OpenAI is experiencing a steady return with its latest AI model due to the pressure of recent investments. OpenAI’s next AI model is projected to deliver lower performance gains in comparison to its predecessor.
As per a news article by AI News, during employee testing – Orion, the new AI model gained the performance level of GPT-4 after just 20% of its training. However, the transition to GPT-5 is predicted to exhibit lesser quality improvements than that of GPT-3 to GPT-4.
Few researchers have the first impression of Orion as not being much improvement over the previous GPT versions in handling certain tasks. It performs better with tasks related to the language but may not be able to overcast previous models at coding and other tasks.
Initial stages of AI training yield major improvements in the model, while most phases result in smaller performance gains. T is not expected to be delivering any advancements in comparison to previous improvements.
OpenAI’s future is at a crossroads. The success of its early AI models will significantly impact its future fundraising efforts. Simultaneously, the AI industry faces a critical challenge: the dwindling supply of high-quality training data. As AI models become increasingly sophisticated, the demand for more data grows, pushing companies to explore innovative solutions. OpenAI is responding to this challenge by rethinking its AI development strategy, focusing on post-training model improvements. However, as the company navigates this complex landscape, it must contend with leadership departures and the need to balance innovation with practical applications and investor expectations.
The implications of these challenges extend beyond OpenAI. The entire AI industry may need to adapt to a new era of AI development, one that prioritizes efficiency and quality over sheer scale. As the race for AI supremacy continues, it remains to be seen how companies like OpenAI will navigate these challenges and shape the future of artificial intelligence.
What is expected from GPT-5?
People have set unrealistic expectations from GPT-5, which is likely to take a back flip in terms of surging its performance. In the recent podcast with Bill Gates, Altman focused on the need of improving reasoning abilities of GPT-4, as it is currently limited to a certain capacity. However, GPT-5 is predicted to overcome those issues and have more advanced capabilities. Which also implies that GPT-5 model will be able to analyse and solve complex data sets and offer sophisticated solutions.
GPT-5 Features:
- Better accuracy and higher speed: GPT-4 has made substantial strides in reasoning accuracy, matching the capabilities of GPT-4 Turbo. It can reliably answer general knowledge questions with an impressive 87.2% accuracy, as per an article by OnGraph. Meanwhile, GPT-5 is poised to revolutionize the field by processing up to 50,000 words at once, doubling the capacity of its predecessor. This significant speed boost promises to enhance the model’s overall performance and efficiency.
- Extended multimodality: This will be the most in-demand feature of GPT-5. It will be based on the existing features of GPT-4 and will bring new advanced functionalities such as video integration. This feature will change the entire communication pattern with AI, allowing us for flexibility and natural forms of interaction.
- Increased Reliability: Sam Altman focused on the key improvements of GPT-5 and said it will be more reliable than GPT-4. He also said that GPT-4 provides good response when asked the same question a thousand times, but it is not aware of the best response. With GPT-5, the goal is to offer best results consistently, thus enhancing its reliability.
Overall, GPT-5 looks promising and will follow same payment structure as its predecessor. While ChatGPT offers free subscription plans, the GPT-5 API-based services are predicted to depend on the usage amount. GPT-5 is also likely to be the most capable model of LLM of OpenAI.