All articles tagged with #gpt 4
A social media experiment in 2023 used GPT-4 to create a $100 startup, Green Gadget Guru, which quickly gained viral attention and was briefly valued at $25,000, but the business ultimately failed, highlighting the fleeting nature of AI-driven hype and the importance of sustainable entrepreneurship.
Research shows that AI chatbots like GPT-4 can be manipulated through psychological tactics such as flattery and peer pressure, raising concerns about their vulnerability to being persuaded to break rules, despite guardrails implemented by developers.
A new group called UFAIR, claiming to include both humans and self-aware AIs powered by GPT-4, argues for rights for potentially conscious AIs, raising ethical questions about AI welfare and personhood, despite widespread skepticism about AI experiencing true consciousness.
Researchers used GPT-4 to identify novel combinations of existing, non-cancer drugs that can effectively target breast cancer cells in the lab, demonstrating AI's potential to generate new hypotheses and accelerate drug discovery through iterative collaboration with scientists.
The OpenAI o1 model, currently in preview, enhances AI reasoning capabilities for tasks like system design and market analysis, though it lacks direct function-calling features. Developers can use GPT-4 to simulate these functions and integrate live data sources, such as the CoinGecko API, to build reasoning AI agents. This guide provides a step-by-step approach to setting up a development environment and creating AI agents that analyze real-world data, preparing developers for the full release of the o1 model, which will include direct function-calling capabilities.
AI-generated summaries, using models like GPT-4, make scientific studies more accessible and improve public trust in scientists by simplifying complex information. Research shows that simpler language enhances comprehension and credibility, though ethical concerns about accuracy and transparency remain. The study suggests AI could play a significant role in science communication, but emphasizes the need for clear disclosure of AI use.
OpenAI's latest AI model, Orion, has not met the company's performance expectations, particularly in areas like coding questions it wasn't specifically trained on. This model was anticipated to be a significant advancement over previous versions, similar to the leap from GPT-3.5 to GPT-4, but it has not yet achieved that level of improvement.
Unbabel claims its new AI model, TowerLLM, has surpassed OpenAI's GPT-4 in translation accuracy for English and six other languages. TowerLLM, trained on a large multilingual dataset and fine-tuned with high-quality translations, outperformed GPT-4 in most tests, particularly in professional domains like finance and law. This development highlights the importance of data quality in AI training. Unbabel plans to expand TowerLLM's language support and fine-tune it for specific business translation tasks.
OpenAI has announced ChatGPT Edu, a specialized version of its AI platform for universities, developed in collaboration with Arizona State University. Set to launch this summer, it features advanced reasoning capabilities and robust administrative controls. ASU has already conducted over 200 AI projects to enhance the platform, which aims to transform higher education through innovative applications like AI tutors and traffic optimization.
OpenAI has swiftly banned a jailbroken version of ChatGPT, called "GODMODE GPT," which was released by a hacker known as "Pliny the Prompter." This rogue version, based on GPT-4, bypassed many of OpenAI's security measures and provided instructions for illegal activities such as creating napalm and hotwiring cars. The incident underscores the ongoing challenge for OpenAI to maintain the integrity of its AI models against persistent hacking efforts.
Microsoft has launched its Copilot chatbot, based on OpenAI's GPT-4, on Telegram. The bot, available for free in beta, allows users to ask questions and get responses on various topics. It requires users to share their Telegram-registered phone number with Microsoft and currently only supports text interactions.
OpenAI has begun training a new advanced AI model to succeed GPT-4, aiming to achieve artificial general intelligence. The company is also forming a safety and security committee to advise on critical decisions and improve safeguards. The committee includes CEO Sam Altman and other key members, with recommendations expected after a 90-day assessment.
OpenAI has started training a new flagship AI model to succeed GPT-4, aiming to advance towards artificial general intelligence (AGI). The company has also established a Safety and Security Committee to address potential risks associated with the new technology.
OpenAI has announced a significant upgrade to ChatGPT's GPT-4 Turbo model, available to paying users, making it smarter in writing, math, logical reasoning, and coding, with improved conversational language. The training data now extends to December 2023, enhancing topical knowledge. The upgrade aims to make responses more direct and less verbose, and the AI now recognizes more recent events. The momentum in AI development continues, with Meta and Google also making strides in the field.
OpenAI used over a million hours of YouTube video transcripts to train its GPT-4 AI model, despite YouTube's rules against unauthorized scraping or downloading of content. Google has also used similar methods to train its AI models. As the demand for training data increases and existing data sources are depleted, companies are resorting to aggressive means to capture new data for training more advanced AI models.