The past year has been an exhilarating sprint into a new era of technological transformation. ChatGPT has reached its one-year milestone, marking a period of immense advancements that have significantly impacted our lives. Transitioning from a simple chatbot that occasionally made errors to an advanced and multimodal AI system, ChatGPT has undergone remarkable growth in just a year.
As we commemorate the first anniversary of ChatGPT, we delve into its evolution, exploring both its progress and challenges, from its initial launch to the current array of features it offers. Let’s embark on this retrospective journey.
ChatGPT Goes Live
On November 30, 2022, OpenAI introduced ChatGPT to the public through a research preview, offering it for free. Surprisingly, within a mere two months, it attracted over 100 million monthly users, establishing itself as the world’s fastest-growing service, even surpassing the growth rate of Facebook.
Unlike abstract AI concepts that felt distant from everyday users, ChatGPT brought AI interaction within reach through a user-friendly web interface, enabling individuals to engage with an AI-powered chatbot in a more accessible manner.
ChatGPT, developed on OpenAI’s GPT-3.5 model, was designed to engage in conversations. It shared similarities with InstructGPT, which responded based on user instructions. However, unlike InstructGPT, ChatGPT was trained to engage in dialogues with users while considering context. OpenAI mentioned in its initial blog that:
ChatGPT generates responses that may sound plausible but could be incorrect or nonsensical.
The initial iteration of ChatGPT, based on the GPT-3.5 model, grappled with the challenge of generating inaccurate information, often referred to as “hallucination.” However, with the introduction of the GPT-4 model (more details below), OpenAI has notably reduced instances of factually incorrect responses.
At the time of its launch, there was considerable hype around ChatGPT, sparking discussions about its potential to surpass Google Search and other online products. Some even created comparisons between ChatGPT and Google Search in YouTube videos. Now, a year later, it feels like we’re merely scratching the surface of the AI era, with numerous opportunities for new product discoveries and advancements awaiting exploration.
AI Classifier Launched to Detect AI-written Text
ChatGPT gained rapid popularity for its adeptness in creative tasks like crafting academic papers, composing marketing emails, and even generating misinformation campaigns. With the proliferation of AI-generated content across the web, the necessity for AI plagiarism detectors and text verifiers became urgent.
Consequently, two months after ChatGPT’s launch, OpenAI introduced an official AI Classifier tool to aid in distinguishing between AI-generated and human-written text.
However, in July, OpenAI quietly discontinued the service, citing low accuracy rates. Those seeking similar tools can refer to curated lists of the best AI plagiarism checkers available. Notably, these tools often produce false positives and inconsistent outcomes, as mentioned in our compilation.
Hence, even a year after ChatGPT’s release, the quest for accurately identifying AI-generated text persists, with ongoing efforts to enhance detection capabilities.
ChatGPT Plus Subscription Launched
In February 2023, four months post its release, OpenAI capitalized on the buzz around ChatGPT by introducing its inaugural subscription plan named ChatGPT Plus, priced at $20 per month.
Upon its launch, ChatGPT+ provided users with uninterrupted access to the chatbot, even during peak usage periods, ensuring faster response times. Subscribers were also promised early access to forthcoming features and enhancements in the subsequent months. Initially, the subscription was exclusively offered to customers in the US but was later expanded to encompass users globally.
ChatGPT API Released for Developers
In March 2023, OpenAI unveiled the ChatGPT API, granting developers access to the chatbot’s robust capabilities. This API empowered developers to craft diverse products harnessing ChatGPT’s functionalities. Among the early adopters, Snapchat utilized the API to devise a customizable chatbot capable of offering suggestions and crafting haikus for friends.
Additionally, we authored an in-depth article detailing the process of training an AI chatbot with a personalized knowledge base using the ChatGPT API. Notably, OpenAI’s initiative included providing free users with $5 worth of API credit. This gesture encouraged both developers and general users to leverage the ChatGPT API, resulting in the creation of numerous intriguing projects.
ChatGPT Gets Upgraded to GPT-4 Model
In March 2023, OpenAI marked a significant leap in the evolution of its chatbot by introducing its next-generation model. This innovation arrived in the form of GPT-4, acclaimed as the world’s most powerful large language model, accessible to ChatGPT Plus users right from its launch.
The debut of GPT-4 stood as a pivotal moment in the development of Large Language Models (LLMs). It exhibited remarkable improvements over the GPT-3.5 model, notably excelling in addressing the issue of hallucination. GPT-4 achieved an accuracy rating exceeding 80% across all test categories, signifying a substantial advancement.
Furthermore, GPT-4 established itself as the inaugural multimodal model capable of comprehending both text and images. Even though the multimodal capability was formally introduced later in September (as mentioned below), OpenAI had already achieved this feat as early as March. Remarkably, this groundbreaking advancement took place a mere five months after its launch.
OpenAI’s announcement of GPT-4 as the pioneering multimodal Large Language Model (LLM) indicated a breakthrough achieved as early as March.
Additionally, the word limit for ChatGPT Plus users was expanded to 25,000 words, a substantial increase compared to the free version of ChatGPT (based on GPT-3.5), which maintained an 8,000-word limit. Furthermore, GPT-4 exhibited enhanced proficiency in understanding a spectrum of 26 languages, encompassing both major and regional languages worldwide.
However, GPT-4’s knowledge base was limited to events up to September 2021. Nevertheless, GPT-4’s unveiling marked a significant turning point in AI development, setting a benchmark that both proprietary and open-source LLMs endeavor to surpass.
ChatGPT Plugins Debut
As mentioned earlier, while GPT-4 boasted impressive capabilities, its knowledge was constrained to events until September 2021. Moreover, its enclosed nature restricted its interaction with the external world, a significant limitation for an AI chatbot with boundless potential.
To circumvent this issue, OpenAI devised a solution by introducing ChatGPT Plugins in March. These plugins facilitated the LLM’s interaction with third-party services and the internet, enabling access to real-time data.
Through these plugins, ChatGPT gained the ability to retrieve current information via a Web Browser plugin initially powered by Microsoft Bing, later renamed Browse with Bing. Additionally, it could perform computations utilizing the Code Interpreter plugin.
The introduction of the Code Interpreter proved to be a game-changer, allowing users to upload files to ChatGPT and execute various tasks through a sandboxed Python interpreter. This feature not only enabled data analysis and visualization within ChatGPT but also garnered widespread excitement. Users were now able to generate graphs, charts, and diagrams directly within the chat interface.
Simultaneously during the launch, OpenAI unveiled 11 third-party plugins, including Zapier, Expedia, Instacart, OpenTable, among others, which have now expanded to hundreds. These plugins empowered users to accomplish tasks such as flight bookings, restaurant reservations, and more, directly within the ChatGPT interface.
In essence, through the integration of ChatGPT Plugins, web browsing capabilities, and the Code Interpreter, OpenAI took the initial strides toward enhancing the versatility of the LLM. This initiative not only showcased the true potential of the chatbot but also laid the groundwork for a feature-rich plugin store atop ChatGPT.
ChatGPT Chat History Leak
In March, OpenAI experienced its most active period, marking the launch of the remarkable GPT-4 model and the expansion of its functionalities through plugins, web browsing, and the Code Interpreter. However, amid the showcase of its new features, OpenAI faced a critical setback when it had to temporarily take ChatGPT offline due to a bug that resulted in the leakage of users’ chat history.
Regrettably, this incident represented OpenAI’s initial significant security lapse. The company acknowledged that the bug may have exposed certain users’ payment-related information, affecting approximately 1.2% of the user base. OpenAI promptly addressed the issue by rectifying the bug and provided a detailed explanation of the technical intricacies in a dedicated blog post.
Manage Your Data in ChatGPT
In response to the incident, OpenAI implemented measures to safeguard user privacy. They introduced an option within ChatGPT that allows users to completely disable chat history. By toggling this feature off, users could prevent their chats from being utilized for training and enhancing OpenAI models.
However, it’s essential to note that despite deactivating the chat history toggle, the history isn’t entirely disabled. OpenAI clarified in its blog post that even with chat history disabled, “we will retain new conversations for 30 days and review them only when needed to monitor for abuse, before permanently deleting.” This approach ensures a level of oversight while preserving user privacy by deleting conversations after a specified period.
OpenAI Introduces the ChatGPT App for iOS
The escalating popularity of ChatGPT gave rise to a significant issue: the proliferation of copycats and low-quality clones. As ChatGPT was initially confined to web browsers, it provided an opportunity for unscrupulous actors to deceive users and exploit the situation for monetary gain.
This issue was particularly prevalent on mobile platforms, where users encountered dubious apps and clones laden with malware. To combat this problem, OpenAI took action in May 2023 by launching a native ChatGPT app exclusively designed for iOS, developed from the ground up.
The iOS app, distinct from a mere web wrapper, introduced support for chat history, voice input, and incorporated the GPT-4 model for ChatGPT Plus users using iPhones. Although the Android app was not concurrently released, OpenAI assured users of its imminent arrival. Subsequently, in July 2023, OpenAI rolled out the ChatGPT app for Android devices.
Both the Android and iOS versions of the ChatGPT app now boast feature parity, offering support for GPT-4, DALL-E 3, Data Analysis, image analysis, and web browsing functionalities.
OpenAI Disables Browse With Bing Plugin
In early July 2023, OpenAI encountered a significant hurdle when its “Browse with Bing” plugin began displaying content from paywalled websites within ChatGPT. Fearing potential legal repercussions from these publications, OpenAI swiftly disabled the feature, acknowledging that it “occasionally presented content in unintended ways.”
The deactivation of this crucial plugin left ChatGPT+ users without an official means to access the internet and provide real-time information to the chatbot. However, users could still utilize third-party plugins to establish connections between ChatGPT and the internet. Finally, after a three-month hiatus, in September 2023, the “Browse with Bing” feature was reinstated on ChatGPT following the necessary corrective measures.
Custom Instructions Come to ChatGPT
In July, OpenAI resumed its pattern of introducing new features to ChatGPT, this time unveiling the capability for custom instructions. This feature empowers ChatGPT to gather more information about users and tailor its responses accordingly. Users can input their preferences, enabling ChatGPT to consider these specifications before generating a response.
For instance, if a user queries the chatbot for recommendations on a quality headphone priced under $100, ChatGPT can utilize the information provided in the custom instructions, such as the user’s location, to offer a personalized response based on local availability. Additionally, users can prompt ChatGPT to deliver succinct and direct responses by requesting a reduction in unnecessary details.
ChatGPT Can See, Hear, and Speak
In September 2023, OpenAI introduced a significant upgrade to ChatGPT, initially revealed on its iOS app. This update introduced two major functionalities: the ability to engage in voice conversations with the model and the capability to upload images for analysis within ChatGPT.
Essentially, the multimodal capabilities previously showcased in March were integrated into ChatGPT, employing its Whisper (speech recognition model) and GPT-4V (GPT-4 Vision model). While initially available on the iOS app, this update was later rolled out to both iOS and Android users around November 2023.
Presently, web users can also leverage image analysis within ChatGPT. However, the voice chat functionality remains exclusive to the mobile apps.
GPT-4V (Vision) Released
In the same month, OpenAI published a dedicated research paper outlining the workings of their newly developed GPT-4V system. This system, an extension of GPT-4 with incorporated vision capabilities, is designed to interpret uploaded images within ChatGPT and articulate visual information in natural language. Trained in 2022, its development involved rigorous measures to prevent the model from engaging in harmful activities such as CAPTCHA breaking, person identification, gender and race recognition, or illicit behavior.
Despite expectations regarding its potential in medical image diagnosis, OpenAI clarified in the research paper that due to the model’s imperfect performance in this domain and the associated risks of inaccuracies, the current iteration of GPT-4V isn’t deemed suitable for undertaking any medical functions or replacing professional medical advice, diagnosis, treatment, or judgment.
Dall-E 3 Released to ChatGPT
In October, OpenAI introduced DALL-E 3, a significant upgrade to its text-to-image model. This latest iteration marked a substantial leap forward, edging closer to comprehending nuanced textual cues and abstract concepts to produce more precise images, rivaling the capabilities of models like Midjourney.
An exciting development accompanied DALL-E 3’s release, its integration directly into ChatGPT. Subscribers of ChatGPT Plus gained access to this impressive feature, which effortlessly generates high-quality AI images without the necessity for visiting external websites.
One notable enhancement with this integration is the elimination of the need for highly detailed prompts. ChatGPT can now transform basic prompts into intricate instructions, enabling the generation of a wide array of diverse and unique images using the power of DALL-E 3. Check out our step-by-step guide for harnessing the potential of DALL-E 3 within ChatGPT.
GPT-4 Turbo Model and GPTs Announced
In November 6, 2023, OpenAI held its inaugural developer conference, DevDay, in San Francisco. During the event, the company unveiled significant updates, notably introducing the new GPT-4 Turbo model. This enhanced iteration boasted an expanded knowledge base, encompassing world events up to April 2023. Its context window was significantly enlarged to 128K, a substantial leap from the previous 8K and 32K token limits of GPT-4.
A remarkable aspect of the GPT-4 Turbo model was its cost-effectiveness, priced at three times lower than the standard GPT-4 model for input tokens. This reduction in operating costs made it an exciting prospect, particularly considering the high expense associated with running the original GPT-4.
Although the GPT-4 Turbo model hasn’t yet been integrated into ChatGPT, OpenAI disclosed “plan[s] for its imminent release as a stable, production-ready model in the forthcoming weeks“. Additionally, the GPT-3.5 Turbo model received an upgrade, now supporting a 16K context window, enhancing its capabilities.
The ChatGPT website underwent a comprehensive overhaul, streamlining the user experience by eliminating the need to manually select different options for various tasks when utilizing the GPT-4 model. With this update, functionalities like DALL-E 3, web browsing, and data analysis remain active concurrently and are summoned as required, removing the need for constant switching between options.
At DevDay, the spotlight was on GPTs, introducing the capability to craft personalized versions of ChatGPT. This advanced tool empowers users to create custom instructions, integrate specialized knowledge from PDFs and other documents, and imbue the bot with unique skills tailored to specific tasks. For those interested in developing their personalized GPT, a comprehensive guide on creating customized GPTs is available.
OpenAI is also launching the GPT Store, an online repository where publicly available GPTs for diverse purposes will be accessible, further enhancing the versatility and applicability of these models.
OpenAI Stops New ChatGPT Plus Sign-ups
After DevDay, OpenAI took a surprising step by temporarily halting new sign-ups for ChatGPT Plus. Initially citing a substantial surge in traffic following the event, the company later disclosed that the pause was a response to a significant Distributed Denial of Service (DDoS) attack aimed at ChatGPT by malicious entities.
This aggressive cyber assault led to an extensive outage of ChatGPT services for several hours. OpenAI acknowledged the disruption, stating, “We are contending with intermittent service disruptions due to an abnormal traffic pattern indicative of a DDoS attack. Efforts are ongoing to mitigate this issue.” Finally, on November 9, the service was restored after the problems were resolved.
In just a year, ChatGPT has undergone remarkable transformations, hosting the world’s most capable large language model. The platform introduced impressive features like plugins, web browsing, code interpreter, custom instructions, multimodal capabilities, a larger context window, GPTs, and more. The future developments of ChatGPT and its next-gen model under the guidance of Sam Altman, who has recently returned to the helm at OpenAI, are highly anticipated, given the recent turbulence within the organization. For further insights into the chaos at OpenAI in recent weeks, explore the linked article below.
What the Future Holds for ChatGPT?
While ChatGPT has made significant strides, it’s important to acknowledge that this is just the beginning of the next phase in AI evolution. Many experts envision a future where dialogue-based chatbots represent only one facet of interacting with Generative AI.
The next generation of AI systems is anticipated to be incredibly powerful, embracing multimodal capabilities to comprehend text, voice, visual inputs, ambient data, and even interpret sensory data from human brain activity. These systems will adeptly draw contextual inferences and execute real-time actions.
Looking ahead, AI integration into augmented reality (AR) and virtual reality (VR) interfaces holds promise for creating highly interactive and immersive experiences. Moreover, beyond mere integration into applications, smartphones, and devices are expected to operate AI systems at the operating system (OS) level. This will enable these devices to automatically understand user preferences and execute tailored actions seamlessly.
The potential of AI spans far beyond managing emails or organizing schedules; it’s poised to automate various tasks based on user behavior and priorities. Qualcomm has outlined the future of personalized on-device AI, showcasing the possibilities.
Moreover, strides are being made towards AGI (Artificial General Intelligence) by OpenAI and other entities. AGI aims to exceed human capabilities, incorporating intellectual input and reasoning.
OpenAI is also exploring ways to mitigate training data limitations by generating high-quality synthetic data. Whether this aligns with the speculated Project Q* remains to be clarified.
ChatGPT, once confined to conversation, has evolved into a versatile tool with numerous applications, achieving significant status. However, when viewed within the larger scope of AI possibilities, it may seem limited. The future holds promise for countless novel AI experiences yet to emerge.
0 Comments