OpenAI, the leading AI research laboratory, has launched its latest language model, the GPT-4. The model is a more advanced and multimodal version of its predecessor, the GPT-3.5, with the ability to process input in both text and image form. OpenAI has embedded the model into several apps, including Duolingo, Khan Academy, and Morgan Stanley Wealth Management, to develop conversational language bots, online tutors, and internal chatbots.
Multimodal Model
The GPT-4 is a significant advancement in language processing as it can now generate detailed descriptions and answer questions based on the contents of a photograph, making it a multimodal model. OpenAI has partnered with Danish start-up Be My Eyes to create a virtual volunteer that can guide people with visual impairments using the GPT-4.
Human-level Performance
OpenAI claims that the GPT-4 exhibits “human-level performance” on academic and professional benchmarks such as the US bar exam, advanced placement tests, and SAT school exams. The model has also scored in the 90th percentile on the Uniform Bar Exam, a significant improvement compared to its predecessor, which only scored in the 10th percentile.
Limitations and Risks
Despite its capabilities, OpenAI acknowledges that the GPT-4 has limitations similar to earlier GPT models. It is not fully reliable, has a limited context window, and does not learn from experience. The company also identified potential risks around bias, disinformation, privacy, and cybersecurity. The GPT-4 can generate potentially harmful content, including advice on planning attacks or hate speech, and represent various biases and world views. It can also generate code that is compromised or vulnerable.
Competitive and Safety Concerns
OpenAI has not revealed any technical details about the GPT-4, including the architecture of the model, the data it was trained on, or the hardware and computing capacity used to deploy it. The company cites competitive and safety concerns as reasons for not disclosing the information.
Investments and Integration
Microsoft has made a multibillion-dollar investment in OpenAI, placing a bet on the future of generative AI. The GPT-4 will underpin Microsoft’s Bing chatbot and is expected to integrate into its consumer products soon. Google has also opened up its conversational chatbot, Bard, to a limited pool of testers and allowed customers of Google Cloud to access its large language model PaLM for the first time to build applications.
The GPT-4 is a significant advancement in language processing and AI technology, with the ability to process input in both text and image form. However, its limitations and potential risks require caution and responsible use. As AI technology continues to evolve, companies like OpenAI must balance innovation with ethical and safety considerations to ensure that the technology is used for positive outcomes.