ChatGPT Turns One: A History of OpenAI’s Groundbreaking Chatbot
Nov. 30, 2023The past year feels like an AI sprint into a new age of technological revolution. ChatGPT has finally turned one, and over this period, we have seen countless new advancements that have changed our lives. From being a simple chatbot that hallucinated a lot to an advanced and multimodal AI system, ChatGPT has grown tremendously over the course of one year.
So, on the first anniversary of ChatGPT, we take a look at all the progress and hiccups ChatGPT has gone through. From its launch to the current set of features it offers, we have included everything in the article. On that note, let’s begin.
ChatGPT Goes Live
Unlike abstract AI concepts that seemed far away from the reach of general users, ChatGPT finally allowed users to interact with an AI-powered chatbot in a friendly web interface.
ChatGPT was trained to be conversational, and it was built on OpenAI’s GPT-3.5 model. In a way, it was similar to InstructGPT, which followed user instructions to give a response. However, ChatGPT was trained to have adialogue with users, while having context in mind. OpenAI in its openingblognoted that,
ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers
Hallucinationwas achallenging issuewith the initial version of ChatGPT, built on the GPT-3.5 model. However, with the launch of the GPT-4 model (more on this below), OpenAI has managed to reduce factually incorrect responses significantly.
As for the ChatGPT hype, when the service was launched, it seemed the AI chatbot was going to kill Google Search and various other online products. We even pit ChatGPT against Google Search in a YouTube video. And now one year down the line, it feels like this is just the beginning of the AI age and a lot of new product discovery is yet to be made.
ChatGPT quickly rose to fame, and it was especially good at creative tasks such as writing academic papers, composing marketing emails, and even creating misinformation campaigns, etc. Seeing the surge in AI-written text on the web, there came the urgent need for AI plagiarism detectors and text checkers.
So, two months after ChatGPT’s launch, OpenAI released an official AI Classifier tool to help people distinguish betweenAI and human-written text.
However, in July, OpenAI quietly shut down the service citing alow rate of accuracy. If you’re in need of such a tool, you can check our list ofbest AI plagiarism checkers. However, in there, we have clearly mentioned that AI-powered plagiarism tools frequently give false positives and inconsistent results.
Thus, the effort to correctly identify AI-written text still continues, even after a year of ChatGPT’s release.
We also wrote a detailed article on how totrain an AI chatbot with a custom knowledge baseusing the ChatGPT API. The best part about this initiative was that OpenAI gave away$5 worth of APIcredit to all free users. So developers and general users alike used the ChatGPT API to create some interesting projects.
It was one of the watershed moments in the history of LLM development. It was much better than the GPT-3.5 model at combating hallucination. GPT-4 scored more than80% in accuracytests across all categories.
Not only that, GPT-4 was also thefirst multimodal modelthat could understand both texts and images. GPT-4 could analyze scribbled notes on a piece of paper and create a mockup of websites in no time. Sure, the multimodal capability was added later in September (see below), but OpenAI had already cracked the secret sauce way back in March. Mind you, it’s only been 5 months since launch.
Announcing GPT-4 as the first multimodal LLM, OpenAI had already cracked the secret sauce back in March.
Apart from that, the word limit was increased to25,000 wordsfor ChatGPT Plus users whereas the free version of ChatGPT (built on GPT-3.5) had a limit of 8,000 words. Finally, GPT-4 also got better at understanding as many as 26 languages, both major and regional languages from around the world.
So, OpenAI found a way around this problem by launching support forChatGPT Pluginsin March itself. The plugins allowed the LLM tointeract with third-party servicesand even the Internet to fetch real-time data.
With the help of plugins, ChatGPT could access up-to-date information using a Web Browser plugin powered by Microsoft Bing (later renamed Browse with Bing). Moreover, it could run computations using theCode Interpreterplugin.
Code Interpreter was especially helpful as you finally had the ability toupload files to ChatGPTand ask ChatGPT to run various tasksusing a sandboxed Python interpreter. It also brought data analysis and visualization to ChatGPT, and everyone went bonkers. With this tool, you cangenerate graphs, charts, and diagrams in ChatGPT.
Besides that, during the launch, OpenAI launched 11 third-party plugins, including Zapier, Expedia, Instacart, OpenTable, etc., and now hostshundreds of plugins. Using these plugins, you can find and book flights, reserve a table at a restaurant, and do much more.
Basically, with ChatGPT Plugins, web browsing, and Code Interpreter, OpenAI was able to take the initial step towards expanding the LLM’s versatility. Furthermore, it opened up our eyes to the true potential of this chatbot, which was now getting a feature plugin store built on top of it.
March was the busiest month for OpenAI. The company launched its impressive GPT-4 model and expanded its utility via plugins, web browsing, and Code Interpreter. However, apart from showing off all its sparkly new features, OpenAI took ChatGPT offline due to a bug thatleaked users’ chat historyon ChatGPT.
Yes, this was OpenAI’s first major security blunder. The company further noted that some payment-related information may also have been visible to other users, but the affected users were only 1.2%. It quickly fixed the bug and explained the technical details in a bloghere.
Following the incident, OpenAI took some privacy-preserving steps to protect user’s interests. It added an option to ChatGPT toturn off chat historycompletely. After turning off chat history, your chats won’t be used to train and improve OpenAI models.
That said, even though you turn off the history toggle, it isn’t really turned off. OpenAI in itsblog postsays, even after disabling chat history, “we will retain new conversations for 30 days and review them only when needed to monitor for abuse, before permanently deleting.“
The soaring popularity of ChatGPT gave rise to another huge problem. Copycats and low-quality clones. Since ChatGPT was so far limited to a web browser, it gave bad actors a chance to rip off users and make a quick buck.
Now, the ChatGPT app for both Android and iOS has feature parity and supports GPT-4, DALL-E 3, Data Analysis, image analysis, and web browsing.
After a break of two months, OpenAI continued its spree of new feature additions to ChatGPT withcustom instructionsin July. It basically allows ChatGPT to know more about you and what kind of response it should generate. You canfeed in your preferences, and ChatGPT will keep them in mind before responding to your query.
For example, if you ask the chatbot to find a good headphone for under $100, it can consider your location from the custom instructions and give a tailored reply, depending on your location and availability. You can also ask it to cut down on fluff and give a concise,to-the-point response.
Now, even web users can perform image analysis on ChatGPT. The voice chat feature is still limited to the mobile apps, though.
The GPT-4V model has been heavily touted as the frontier model formedical image diagnosis, but OpenAI in its research paper says, “Given the model’s imperfect performance in this domain and the risks associated with inaccuracies, we do not consider the current version of GPT-4V to be fit for performing any medical function or substituting professional medical advice, diagnosis, or treatment, or judgment.“
DALL-E has been OpenAI’s proprietary text-to-image model, but in its initial state, it has not been able to compete againstMidjourney. However, with the latest DALL-E 3 model released in October, OpenAI has come very close to understanding textual nuances and abstract concepts and translating them into accurate images.
Moreover, the need for detailed prompts is eliminated with this integration. ChatGPT can take your basic prompts and turn them into detailed prompts for generating diverse and unique images using DALL-E 3. Follow our step-by-step guide to learnhow to generate images with DALL-E 3.
The GPT-4 Turbo model has not been added to ChatGPT yet, but OpenAI has “plan[s] to release the stable production-ready model in the coming weeks.” Also, the GPT-3.5 Turbo model has also been upgraded and now supports a 16K context window.
Furthermore, the ChatGPT website has been completely overhauled and you don’t need to manually select different options for different tasks when using the GPT-4 model. You can use DALL-E 3, web browsing, and data analysis without switching between options. They are all active and will be called upon as required.
Finally, coming to the highlight of DevDay, we have GPTs. You can now create a custom version of ChatGPT with your custom instructions,custom knowledgethrough PDFs and other documents, and assign special skills to the bot. It’s a powerful new tool that lets you create a tailored version of ChatGPT for your specific task, and you can learnhow to create a personalized GPThere.
OpenAI is rolling out a GPT Store where you can find publicly available GPTs for various purposes.
Just after DevDay, OpenAI announced that it’spausing new sign-ups for ChatGPT Plus. Initially, the company said the surge in traffic has been tremendous, especially after the DevDay event. However, later, it was confirmed that ChatGPT was facing an intenseDDoS attackfrom malicious actors.
As a result, ChatGPT faced a massive outage for several hours. In a statement, OpenAI said, “We are dealing with periodic outages due to an abnormal traffic pattern reflective of a DDoS attack. We are continuing work to mitigate this.” Finally, on November 9, the service resumed after resolving the issues.
Within one year, ChatGPT has undergone incredible change and now hosts the world’s most capable large language model. In addition, it has gotten impressive features like plugins, web browsing, code interpreter, custom instructions, multimodal capability, larger context window, GPTs, and so much more. We’re excited about what OpenAI does next with ChatGPT and its next-gen model, especially with Sam Altman back at the helm. Yeah, back, read more about the chaos at OpenAI over the past couple of weeks in the article linked below.
While ChatGPT has gotten pretty powerful, let’s be clear that this is “just the beginning” before the next explosion of AI. Many experts believe that a dialogue-based, conversational chatbot is not the only interface to interact with Generative AI.
In the future, we may have immensely powerful multimodal AI systems that can understand text, voice, visual elements, ambient data, andsensory data from human brain activity to draw contextual inferencesand perform actions in real time.
Apart from that, AI can be tightly integrated into AR and VR interfaces to create a highly interactive and immersive experience. Also, beyond AI chatbot integration into apps, smartphones, and devices will runAI systems at the OS levelto understand and perform tailored actions automatically.
From managing emails to creating schedules based on your usage and priority, AI can automatically do things for you. Qualcomm has alreadyexplainedhow personalized, on-device AI will work in the future.
Not to forget, OpenAI and many other companies are alreadyworking onAGI(Artificial General Intelligence) that can surpass human capabilities that require intellectual input and reasoning.
In fact, the ChatGPT-maker is already working on a solution to overcome training data limitations by generating high-quality synthetic data. Whether it’s connected to the ongoing discussion on the talk-of-the-town that isProject Q*is yet to be established.
Overall, I would say that ChatGPT as a conversational chatbot has gone from being limited to having quite the many applications, sitting at the top like a king. Still, if we zoom out and look at the bigger picture, it can feel limited, and we may see countless new forms of AI experiences in the future.
Passionate about Windows, ChromeOS, Android, security and privacy issues. Have a penchant to solve everyday computing problems.
