97+ ChatGPT Statistics & User Numbers in June 2023 (New Data)

ChatGPT userbase grew by an explosive 9,900% in just 60 days of launch, taking the whole world by a storm.

Here, I’ve added the most up to date user statistics and facts about ChatGPT, the AI chatbot launched by OpenAI, with sources. Let’s get started!

ChatGPT Stats Overview

Note: Some websites are wrongly equating website visits to user count. I have clearly mentioned the distinction wherever possible.

ChatGPT User Stats

• ChatGPT crossed 1 million users in just 5 days of launch and gained 100 million active users by January 2023.
• Thus, ChatGPT users grew by 9,900% in 60 days setting the record for the fastest-growing platform in history!
• Applying linear regression, ChatGPT had 173 million active users in April 2023.
• ChatGPT’s website chat.openai.com saw a 576.6% increase in visits from December 2022 to April 2023. ChatGPT got 1.8 billion visits in April 2023 which is 7 times higher than the number of visits in December 2022 (266 million).

• The highest percentage of ChatGPT users belong to USA (15.22%), followed by India (6.32%), Japan (4.01%) and Canada (2.75%).
• Each user spends an average of 8 minutes and 32 seconds on the website.

Sources: Reuters, Similarweb

ChatGPT Updates

Since its launch, Open AI has been continuously adding new features and the model is also updated frequently. Hence users can expect to see changes in interface or even in quality of responses.

GPT-4 & ChatGPT Plugins Update

• OpenAI has launched GPT-4, a successor to ChatGPT’s GPT 3.5 model. GPT-4 understands both images and text input and beats GPT 3.5 at every task (including passing US Bar Exam).
• Read our coverage of GPT-4 statistics, abilities, and limitations.
• ChatGPT plugins have been launched by OpenAI, enabling ChatGPT to talk to other apps like search engines, code interpreters, food apps, etc.
• This effectively connects ChatGPT to the internet and the world and could mark beginning of an AI Appstore.

Users Can Now Disable ChatGPT History

• ChatGPT now allows users to opt out of their chats being used for training ChatGPT.
• Click Settings > Show and toggle off the Chat History & Training setting to turn off conversation history.

About Open AI and Investments

• Open AI created ChatGPT. Open Al also created Dall-E, an image-generation AI.
• Open AI was co-founded by Elon Musk and is based out of San Francisco (HQ: Pioneer Building in Mission District, San Francisco).
• OpenAI has closed a $300 million funding round at a valuation between $27 billion and $29 billion.
• Microsoft has already invested $1 Billion and confirmed an additional investment of $10 billion in Open AI. Bing got a 15% Daily Traffic Bump from ChatGPT Integration.

Cost to Run and Revenue Model

• The average cost of a single chat with ChatGPT is likely in single-digit cents. (Sam Altman’s Twitter, Sam Altman is CEO of OpenAI)
• Tom Goldstein, an AI ML Professor at Maryland University, has estimated the daily cost of running ChatGPT to be approximately $100,000 and the monthly cost to be $3,000,000 (3 Million USD). His estimates are based on Azure Cloud costs (server infrastructure on which ChatGPT runs).
• He estimates the ChatGPT model has 175 Billion parameters and receives 10 Million queries per day assuming 1 million users make an average of 10 queries/user. (Tom Goldstein’s Twitter)
• Extrapolating with January 2023 user’s count of 100 million (source: Reuters), it is estimated that ChatGPT gets 1 billion queries per day.
• ChatGPT is completely Free to use during its research preview.
• ChatGPT Plus has been recently launched at a pilot subscription price of $20.
• Open AI has added “Turbo” mode for ChatGPT plus users. It gives faster responses.
• ChatGPT once monetized is predicted to earn a revenue of $200 million by end of 2023 and $1 billion dollars by the end of 2024. (Reuters)

Model Training

• ChatGPT is a fine-tuned model based on GPT-3.5 (text-DaVinci-003).
• Text-DaVinci-003 builds on an earlier open AI model called InstructGPT which is better at understanding and answering human requests than GPT 3.
• An improved version of the ChatGPT model is powering Bing AI. It’s called Prometheus.
• GPT-4 (available in ChatGPT Plus) has shown “Sparks of AGI” and scored in 90th percentile of several hard human examinations like US BAR exam, International Biology and Math Olympiads.

What was the dataset size for training ChatGPT?

• ChatGPT was trained on a dataset of 300 billion words. The dataset was 570 GB in size and consisted of crawled web data, books, Wikipedia, etc. (Analytics India Magazine)

How was ChatGPT trained?

• ChatGPT is a Large Language Model (LLM) trained using Reinforcement Learning from Human Feedback (RLHF)
• RLHF means that during initial model training, human AI trainers rated ChatGPT’s responses as “correct” or “wrong”, helping it improve its responses.

Limitations of ChatGPT

While ChatGPT can help you do fun things like writing movie scripts, launching your own online course, or writing AI-generated stories, it does have its limitations.

• ChatGPT Cannot Access the Internet on its own.

Can ChatGPT read links? No, as an AI language model, ChatGPT cannot access the internet or external links. It can only provide information and answer questions based on the text it was trained on.

Source: ChatGPT

But, with the recent launch of browsing plugin, ChatGPT will be able to search the web for information. The ChatGPT browsing plugin is only available to limited users.

• ChatGPT’s knowledge is limited to its training data which has the cutoff year of 2021.

• ChatGPT can give wrong information. It has a tendency to sound confident despite being incorrect in some cases. Users are advised to fact-check.
• All LLM models are prone to such inaccuracies including alternatives like Bard, Claude, Open Assistant, etc.

Real-World Impact

ChatGPT has passed the US Med licensing exam, law school exams, and Wharton’s MBA exam. Unfortunately, it is also being used for some nefarious purposes like generating hacking scripts, phishing emails, etc.

Job & Labor Market Impacts

GPT-4 improves upon creative and cognitive capabilities of ChatGPT and it is predicted that over 300 million jobs will be impacted by GPT models.

CyberSecurity Threat

• ChatGPT is being used to write convincing phishing messages and create malicious code (even without prior programming knowledge by the prompter). (Source: TechCrunch)
• For example, ChatGPT can be asked to write a medical advice article in the “style of a doctor” or craft legal documents that look “legit” (but aren’t actually written by a lawyer).
• ChatGPT can also “imitate” the writing style of famous personalities.
• It can also generate working code for adversarial attacks, ransomware attacks, and other exploits when given a simple English prompt. This will arm potential hackers who earlier were limited by their programming knowledge.
• 51% of IT professionals predict that we will witness a successful cyberattack done with the help of ChatGPT by the end of this year. (Blackberry Research)
• ChatGPT has already been used to create an info stealer, an encryption tool, and for dark web malware scripts.

Threat to Higher Education

• ChatGPT can be used to create convincing essays and research papers. While tools to detect AI content exist but they are not very reliable.
• Thus, the potential for misuse in academia is a lot: students can cheat by submitting assignments written by AI and pass without acquiring any real knowledge.
• In the near future, Open AI has said that it plans to “watermark” content generated by ChatGPT to prevent academic fraud and spam.
• Springer (a research publisher) has asked researchers
  • not to credit ChatGPT as an author (as a bot cannot be held accountable)
  • to Document the use of chatGPT in the acknowledgments or methods section.

Other Reactions

• Stack Overflow (a popular coding question and answer platform) banned the use of ChatGPT because it can be used to give legit-looking but actually useless/wrong answers. This makes it harder to find relevant answers that actually work.
• Amazon has warned employees not to share sensitive information with ChatGPT. (Source)
• Artificial Intelligence tokens like DeepBrain Chain and Numeraire witnessed a sharp rise in prices after debuting of ChatGPT. According to CoinGecko, prices of AI tokens increased by up to 77%. (Coingecko)
• Italy had banned ChatGPT citing concerns back in Jan. On April 29, 2023 Italy unbanned ChatGPT and it is now accessible again.

Apps Based On ChatGPT

• Bing AI uses a more powerful version of ChatGPT called Prometheus.
• The Indian government is planning to integrate ChatGPT into an educational tool to help students.
• Lexi is India’s first AI-based e-commerce chatbot app that uses ChatGPT.
• Microsoft plans to add ChatGPT-like AI to office apps later this year.
• A business owner on Reddit posted a video of an internal tool they had created based on ChatGPT. It gave responses like ChatGPT but based on company information.
• TweetGPT is a chrome extension that writes tweets for you using ChatGPT.
• Promptheus is a chrome extension that takes voice input and sends it as text prompt to CHatGPT, returning with a response.
• YouTube Summary With ChatGPT summarizes Youtube videos from transcriptions.
• Merlin is a ChatGPT based toolto create replies, shorten the text, provide summarised content and more.
• ShareGPT is an app to share chatGPT prompts with other users.
• ChatGPT writer is an extenstion to create AI replies to emails using ChatGPT.
• ChatGPT for Google collects answers to your search query from across Bing, Google, and DuckDuckGo (again uses Bing).
• Engage AI can read any Linkedin post and generate a comment on your behalf based on the content contained in the post. It is based on ChatGPT.
• GPT-4 is being used in Khan Academy, Duolingo, Stripe, BeMyEyes and many more such apps.

Alternatives & Competitors of ChatGPT

ChatGPT vs Bing vs Bard

Read Bing Statistics for 2023 – learn how Bing is dominating AI search since the arrival of Bing Chat.

Google Vs ChatGPT

Google has launched an AI search assistant called Bard AI. Sundar Pichai, CEO Google, shared a short clip of Bard AI in action.

Bard AI’s demo response was found incorrect and led to Google losing over $100 billion in market cap!

Bard responds with a number of answers, including one suggesting the JWST was used to take the very first pictures of a planet outside the Earth’s solar system, or exoplanets. This is inaccurate.

The first pictures of exoplanets were taken by the European Southern Observatory’s Very Large Telescope (VLT) in 2004, as confirmed by NASA.

Reuters

Background: AI and Search

• ChatGPT since its launch has shown the ability to give direct responses to most user queries with the chance to ask specific follow-up questions. This is an area Google falls behind.
• Google issues a code red on the release of ChatGPT.
• Google Lamda: First demoed in 2020, LAMDA is a cutting-edge language model at the forefront of Google’s AI research.
• It could converse with the user on a variety of topics, make jokes, write poems, etc much like ChatGPT today.
• However, Google had been hesitant to push LAMDA to users or make it part of its core search service as AI models are prone to inaccuracies (ChatGPT, now Bard and Bing are all guilty of sounding confident while being incorrect).
• Google PaLM: PaLM is another language model already demoed by Google. It has been trained on 540 billion parameters (in comparison to only 175 Billion parameters of ChatGPT).

Microsoft and ChatGPT

Microsoft has been a step ahead of Google in launching an AI search assistant. Several users have already been able to preview Bing AI “the co-pilot for the web”. See examples of Bing AI’s responses.

Which model is behind Bing AI?

…A new, next-generation OpenAI large language model that is more powerful than ChatGPT and customized specifically for search. It takes key learnings and advancements from ChatGPT and GPT-3.5 – and it is even faster, more accurate and more capable.

Techcrunch

Later Microsoft confirmed that they have been using GPT-4 in Bing Chat.

Youchat

• Like ChatGPT, Youchat can hold conversations with its users. Unlike ChatGPT, Youchat is connected to the internet.

Wordtune

• One of the problems with ChatGPT is its tendency to give statistics and definitions without sources and citations. Wordtune is a conversational model that fixes this by providing citations.

Others

• Amazon is working on ALexTM 20B model.
• Baidu, a chinese company is also developing an AI chatbot like ChatGPT but their demo was lackluster with pre-recorded presentation.
• Chinese researches also released ChatGLM – a general language model like ChatGPT optimized for Chinese question-answering.
• Anthropic’s Claude is another ChatGPT competitor (bought by Google now).
• AUTOGPT is an autonomous GPT AI which takes an input goal from the user and then autonomously plans and executes the tasks required to achieve it.

Open-Source ChatGPT Alternatives

In this list, only Open Assistant has Apache 2.0 license (that allows commerical usage). Other alternatives used ChatGPT in training and cannot be used for purposes other than personal use or research.

Open Assistant

• Open Assistant is a truly free ChatGPT alternative with open-source/Apache 2.0/commercial usage license.
• Open Assistant understands tasks, can interact with third-party systems, and retrieve information dynamically to do so.
• It is an open-source project meant to give everyone access to a great chat-based large language model.
• The project aims to build the assistant of the future, able to not only write email and cover letters but do meaningful work, use APIs, dynamically research information, and much more, with the ability to be personalized and extended by anyone.

Stanford Alpaca

• Alpaca is a seven-billion parameter variant of Meta’s LLaMA. It was fine-tuned using 52,000 instructions generated by GPT-3.5 (ChatGPT). (Much like the way, human testers were used to fine-tune ChatGPT, Stanford used the model behind ChatGPT to train their Alpaca AI.)
• In tests, Alpaca performed comparably to OpenAI’s model but produced more hallucinations.
• Alpaca is significant because it proved that building and training new AI models can be insanely cheap. (Alpaca was trained using just $600)

Vicuna

• Vicuna-13B, an open-source chatbot trained by fine-tuning LLaMA.
• It was evaluated by using GPT-4 as a judge: Vicuna-13B achieves more than 90% quality of OpenAI ChatGPT and Google Bard while outperforming other models like LLaMA and Stanford Alpaca in more than 90% of cases.
• The cost of training Vicuna-13B is around $300.
• The Vicuna code is at available on Github for non-commercial use.

Dolly 2

• Databricks released a completely open source model Dolly 2 which has been trained by their own employees.

Wrap Up

ChatGPT has taken the world by storm and has ushered in a new era of AI-driven assistants.

People around the world are using ChatGPT in fascinating ways to make money, create courses, write movie scripts, be more productive, boost their creativity, etc.

If you’re interested in cool applications of AI, you might also like my experiments with AI voice generators and AI story assistants.

Frequently Asked Questions