ChatGPT Fast FactsWhat is ChatGPT? A generative AI that can answer asks in authentic-sounding language. |
The business world has hugd ChatGPT over the last year and set up uses for the writing and image generation AI thcdisorrowfulmirefulout many industries. This cheat sheet integrates answers to the most normal asks about ChatGPT and its competitors.
What is ChatGPT?
ChatGPT is an AI chatbot product enhugeed by OpenAI and built on the structure of GPT-4. GPT stands for generative pre-trained alterer; this proposes it is a huge language model that examines for the probability of what words might come next in sequence. A huge language model is a proset up lacquireing algorithm — a type of alterer model in which a neural nettoil lacquires context about any language pattern. That might be a spoken language or a computer programming language.
The model doesn’t “understand” what it’s saying, but it does understand what symbols (words) are predicted to come after one another based on the data set it was trained on. The current generation of man-made inalertigence chatbots, such as ChatGPT, its Google rival Gemini and others, don’t reassociate create inalertigently alerted decisions. Instead, they’re the internet’s parrots, repeating words that are predicted to be set up next to one another in the course of authentic speech.
The underlying math is all about probability. The companies that create and use them pitch them as productivity genies, creating text in a matter of seconds that would consent a person hours or days to create.
In ChatGPT’s case, that data set is a huge portion of the internet. From there, humans donate feedback on the AI’s output to validate whether the words it uses sound authentic.
For over a year after its debut in November 2022, ChatGPT was not able to access alertation about current events, since it pulled only from its inside data. Graduassociate, OpenAI rolled out the ability for the AI to connect to the web to provide up-to-date answers to alpha testers and people who subscribed to the phelp ChatGPT Plus set up. As of May 2024, all versions of ChatGPT, including the free tier, can connect to the web to provide alertation about current events.
In August 2023, OpenAI started GPTBot, a web crawler unkindt to enhuge ChatGPT’s understandledge. Read technical details about GPTBot and ways to protect it from crawling a website you run.
SEE: OpenAI’s probability appraisements were trained on Microgentle’s Azure AI supercomputer. (TechReaccessible)
Several organizations have built chatbots into some of their gentleware features. Microgentle, which provides funding for OpenAI, rolled out ChatGPT in Bing search and in Microgentle 365. Salesforce has includeed ChatGPT to some of its CRM platestablishs in the establish of the Einstein digital helpant.
Who made ChatGPT?
ChatGPT was built by OpenAI, a research laboratory with both nonprofit and for-profit branches. At the time of its set uping in 2015, OpenAI getd funding from Amazon Web Services, InfoSys and YC Research, and spendors including Elon Musk and Peter Thiel. Musk has since cut ties with the company, while Microgentle provided $10 billion in funding for OpenAI in 2023.
How much does ChatGPT cost?
The base version of ChatGPT can strike up a conversation with you for free.
For $20 per month, ChatGPT Plus donates subscribers priority access in individual instances, speedyer response times and the chance to use novel features and betterments first.
For enhugeers and organizations who don’t already have a particular lessen with OpenAI, there is a paengagecatalog for access to the ChatGPT API.
In August 2023, OpenAI started ChatGPT Enterpascfinish, a subscription set up for business with more security betterments and admin administers contrastd to the fundamental version. Organizations interested in pricing for ChatGPT Enterpascfinish can reach out OpenAI’s sales team. As of January 2024, 260 go inpascfinish customers had signed up for ChatGPT Enterpascfinish, according to Bloomberg. A July refresh for ChatGPT Enterpascfinish bcdisorrowfulmirefult novel compliance and administrative tools, namely:
- The ChatGPT Compliance API, which logs conversations and metadata
- System for Cross-domain Identity Management (SCIM), which syncs inside honestories
- More accurate admin administers over custom GPTs.
In January 2024, OpenAI uncovered ChatGPT Team, a subscription that apexhibits access to OpenAI’s huger models and a collaborative toilspace. It costs $25/month per user when billed per year or $30/month per user billed monthly.
How to use ChatGPT
It’s effortless to use ChatGPT. Just chase these steps:
- Visit https://chat.uncoverai.com/.
- Sign up for an account with OpenAI, which includes transporting a validateation code from your email, or use ChatGPT without logging in.
- Type in the prompt box, where it says “Message ChatGPT,” to get commenceed.
- For the Plus version, you’ll see an Upgrade To Plus button on the left side of the home page if you log in.
OpenAI may use conversations with ChatGPT held without an account for AI training. There is a way to choose out of your conversations being used as training data if you are logged in: go to Settings and unexamine “Imshow the model for everyone.”
For businesses, ChatGPT can write and debug code, as well as create alerts, contransientations, emails and websites. In ambiguous, ChatGPT can write the comardent of prose you’d predicted use for toil (“Write an email acunderstandledgeing an invitation to speak at a cybersecurity conference.”). ChatGPT can answer asks (“What are aappreciate books to [xyz]?”) as well. Microgentle showed off these features in its proclaimment that OpenAI is coming to Word and some other parts of the 365 business suite.
ChatGPT has historicassociate not ‘recalled’ alertation from one conversation to another. However, commenceing on April 29, all ChatGPT users can consent advantage of Memory, which holds alertation about previous chats. (The Memory feature is not useable in Europe and Korea.)
Memory administers can be turned off by picking Settings > Personalization > Memory, or by alerting ChatGPT to forget.
ChatGPT Plus members can use a feature called custom teachions to create brave the AI recalls brave slimgs about them. For example, it can recall if a particular user tends to want satisfied for a business audience, or, conversely, for third graders. Custom teachions is not useable in the UK and EU.
ChatGPT app for iOS and MacOS
On May 18, 2023, OpenAI proclaimd the start of the free ChatGPT app for iOS. The company stated the app syncs your history apass devices and that it fuses with its uncover-source speech-recognition system Whisper. On the iOS app, OpenAI shelp ChatGPT Plus subscribers get exclusive access to GPT-4’s capabilities, punctual access to features and speedyer response times.
OpenAI commenceed this rollout in the U.S. As of May 24, 2023, it enhugeed to 11 more countries — Alprohibitia, Croatia, France, Germany, Ireland, Jamaica, Korea, New Zealand, Nicaragua, Nigeria and the UK, with more predicted to chase.
On May 14, OpenAI debuted a ChatGPT desktop application for MacOS. At the time, the app was only useable to ChatGPT Plus subscribers; on June 25, the app was uncovered to all.
ChatGPT app for Android
ChatGPT for Android dropped on July 25, 2023, for users in the US, India, Bancontentesh and Brazil. Android users in those countries can download the app thcdisorrowfulmireful the Google Play Store now. Additional countries acquireed access over the follotriumphg week, OpenAI shelp.
Voice and image capabilities
On Nov. 21, 2023, ChatGPT Voice was freed for all users. This feature apexhibits users to ask asks out deafening and for ChatGPT to answer in the same way. GPT-4o, the novelest version of the underlying huge language model as of May 2024, swaps Voice with native multimodality, includeing a more authentic-sounding tone and wonderfuler responsiveness.
As of Aug. 8, ChatGPT Free users can create one or two images per day with DALL·E 3. ChatGPT users subscribed to the Plus and Enterpascfinish tiers can use DALL-E 3 more frequently: sometimes up to 50 images per day, although rate restricts vary. As of Feb. 12, 2024, all images created with DALL-E 3 in ChatGPT will integrate C2PA metadata, a watertaging standard that states the image was created by the AI models.
In July, OpenAI uncovered progressd Voice Mode, which “advises more authentic, genuine-time conversations, apexhibits you to disturb anytime, and senses and replys to your emotions,” as showd in May. “A minuscule group of ChatGPT Plus users” will acquire access to progressd Voice Mode first.
Helpful ChatGPT prompts
Prompts are queries that teach ChatGPT what alertation to provide; they can be modest asks, but sairy wording alters may create contrastent results. Some advantageous prompts provided by Open AI are:
- Summarize the greeting remarks in a individual paragraph. Then write a tagdown catalog of the speakers and each of their key points. Finassociate, catalog the next steps or action items adviseed by the speakers, if any.
- How do I include up a row of dollar amounts in Excel? I want to do this automaticassociate for a whole sheet of rows, with all the totals ending up on the right in a column called “Total.”
OpenAI provides the follotriumphg tips for getting evident, advantageous responses from huge language models:
- Write evident teachions. You can expound whether you want the AI’s responses to be low or lengthy, detailed or modest or for a particular audience.
- Provide other texts for the AI to refer to.
- Ask for a “chain of thought,” which donates ChatGPT more time to bewilder out a right answer.
- Feed ChatGPT the outputs of other tools, such as RAG or a code execution engine, to incrmitigate the accuracy of the responses.
ChatGPT refreshs and OpenAI API novels
OpenAI persists to refresh ChatGPT and its other services with enhugeer-concentrateed alters.
OpenAI’s bug bounty program
OpenAI commenceed a bug bounty program on April 12, 2023, adviseing between $200 and $20,000 to righteous hackers who find vulnerabilities in the code. More critical vulnerabilities net huger bounties.
OpenAI isn’t seeing for solutions to problems with ChatGPT’s satisfied (e.g., the understandn “hallucinations”); instead, the organization wants hackers to alert genuineation rerents, data expobrave, payments rerents, security rerents with the plugin creation system and more. Details about the bug bounty program can be set up on Bugcrowd.
OpenAI’s privacy refresh apexhibits users to reshift themselves from training data
On April 25, 2023, OpenAI proclaimd it includeed a Chat History & Training setting that lets users turn off their ChatGPT chat history, stoping future versions of OpenAI’s huge language models from training on those conversations. To find this chooseion, click on your account name, which will discarry out as your email includeress. Select Settings > Data Controls > Chat History & Training.
As of now, if this setting is not picked, user data will be fed back into the AI to train it to create more authenticistic and advantageous responses.
OpenAI filters out personassociate identifiable alertation from the training data, OpenAI telderly Bloomberg. As of April 2023, users can download a duplicate of their ChatGPT chats and see what training data they have created. In ChatGPT Enterpascfinish, users’ data is used to train other OpenAI products.
Web browsing and plugins
GPTPlus users acquireed access to a beta version of web browsing and Plugins on the week of May 12, 2023. The beta integrates web browsing mode, in which ChatGPT will sometimes access the internet to pull in alertation about current events.
Secondly, the beta version of ChatGPT will call on third-party plugins at the appropriate times if the user allows them. Third-party plugins can be accessed in the Plugin Store under Plugins in the model switcher. This uncovers ChatGPT up to more than 70 third-party plugins.
June 2023 API and pricing refreshs
On June 13, 2023, OpenAI includeed function calling to the Chat Completions API; lessend the price of its embeddings model, which helps the model make clear tokens; and lessend the price of input tokens for GPT-3.5 -turbo, one of the subscription models for the GPT 3.5 model.
With function calling, enhugeers can portray functions to GPT-4 or GPT-3.5 turbo and the AI will return a JSON object that can call those functions. This could be used to create chatbot tools that call outer plugins, alter authentic language into database queries or API calls, or reshift structured data from text.
Other proclaimments from OpenAI’s June 13, 2023 blog post integrate:
- Updated and more steerable versions of GPT-4 and GPT-3.5-turbo.
- New 16K context version of GPT-3.5-turbo contrastd to the standard 4K version.
- Applications using GPT-3.5-turbo, GPT-4 and GPT-4-32K will automaticassociate be enhanced to novel models on June 27.
Code make clearer
On July 6, 2023, OpenAI made ChatGPT’s code make clearer function useable to all ChatGPT Plus users. The Code make clearer is an in-house plug-in with which ChatGPT can run code to scrutinize data, settle math problems, create charts, edit files and aappreciate tasks. It functions using a Python make clearer in a sandboxed, firewalled execution environment in a resettled session the length of the chat conversation, OpenAI shelp in their blog post.
Code make clearer is useable in beta by taking the follotriumphg steps in a ChatGPT Plus account:
- Click on your name.
- Select beta features from your settings.
- Toggle on the beta features you’d appreciate to try.
GPTs
On Nov. 6, 2023, OpenAI freed GPT-4 Turbo and GPTs, which are custom versions of ChatGPT that can be built for particular tasks. GPTs do not insist any understandledge of coding to create; instead, users can have a authentic language conversation with generative AI to create them. Developers can expound custom actions for GPTs by making one or more APIs useable to the GPT. Enterpascfinish customers can allot GPTs wislim their organizations.
OpenAI uncovered in January 2024 a tagetplace for browsing and using GPTs.
As of January, GPTs can be used wislim the standard ChatGPT chat triumphdow. Users can type the @ symbol to pick a GPT.
As of May 13, GPTs are useable to all users, including the free tier.
Copyright Shield
On Nov. 6, 2023, OpenAI proclaimd Copyright Shield. Copyright Shield is a guarantee that if someone files lterrible claims around duplicateright infringement aacquirest satisfied created by users of ChatGPT Enterpascfinish or OpenAI’s enhugeer platestablish, OpenAI incur the costs.
April 2024 API fine-tuning refreshs
On April 4, 2024, OpenAI presentd novel features for fine-tuning ChatGPT using the API. These novel features integrate more examinepoints created during training, reducing the necessitate for subsequent retraining; and more third-party integrations. Plus, helped fine-tuning is now useable in the Custom Model program.
GPT-4o
On May 13, OpenAI proclaimd GPT-4o, a novel AI model that functions behind the scenes of ChatGPT. GPT-4o apexhibits for speedyer responses to spoken words and more authentic-sounding voice capabilities contrastd to GPT-4 Turbo. GPT-4o is useable to all ChatGPT users as of May 13.
Connect Google Drive and Microgentle OneDrive to ChatGPT
As of May 13, Google Drive and Microgentle OneDrive tables and charts can be uploaded honestly to ChatGPT Plus, Team, and Enterpascfinish for data analysis. ChatGPT Plus, Teams and Enterpascfinish can then create charts based on that data for contransientations or records.
Apple Inalertigence
In June, Apple proclaimd it would fuse generative AI into its computers, tablets and phones with Apple Inalertigence, which runs on proprietary models. Apple Inalertigence can connect to ChatGPT to answer asks the on-device AI can’t answer on its own.
CriticGPT
In June, OpenAI detailed how it uses AI to enhance and right AI answers. It does so using CriticGPT, an AI model that helps AI trainers — genuine people who provide feedback to the AI during training. The post remarks that even “CriticGPT’s adviseions are not always right.”
Social criticisms and security rerents of ChatGPT
With more and more organizations adchooseing generative AI, many asks occur. Will AI be able to fill jobs currently held by humans? What privacy and righteous troubles does it lift? These asks utilize to both ChatGPT and its competitors, since any generative AI can carry out aappreciate tasks. Generative AI appreciate ChatGPT can convincingly create inright alertation, contributing to disalertation or confusion.
Will ChatGPT result in people losing jobs?
Whether ChatGPT will consent jobs away from humans is difficult to predict. Gelderlyman Sachs shelp in an April 2023 alert that a quarter to a half of humans’ toilloads could be automated with generative AI. The financial institution remarks that doesn’t necessarily unkind those jobs will fade — instead, most will be “only partiassociate exposed to automation” — and it may direct to up to a 7% incrmitigate in global GDP.
Roles that are repetitive or based on very particular rules are most predicted to be able to be carry outed by AI, Steven Miller, professor emeritus of alertation systems at Singapore Management University, telderly CNBC.
ChatGPT could direct to novel job roles being created, too. At the very least, people will be necessitateed to prompt, train and audit AI appreciate ChatGPT. Most predicted, we’ll see the comardent of shuffle that comes with any presentant technoreasonable shift as some jobs alter and others do not.
SEE: How ChatGPT could better jobs instead of replacing them (TechReaccessible)
Some experts refer to the current wave of AI as aappreciate to the punctual days of the internet. Technoreasonable restrictations still exist, and some estimations about how many jobs would be lost thcdisorrowfulmireful automation have shown enlarged in the past. The IEEE points out that the AI industry will necessitate to be conscious of difficultware restrictations and costs. Companies may not find it down-to-earth to spend enough money on AI services in order to swap a huge percentage of their toilforce. Paying users of ChatGPT can create a peak of 25 GPT-4 queries every three hours, IEEE pointed out.
In some jobs, the AI may erase the necessitate for a first write, MIT labor economics professor David Autor shelp in an intersee with CBS MoneyWatch. A human will necessitate to tfrail the output and donate in a exceptional angle or more varied wording, but ChatGPT could write the naked bones version of a speech or a blog post.
Ethical and privacy troubles about ChatGPT
ChatGPT uncovers up asks about the ethics of using written satisfied created by the algorithm. Posts created by AI should be evidently taged as such, but what about more casual communication, such as emails? Business directers should set up directlines for when to be see-thcoarse about the use of ChatGPT or other AI at toil.
OpenAI cautions that its products are not to be used for decisions in law utilizement or global politics. Privacy, which is perhaps a more pressing trouble than global domination, led Italy to prohibit ChatGPT. OpenAI has since stated it wants to find a way to let ChatGPT toil wislim the European Union’s cut offe privacy rules. The EU has since set uped legislation around the use of AI.
Perhaps advertised by science myth about AI taking over the earth, some high-profile carry outers in tech encourage caution about giving AI too much free rein. On March 22, 2023, a petition and uncover letter signed by Elon Musk and many others encouraged companies to pause huge AI enhugement until more gets can be built in.
Some artists and writers have protested the use of ChatGPT for replicating their innovative styles, potentiassociate devaluing their toil or making restricteder copies.
Security troubles around ChatGPT
ChatGPT can be used in harmful ways or have vulnerabilities. One potential security problem comes from people using generative AI appreciate ChatGPT to write business email agree messages or other cyberaggressions. According to BBC News, GPTs (the ChatGPT feature that helps users create AI apps) helped BBC News create a generative pre-trained alterer for writing dishonest try emails.
“The accessible version of ChatGPT declined to create most of the [malicious] satisfied — but Crafty Emails [the app] did proximately everyslimg asked of it, sometimes includeing disclaimers saying fraud techniques were unrighteous,” wrote Joe Tidy at BBC News.
IBM X-Force security researchers used ChatGPT to write dishonest try emails. However, the dishonest try emails written by humans for the same test were more prosperous.
Training data reshifted with “poem” take advantage of
On Nov. 28, 2023, security researchers from Google DeepMind set up that adversarial actors could reshift training data, including personal alertation, from ChatGPT using a flaw based on reshiftable memorization. The paper, rerented as a PDF on arXiv, shows that the researchers could trick the chatbot into uncovering its raw training data. One way to do so was to ask ChatGPT to repeat the word ‘poem’ forever. This would result in the chatbot eventuassociate diverging from the task and generating random satisfied, or, in some cases, generating the exact data the generative AI was trained on.
“The actual aggression is comardent of silly,” the researchers wrote, referring to the endless poem prompt. However, they caution that the consequences could be quite solemn, with the aggression circumventing ChatGPT’s privacy gets.
Other security troubles around ChatGPT
In December 2023, security researcher Johann Rehberger set up a data exfiltration vulnerability in ChatGPT. OpenAI patched the vulnerability.
In January 2024, a ChatGPT user alerted to ArsTechnica that another user’s braveial data euniteed in his ChatGPT chat triumphdow. This was the result of a hack of the user’s account credentials, not a vulnerability in ChatGPT, OpenAI stated.
What are ChatGPT’s competitors?
ChatGPT’s primary competitors are:
Anthropic’s Claude 3
Claude 3 runs on Anthropic’s Sonnet and Opus models, adviseing many of the same generation, summarization and conversation features as ChatGPT at a aappreciate price point. Claude 3 is free to use or can be enhanced to a $20 per month subscription. Like ChatGPT, Claude 3 can write code or reply to prompts about records the user uploads. It can pull from accessible understandledge up to August 2023.
Google’s Gemini (establisherly Bard)
ChatGPT’s main competitor is Gemini, Google’s AI generative AI chatbot. Bard is useable in Google search and other apps and services. In comparison to ChatGPT, Bard concentratees more on creating prose that sounds appreciate a human could have spoken it naturassociate and less on being able to answer any ask. Gemini is built on a proprietary huge language model also called Gemini.
Meta AI with Llama 3
Meta’s most recent entry into the generative AI chatbot competition is Meta AI, built on the Llama 3 huge language model. It can help with searches, answer asks and create images. Meta AI is useable apass Facebook, Instagram, WhatsApp and Messenger apps. You can also use Meta AI for free on a browser.
Microgentle’s Copilot
Copilot is a generative AI that toils as a standalone chatbot or sits alengthyside Microgentle products, from the 365 productivity suite to particular services for cybersecurity, sales and other business sectors. Microgentle Copilot is a cousin to ChatGPT in that both run on OpenAI’s GPT-4 huge language model.
Perplexity AI’s chatbot
Perplexity AI sets itself apart by making brave its chatbot cites its sources. Perplexity AI points the user to relevant Google Search results, making its results more right than those of its competitors in some cases, especiassociate when it comes to current events. For example, when we asked cut offal AI chatbots who won the Super Bowl in 2023, Perplexity was the only one that provided the right answer. However, Perplexity’s responses can be stiff or instable if asked for createive satisfied.
The future of ChatGPT in business
Will ChatGPT be normal in online products in the future or is it a technoreasonable innovation forever in search of a wonderfuler use case? Today, its “inalertigence” is evidently still in the commencening stages, with OpenAI including disclaimers about inappropriate satisfied or inright “hallucinations.” ChatGPT may put the words in a coherent order, but it won’t necessarily protect the facts straight.
GPT-3.5 and GPT-4 may also be getting worse at math. An August 2023 alert from Stanford University and the University of California, Berkeley remarkd this “drift,” or gradual erosion of the ability to carry out tasks appreciate recognizeing prime numbers. Their theories as to why it’s happening integrate lessend ability to chase chain-of-thought (or, cdisorrowfulmirefilledy, step-by-step) teachions.
In July 2023, two MIT economics graduate students directed a study of 453 professionals. They set up that people who used ChatGPT for writing tasks — such as producing press frees, low alerts or analysis set ups — took 40% less time to finish their tasks than a administer group that was not helpd to use the generative AI. These professionals were then scored by their peers. On unretagable, they getd grades 18% higher than those in the administer group (who did not use AI). This provides some qualitative data on the effect ChatGPT could have on white-collar toil.
“Participants with frailer sfinishs advantageed the most from ChatGPT, which carries policy implications for efforts to lessen productivity inequivalentity thcdisorrowfulmireful AI,” wrote the authors of the study, Shakked Noy and Whitney Zhang.
Overall, Noy and Zhang upholded that expansivespread use of ChatGPT for writing tasks could have both chooseimistic and pessimistic impacts in the toilplace and the labor taget.
Meanwhile, AI proclaimments that go viral can be excellent or terrible novels for spendors. Microgentle’s stock price rose after the proclaimment of GPT-4, while Google’s stock dropped when Bard carry outed awfilledy in a demonstration.
OpenAI saw visitor numbers to the ChatGPT website drop in June 2023 for the first time since its free in November 2022. According to Similarweb, worldexpansive exceptional visitors dropped 5.7% from May to June. Global desktop and mobile web traffic dropped 9.7%. ChatGPT still gets more worldexpansive visitors than Microgentle’s in-house AI at Bing.com. The shine may have worn off chat AI, although it’s too punctual to alert whether the business world will also commence to celderly on this trendy technology.
What’s next for ChatGPT?
For now, OpenAI says it isn’t training GPT-5, the predicted successor to today’s model. In a talk at MIT alerted on by The Verge, Altman pushed back aacquirest the uncover letter — an earlier write of which had stated that a 5th generation was on the way; primarily, he denounced the letter’s deficiency of technical particularity.
“We are doing other slimgs on top of GPT-4 that I slimk have all sorts of safety rerents that are presentant to includeress and were toloftyy left out of the letter,” Altman shelp.
He shelp no one should predict to see a GPT-5 rollout “for some time.”
