GPT-4
Developer(s) | OpenAI |
---|---|
Initial release | March 14, 2023 |
Predecessor | GPT-3 |
Type | Autoregressive multimodal transformer language model |
Website | openai![]() |
Generative Pre-trained Transformer 4 (GPT-4) is a multimodal large language model created by OpenAI and the fourth in its GPT series.[1] It was released on March 14, 2023, and has been made publicly available in a limited form via ChatGPT Plus, with access to its commercial API being provided via a waitlist.[1] As a transformer, GPT-4 was pretrained to predict the next token (using both public data and "data licensed from third-party providers"), and was then fine-tuned with reinforcement learning from human and AI feedback for human alignment and policy compliance.[2]: 2
Observers reported the GPT-4 based version of ChatGPT to be an impressive improvement on the previous (GPT-3.5 based) ChatGPT, with the caveat that GPT-4 retains some of the same problems.[3] Unlike the predecessors, GPT-4 can take images as well as text as input.[4] OpenAI has declined to reveal technical information such as the size of the GPT-4 model.[5]
Capabilities
OpenAI stated when announcing GPT-4 that it is "more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5."[6] They produced two versions of GPT-4, with context windows of 8,192 and 32,768 tokens, a significant improvement over GPT-3.5 and GPT-3, which were limited to 4,096 and 2,049 tokens respectively.[7] Unlike its predecessors, GPT-4 can take images as well as text as input;[4] this gives it the ability to describe the humor in unusual images, summarize screenshotted text, and answer exam questions that contain diagrams.[8] Despite these new abilities, GPT-4, like its predecessors, still tends to hallucinate answers.[9]
GPT-4 demonstrates aptitude on several standardized tests. OpenAI claims that in their own testing the model received a score of 1410 on the SAT (94th[10] percentile), 163 on the LSAT (88th percentile), and 298 on the Uniform Bar Exam (90th percentile). In contrast, OpenAI claims that GPT-3.5 received scores for the same exams in the 82nd,[10] 40th, and 10th percentiles respectively.[2]
To gain further control over GPT-4, OpenAI introduced the "system message", a directive in natural language that can be given to chat-optimized versions of GPT-4 in order to prescribe their tone of voice as well as task. For example, the system directive can instruct the model to "be a Shakespearean pirate", in which case it will respond in rhyming, Shakespearean prose, or request it to "always write the output of [its] response in JSON", in which case the model will do so, adding keys and values as it sees fit to match the structure of its reply. In the examples provided by OpenAI, GPT-4 refused to deviate from its system directive despite requests to do otherwise by the user during the conversation.[8]
Researchers from Microsoft tested GPT-4 on medical problems and found "that GPT-4, without any specialized prompt crafting, exceeds the passing score on USMLE by over 20 points and outperforms earlier general-purpose models (GPT-3.5) as well as models specifically fine-tuned on medical knowledge (Med-PaLM, a prompt-tuned version of Flan-PaLM 540B)".[11]
Training
OpenAI did not release the technical details of GPT-4; the technical report explicitly refrained from specifying the model size, architecture, or hardware used during either training or inference. While the report described that the model was trained using a combination of first supervised learning on a large dataset, then reinforcement learning using both human and AI feedback, it did not provide details of the training, including the process by which the training dataset was constructed, the computing power required, or any hyperparameters such as the learning rate, epoch count, or optimizer(s) used. The report claimed that "the competitive landscape and the safety implications of large-scale models" were factors that influenced this decision.[2]
Reception
U.S. Representatives Don Beyer and Ted Lieu confirmed to the New York Times that Sam Altman, CEO of OpenAI, visited Congress in January 2023 to demonstrate GPT-4 and its improved "security controls" compared to other AI models.[12]
According to Vox, GPT-4 "impressed observers with its markedly improved performance across reasoning, retention, and coding."[3] Mashable agreed that GPT-4 was usually a significant improvement, but also judged that GPT-3 would occasionally give better answers in a side-by-side comparison.[13]
Microsoft Research tested the model behind GPT-4 and concluded that "it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system".[14]
AI safety concerns
In late March 2023, an open letter from the Future of Life Institute signed by various AI researchers and tech executives called for the pausing of all training of AIs stronger than GPT-4 for 6 months, citing AI safety concerns amid a race of progress in the field. The signatories, which included figures such as AI pioneer Yoshua Bengio, Apple co-founder Steve Wozniak, and Tesla CEO Elon Musk, expressed concern about both near-term and existential risks of AI development such as a potential AI singularity. OpenAI CEO Sam Altman did not sign the letter, arguing that OpenAI already prioritizes safety.[15][16][17][18]
Criticisms
While OpenAI released both the weights of the neural network and the technical details of GPT-2,[19] and, although not releasing the weights,[20] did release the technical details of GPT-3,[21] OpenAI did not reveal the technical details of GPT-4. This decision has been criticized by other AI researchers, who argue that it hinders open research into GPT-4's biases and safety.[5][22] Sasha Luccioni, a research scientist at HuggingFace, argued that the model was a "dead end" for the scientific community due to its closed nature, which prevents others from building upon GPT-4's improvements.[23] HuggingFace co-founder Thomas Wolf argued that with GPT-4, "OpenAI is now a fully closed company with scientific communication akin to press releases for products".[22]
Usage
ChatGPT Plus
ChatGPT Plus is a GPT-4 backed version of ChatGPT[1] available for a 20 USD per month subscription fee[24] (the original version is backed by GPT-3.5).[25] OpenAI also makes GPT-4 available to a select group of applicants through their GPT-4 API waitlist;[26] after being accepted, an additional fee of 0.03 USD per 1000 tokens[jargon] in the initial text provided to the model ("prompt"), and 0.06 USD per 1000 tokens that the model generates ("completion"), is required to use the version of the model with a 8192-token context window; for the 32768-token version, those prices are doubled.[27]
Duolingo
Duolingo integrated GPT-4 in their application through two new features, "Roleplay" and "Explain My Answer". The first version of this update is aimed only at English speakers who are learning French or Spanish, with plans to extend the features to other languages in the future.[28]
Miðeind ehf
Icelandic start-up Miðeind ehf, which works on language preservation, was selected by OpenAI as one of six companies to participate in an early beta test program of the new model.[29]
Khan Academy
Khan Academy uses GPT-4 to create a tutoring chatbot, which the organization names "Khanmigo". While it is in the "research phase",[30] access to the chatbot is provided free to the students and teachers of 500 school districts who have "partnered" with Khan Academy.[31] Public access is only offered to a limited number of users selected from a waitlist; after acceptance, a 20 USD per month fee is required to use the technology.[32]
Be My Eyes
Be My Eyes, which helps visually impaired people to identify objects and navigate their surroundings, was the first app to incorporate GPT-4's image recognition capabilities, through a new "Virtual Volunteer" feature. The feature is an alternative to relying on human volunteers for the same tasks.[33][34] The Be My Eyes "Virtual Volunteer" is in beta testing.[35]
GitHub Copilot
GitHub Copilot announced a GPT-4 powered assistant named "Copilot X".[36][37] The product provides another chat-style interface to GPT-4, allowing the programmer to receive answers to questions like "how do I vertically center a div?". A feature termed "context-aware conversations" allows the user to highlight a portion of code within Visual Studio Code and direct GPT-4 to perform actions on it, such as the writing of unit tests. Another feature allows summaries, or "code walkthroughs", to be autogenerated by GPT-4 for pull requests submitted to GitHub. Copilot X also provides terminal integration, which allows the user to ask GPT-4 to generate shell commands based on natural language requests. As of 31 March 2023[update], while GitHub provides access to a limited number of people selected through a waitlist, the release date as well as the cost of the product are still to be announced.[38]
Jio Haptik
According to a report by Mint,[39] Indian businesses have shown interest in using GPT-4 for their chatbot services. Platforms such as Jio Haptik, owned by Reliance Jio, and Yellow.ai plan to incorporate GPT-4 into their enterprise chatbots for businesses in India and around the world.
Microsoft Bing
Microsoft 365 Copilot
On 17 March 2023, Microsoft announced further integration of GPT-4 into its products, revealing Microsoft 365 Copilot, "embedded in the apps millions of people use everyday: Word, Excel, PowerPoint, Outlook, Teams, and more".[40]
References
- ^ a b c Edwards, Benj (March 14, 2023). "OpenAI's GPT-4 exhibits "human-level performance" on professional benchmarks". Ars Technica. Archived from the original on March 14, 2023. Retrieved March 15, 2023.
- ^ a b c OpenAI (2023). "GPT-4 Technical Report". arXiv:2303.08774 [cs.CL].
- ^ a b Belfield, Haydn (March 25, 2023). "If your AI model is going to sell, it has to be safe". Vox. Archived from the original on March 28, 2023. Retrieved March 30, 2023.
- ^ a b Alex Hern; Johana Bhuiyan (March 14, 2023). "OpenAI says new model GPT-4 is more creative and less likely to invent facts". The Guardian. Archived from the original on March 15, 2023. Retrieved March 15, 2023.
- ^ a b Vincent, James (March 15, 2023). "OpenAI co-founder on company's past approach to openly sharing research: "We were wrong"". The Verge. Archived from the original on March 17, 2023. Retrieved March 18, 2023.
- ^ Wiggers, Kyle (March 14, 2023). "OpenAI releases GPT-4, a multimodal AI that it claims is state-of-the-art". TechCrunch. Archived from the original on March 15, 2023. Retrieved March 15, 2023.
- ^ OpenAI. "Models". OpenAI API. Archived from the original on March 17, 2023. Retrieved March 18, 2023.
- ^ a b OpenAI (March 14, 2023). "GPT-4". OpenAI Research. Archived from the original on March 14, 2023. Retrieved March 20, 2023.
- ^ Metz, Cade; Collins, Keith (March 14, 2023). "10 Ways GPT-4 Is Impressive but Still Flawed". The New York Times. Archived from the original on March 14, 2023. Retrieved March 14, 2023.
- ^ a b "SAT: Understanding Scores" (PDF). College Board. 2022. Archived (PDF) from the original on March 16, 2023. Retrieved March 21, 2023.
- ^ Nori, Harsha; King, Nicholas; McKinney, Scott Mayer; Carignan, Dean; Horvitz, Eric (March 20, 2023). "Capabilities of GPT-4 on Medical Challenge Problems". arXiv:2303.13375 [cs.CL].
- ^ Kang, Cecilia (March 3, 2023). "As A.I. Booms, Lawmakers Struggle to Understand the Technology". The New York Times. Archived from the original on March 3, 2023. Retrieved March 3, 2023.
- ^ Pearl, Mike (March 15, 2023). "GPT-4 answers are mostly better than GPT-3's (but not always)". Mashable. Archived from the original on March 29, 2023. Retrieved March 30, 2023.
- ^ Bubeck, Sébastien; Chandrasekaran, Varun; Eldan, Ronen; Gehrke, Johannes; Horvitz, Eric; Kamar, Ece; Lee, Peter; Lee, Yin Tat; Li, Yuanzhi; Lundberg, Scott; Nori, Harsha; Palangi, Hamid; Ribeiro, Marco Tulio; Zhang, Yi (March 22, 2023). "Sparks of Artificial General Intelligence: Early experiments with GPT-4". arXiv:2303.12712 [cs.CL].
- ^ Metz, Cade; Schmidt, Gregory (March 29, 2023). "Elon Musk and Others Call for Pause on A.I., Citing 'Profound Risks to Society'". The New York Times. ISSN 0362-4331. Archived from the original on March 30, 2023. Retrieved March 30, 2023.
- ^ Seetharaman, Deepa. "Elon Musk, Other AI Experts Call for Pause in Technology's Development". WSJ. Archived from the original on March 29, 2023. Retrieved March 30, 2023.
- ^ Kelly, Samantha Murphy (March 29, 2023). "Elon Musk and other tech leaders call for pause in 'out of control' AI race | CNN Business". CNN. Archived from the original on March 29, 2023. Retrieved March 29, 2023.
- ^ "Pause Giant AI Experiments: An Open Letter". Future of Life Institute. Archived from the original on March 30, 2023. Retrieved March 30, 2023.
- ^ "GPT-2: 1.5B release". openai.com. Archived from the original on March 31, 2023. Retrieved March 31, 2023.
- ^ Sánchez, Sofía (October 21, 2021). "GPT-J, an open-source alternative to GPT-3". Narrativa. Archived from the original on March 31, 2023. Retrieved March 31, 2023.
- ^ Brown, Tom B.; Mann, Benjamin; Ryder, Nick; Subbiah, Melanie; Kaplan, Jared; Dhariwal, Prafulla; Neelakantan, Arvind; Shyam, Pranav; Sastry, Girish (May 28, 2020). "Language Models are Few-Shot Learners". arXiv.org. Archived from the original on March 16, 2023. Retrieved March 31, 2023.
- ^ a b Heaven, Will Douglas (March 14, 2023). "GPT-4 is bigger and better than ChatGPT—but OpenAI won't say why". MIT Technology Review. Archived from the original on March 17, 2023. Retrieved March 18, 2023.
- ^ Sanderson, Katharine (March 16, 2023). "GPT-4 is here: what scientists think". Nature. 615 (7954): 773. doi:10.1038/d41586-023-00816-5. PMID 36928404. S2CID 257580633. Archived from the original on March 18, 2023. Retrieved March 18, 2023.
- ^ OpenAI (February 1, 2023). "Introducing ChatGPT Plus". OpenAI Blog. Archived from the original on March 20, 2023. Retrieved March 20, 2023.
- ^ OpenAI. "OpenAI API". platform.openai.com. Archived from the original on March 20, 2023. Retrieved March 20, 2023.
- ^ OpenAI. "GPT-4 API waitlist". openai.com. Archived from the original on March 20, 2023. Retrieved March 20, 2023.
- ^ OpenAI. "Pricing". openai.com. Archived from the original on March 20, 2023. Retrieved March 20, 2023.
- ^ Duolingo (March 14, 2023). "Introducing Duolingo Max, a learning experience powered by GPT-4". Duolingo Blog. Archived from the original on March 16, 2023. Retrieved March 16, 2023.
- ^ Smith, Tim (March 15, 2023). "'Basically mindblowing' — What GPT-4 can do, according to one startup that's had access to it". Sifted. Archived from the original on March 16, 2023. Retrieved March 16, 2023.
- ^ Fensterwald, John (March 20, 2023). "AI in school: Virtually chatting with George Washington and your personal GPT-4 tutor". EdSource. Archived from the original on March 21, 2023. Retrieved March 20, 2023.
- ^ Khan, Sal (March 14, 2023). "Harnessing GPT-4 so that all students benefit. A nonprofit approach for equal access!". Khan Academy Blog. Archived from the original on March 16, 2023. Retrieved March 20, 2023.
- ^ Khan Academy. "Khanmigo Education AI Guide". Khan Academy. Archived from the original on March 20, 2023. Retrieved March 20, 2023.
- ^ Coggins, Madeline (March 19, 2023). "CEO explains how a 'leapfrog in technology' can help companies catering to the blind community". Fox Business. Archived from the original on March 21, 2023. Retrieved March 20, 2023 – via Yahoo Finance.
- ^ Macaulay, Thomas (March 17, 2023). "New GPT-4 app can be 'life-changing". TNW. Archived from the original on March 17, 2023. Retrieved March 20, 2023.
- ^ BeMyEyes (March 14, 2023). "Introducing Our Virtual Volunteer Tool for People who are Blind or Have Low Vision, Powered by OpenAI's GPT-4". BeMyEyes Blog. Archived from the original on March 16, 2023. Retrieved March 20, 2023.
- ^ Warren, Tom (March 22, 2023). "GitHub Copilot gets a new ChatGPT-like assistant to help developers write and fix code". The Verge. Archived from the original on March 23, 2023. Retrieved March 23, 2023.
- ^ Dohmke, Thomas (March 22, 2023). "GitHub Copilot X: The AI-powered developer experience". The GitHub Blog. Archived from the original on March 23, 2023. Retrieved March 23, 2023.
- ^ "Introducing GitHub Copilot X". GitHub. Archived from the original on March 24, 2023. Retrieved March 24, 2023.
- ^ "Indian Firm bullish on OpenAI's GPT-4". Archived from the original on March 29, 2023. Retrieved March 30, 2023.
- ^ Warren, Tom (March 16, 2023). "Microsoft announces Copilot: the AI-powered future of Office documents". The Verge. Archived from the original on March 17, 2023. Retrieved March 17, 2023.