with ChatGPT: my uses, developments and reflections
Greg Taieb · · 6 min read
On November 30, 2022, a company relatively unknown to the general public (unless you were listening my podcast), OpenAI, created a shock wave by launching ChatGPT. For those unfamiliar with OpenAI, it is an organization committed to developing artificial intelligence (AI) for the benefit of all.
A major innovation, not just an update
Although ChatGPT may seem like a packaged version of GPT-3 technology that was already two years old in its active, he represented much more than that. Before this release on the market, GPT-3 was accessible via APIs, and a few copywriting applications had developed around this technology, but its use remained largely niche.
The genius of OpenAI was twofold:
- Offering a conversational interface similar to those we use every day, reinforcing the “human” aspect of the interaction.
- Opening ChatGPT to everyone so that everyone can do it the experience.
A record adoption
The adoption statistics are there to bear witness to this: only 2 months to reach 100 million users, when it took TikTok 9 months and Instagram 2 years. But what has made ChatGPT such a phenomenon?
- Natural Language Understanding: ChatGPT has been trained on a wide range of texts available on the internet, allowing it to understand and generate responses in natural language that is often fluent and coherent.
- Versatility: Capable of discussing a variety of topics, solving problems, writing creative texts and much more, ChatGPT finds its usefulness in many different contexts.
- Conversational Approach: Unlike traditional search engines, ChatGPT allows an iterative approach to refine, deepen or modify a query based on previous interactions, making it easier to obtain precise answers.
How have I been using ChatGPT for 1 an
When I started writing this part I said to myself that I would need an entire article dedicated to my use of ChatGPT as this application has become a reflex for me on a daily basis.
ChatGPT has in fact largely supplanted Google for many use cases. I have also subscribed to the paid offer which has some very practical advantages such as access to GPT 4 and the beta functionalities which I will talk about a little further down.
Here are some of my use cases:
- Plan a trip from A to Z (route, districts to choose for the hotel, places to visit, order of visit, restaurants, etc.)
- Create a detailed structure of a document (plan type of com, call for tenders,...)
- Write advertisements for Vinted
- Summarize a text, meeting notes
- Translate a text into several languages
- Create exercises and their answers to help children with their homework
- Write an email or a text
- Reread and improve an article or newsletter before publication
- Create a macro in Excel to meet a very specific
- Analyze Excel files (like the stats of this site) to identify trends and lessons
Challenges remain
With everything I have just said you could say that ChatGPT is a free little genius that you would be crazy not to use.
And yet I did the test, passed the effect wow, few people around me use it regularly.
There are several good reasons for this:
- Lack of use cases: Lack of practice, difficult to know everything you can do with ChatGPT
- Privacy and security concerns: In the free version, OpenAI is murky about data privacy and how it is used by AI systems to improve.
- Reliability: AIs don't think, they generate responses according to a probability model. As a result, they can tell big bullshit.
So we see that between testing a solution and adopting it in your daily life there is a gap but that could change quickly.
The boom in uses in business is slow in coming...
The reasons mentioned above partly explain the reason why the uses struggle to take off in business. Indeed, it's not easy to trust an AI that can potentially share my data and hallucinate.
Add to that the legitimate uncertainties about employment and the transformation of certain professions.
Overall the use cases that work well are those that do not require feeding the AI with sensitive data:
- Create content for social networks
- Write or translate content for websites
- Questioning public knowledge bases
...But the solutions are coming
All of this is fortunately changing for several reasons.
First of all, in recent months we have seen LLM (Large Language Models) alternatives arrive that can be installed on internal servers. I am thinking for example of Zephyr the model developed by the French Mistral.ai whose version 7B seems very promising.
But the real accelerator risks coming from Microsoft which invested 10 billion in OpenAI at the end of January 2023 but above all made it the pillar of its software strategy: integration in Bing, in Windows 11, in Paint but especially in Office with Copilot officially launched on November 1st. https://x.com/petergyang/status/1720956604676485292?s=46
It's quite simply the best integration of AI into the tools in which millions of people work every day.
AI and its contributions will become very concrete thanks to the integration of GPT into the tools of the Office suite: creation of a PowerPoint, responses to emails, analysis of Excel files,... all tasks which will be facilitated by this integration.
Important clarification: the data shared with OpenAI as part of the agreement with Microsoft 365 are not used to train the model.
How ChatGPT has changed in 1 year
And since we're taking stock, it's the perfect opportunity to recall how much ChatGPT has changed in 1 year.
It's very simple almost every month given the arrival of a major feature:
- Version paid
- Release of GPT4
- Support for plugins to extend the functionalities
- Release of the mobile application
- Integration of Bing to have real-time results from the web
- Audio assistant in Siri mode to interact only by voice
- Multi modal: allows you to rely on a photo, an image or a document for a query
The capabilities of ChatGPT have increased tenfold in the space of a year by allowing a greater variety in prompt formats (text, document, audio, image, etc.) and output.
But that could only be the beginning. Indeed the next move will consist of allowing anyone to create THEIR ChatGPT with their set of data, their tone and to share it via a store.
If these leaks are real, OpenAI will let you create and share your own GPT starting tomorrow:
— Peter Yang (@petergyang) November 5, 2023
You can give your GPT:
- A name
- A welcome message
- Example prompts to start the conversation
- Additional files for the GPT to reference
- Actions like web browsing, DALL-E, code… pic.twitter.com/9VJMLy0Z3o
ChatGPT has redefined the way we interact with technology, making AI more accessible and useful in everyday life.
And you, have you adopted ChatGPT? What tasks does it help you accomplish? Share your experience in the comments below!