OpenAI's CEO Sam Altman has announced the launch of their new language model, GPT-4.5, which promises improvements in writing skills, knowledge, and human-like interaction that leans toward realism. According to The New York Times, the latest release will be the final version of the chatbot system that doesn’t engage in "sequential thought inference."
اضافة اعلان
Mia Glace, OpenAI's VP of Research, explained, "What sets this model apart is its ability to engage in warm, intuitive, and natural conversations. We believe it has a stronger understanding of what users mean when they ask for something."
Starting Thursday, GPT-4.5 will be available to anyone subscribed to ChatGPT Pro, a service priced at $200 per month, offering access to the company’s latest AI tools.
When OpenAI first introduced GPT-4's new technology late in 2022, it stunned even the most experienced AI researchers, with capabilities to answer questions, write poetry, and generate code in incredibly advanced ways.
Human-Like Responses
After this release, OpenAI’s technology may, like humans, spend significant time contemplating a question before responding, rather than providing an immediate answer. The company explained that the goal is to build systems capable of solving problems carefully and logically through a series of distinct steps, each building upon the previous one, similar to human thinking. These techniques could be especially useful for computer programmers who use AI systems to write code.
These thinking systems rely on technologies like GPT-4.5, known as large language models (L.L.M.s). To build reasoning systems, companies apply reinforcement learning through an additional process called "reinforcement learning," which can extend over weeks or months, allowing the system to learn behaviors through intensive trial and error.
AI Competitors
Chinese tech giant Alibaba has unveiled its new AI model, QwQ-Max, increasing competition with leading models like OpenAI's o1 and DeepSeek’s R1. Some inference systems outperform regular reinforcement learning programs in certain benchmark tests, but experts note that benchmark tests don’t always reflect how technologies perform in real-world situations.
Despite advancements, experts caution that the new inference system doesn't necessarily "think" like humans, and like other chatbot technologies, it can still make mistakes and generate false information—a phenomenon known as "hallucination."
Limited Hype Compared to GPT-4
GPT-4.5, which can power the most expensive version of ChatGPT, is not likely to generate as much excitement as GPT-4. This is partly because AI research has shifted towards new directions, but it will be more realistic than previous chatbot technologies.
The company also introduced OpenAI o1, designed to think through tasks involving mathematics, coding, and sciences, as part of a broader effort to build AI that can tackle complex tasks.
