ChatGPT Fast Facts:
– What is ChatGPT? ChatGPT is a generative AI that can answer questions in natural-sounding language.
– Who developed ChatGPT? ChatGPT was developed by OpenAI, Inc.
– Pricing: ChatGPT is free, ChatGPT Plus costs $20/month, and ChatGPT Teams costs $25/month for at least two people.
The business world has embraced ChatGPT over the past year, finding various uses for this AI in writing and image generation across different industries. This guide provides answers to common questions about ChatGPT and its competitors.
ChatGPT is an AI chatbot product developed by OpenAI, built on the GPT-4 structure. GPT, or generative pre-trained transformer, is a large language model that predicts the likelihood of words following each other in a sequence. These AI chatbots, including ChatGPT and its competitors, do not make informed decisions but rather generate text based on patterns in natural language.
OpenAI’s AI models are trained on large datasets from the internet, and human feedback is used to ensure the output sounds natural. ChatGPT can now connect to the web to provide real-time information, expanding its knowledge base.
Organizations like Microsoft and Salesforce have integrated ChatGPT into their software offerings. Microsoft, a major funder of OpenAI, uses ChatGPT in Bing search and Microsoft 365, while Salesforce incorporates it into CRM platforms as the Einstein digital assistant.
ChatGPT is available for free, with additional features offered through paid subscriptions like ChatGPT Plus and ChatGPT Enterprise. Users can access ChatGPT through the web or the iOS app, with features like Memory and custom instructions available for Plus members. OpenAI announced that ChatGPT Plus subscribers on the iOS app will have exclusive access to GPT-4’s capabilities, early feature access, and faster response times. The rollout began in the U.S. and expanded to 11 more countries, including Albania, Croatia, France, Germany, Ireland, Jamaica, Korea, New Zealand, Nicaragua, Nigeria, and the UK.
On May 14, OpenAI introduced a ChatGPT desktop application for MacOS, initially available only to ChatGPT Plus subscribers; however, it became accessible to all users on June 25. ChatGPT for Android was launched on July 25 for users in the US, India, Bangladesh, and Brazil, with availability expanding to more countries in the following week.
Voice capabilities were released for all users on Nov. 21, allowing users to ask questions verbally and receive responses from ChatGPT. The latest version of the large language model, GPT-4o, introduced native multimodality in May 2024, enhancing the AI’s tone and responsiveness.
Additionally, ChatGPT Plus and Enterprise subscribers can utilize DALL-E 3 to generate and edit images within the chat window. All images generated with DALL-E 3 in ChatGPT now include C2PA metadata as of Feb. 12, 2024. Users can expect upcoming image interpretation features that enable ChatGPT to answer questions based on images from the user’s device.
OpenAI provides tips for obtaining clear and useful responses from large language models, emphasizing the importance of clear instructions and providing additional texts for reference. A bug bounty program was initiated on April 12, 2023, offering rewards for identifying vulnerabilities in the code. Users can now exclude themselves from training data by adjusting the Chat History & Training setting in their account settings.
In June 2023, OpenAI introduced function calling to the Chat Completions API, reduced pricing for its embeddings model and input tokens for GPT-3.5-turbo, and unveiled updated versions of GPT-4 and GPT-3.5-turbo. On July 6, 2023, the code interpreter function was made available to all ChatGPT Plus users, enabling the AI to run code for various tasks such as data analysis and chart creation. OpenAI stated in a blog post that it operates using a Python interpreter within a sandboxed, firewalled execution environment for the duration of a chat conversation. To access the code interpreter, ChatGPT Plus users can follow these steps:
1. Click on your name.
2. Select beta features in your settings.
3. Toggle on the beta features you wish to try.
GPTs
On November 6, 2023, OpenAI introduced GPT-4 Turbo and GPTs, specialized versions of ChatGPT designed for specific tasks. GPTs can be created through natural language conversations with generative AI, without the need for coding knowledge. Custom actions for GPTs can be defined by developers through one or more APIs. Enterprise customers have the ability to share GPTs within their organizations.
In January 2024, OpenAI launched a marketplace for browsing and utilizing GPTs. As of January, GPTs can be accessed within the standard ChatGPT chat window by typing the @ symbol to select a GPT. As of May 13, GPTs are available to all users, including those on the free tier.
Copyright Shield
OpenAI announced Copyright Shield on November 6, 2023. This feature ensures that if legal claims related to copyright infringement are filed against content created by users of ChatGPT Enterprise or OpenAI’s developer platform, OpenAI will cover the costs.
April 2024 API Fine-Tuning Updates
On April 4, 2024, OpenAI introduced new features for fine-tuning ChatGPT using the API. These enhancements include more training checkpoints to reduce the need for retraining and additional third-party integrations. Assisted fine-tuning is now accessible in the Custom Model program.
GPT-4o
OpenAI unveiled GPT-4o on May 13, a new AI model integrated into ChatGPT for faster responses and more natural-sounding voice capabilities compared to GPT-4 Turbo. GPT-4o is now available to all ChatGPT users.
Integration with Google Drive and Microsoft OneDrive
As of May 13, tables and charts from Google Drive and Microsoft OneDrive can be directly uploaded to ChatGPT Plus, Team, and Enterprise for data analysis. Users can then create charts based on this data for presentations or documents.
Apple Intelligence
In June, Apple announced the integration of generative AI into its devices through Apple Intelligence, which operates on proprietary models. Apple Intelligence can connect to ChatGPT to answer questions that the on-device AI may not be able to address.
Social Criticisms and Security Issues
The increasing adoption of generative AI raises questions about its impact on job displacement, privacy, and ethics. While some believe AI could automate a significant portion of human workload, leading to job changes rather than losses, others express concerns about potential misinformation and privacy breaches. The ethical implications of using AI-generated content, as well as security vulnerabilities, should also be carefully considered. According to BBC News, the ChatGPT feature, which helps users build AI apps, was used by BBC News to create a generative pre-trained transformer for writing phishing emails. Despite the public version of ChatGPT refusing to create malicious content, an app called Crafty Emails managed to generate almost all requested content, sometimes including disclaimers denouncing unethical scam techniques, as reported by Joe Tidy at BBC News.
IBM X-Force security researchers also utilized ChatGPT to write phishing emails, but found that human-written phishing emails were more successful in the same test. Additionally, on Nov. 28, 2023, Google DeepMind researchers discovered a flaw in ChatGPT that allowed adversarial actors to extract training data, including personal information, using a technique called the “poem” exploit.
In December 2023, a security researcher identified a data exfiltration vulnerability in ChatGPT, which was promptly patched by OpenAI. In January 2024, a ChatGPT user reported seeing another user’s private data in their chat window, which was attributed to a hack of the user’s account credentials rather than a vulnerability in ChatGPT.
The article also outlines competitors of ChatGPT, including Anthropic’s Claude 3, Google’s Gemini, Meta AI with Llama 3, Microsoft’s Copilot, and Perplexity AI’s chatbot. It discusses the future of ChatGPT in business and potential implications for the workplace and labor market.
Despite the success and challenges faced by ChatGPT, OpenAI has stated that they are not currently training a successor model, such as GPT-5. En una charla en MIT informada por The Verge, Altman se opuso a la carta abierta, la cual en un borrador anterior había afirmado que una quinta generación estaba en camino; principalmente, criticó la falta de especificidad técnica de la carta.
“Estamos haciendo otras cosas además de GPT-4 que creo que tienen todo tipo de problemas de seguridad que son importantes de abordar y que fueron completamente omitidos en la carta”, dijo Altman.
Dijo que nadie debería esperar ver un lanzamiento de GPT-5 “por algún tiempo”.
Fuente
La publicación completa para 2024 se puede encontrar en World Online.