In a recent interview, Sam Altman says ‘OpenAI o1-preview is deeply flawed’, thus highlighting the challenges the model faces. The release of OpenAI’s O1-preview has created excitement within the tech industry, yet it remains far from flawless. OpenAI CEO Sam Altman recently shared insights into the development of the model, acknowledging its imperfections while highlighting its potential.
In this recent interview, Altman reflected on OpenAI’s journey, stating that the company’s first product, GPT-3, was not initially something they took great pride in. Despite the attention GPT-3 garnered, the team identified its limitations, which led to the eventual release of GPT-3.5, ChatGPT, and ultimately GPT-4. Altman noted that the company’s progress lies not in incremental improvements but in aiming for breakthroughs that could benefit humanity in the long term.
OpenAI’s mission extends beyond fixing immediate issues. Altman emphasized that the company often falls into the trap of focusing on making products “less bad” instead of striving for more groundbreaking innovations. The company is committed to long-term advancements, to transform how artificial intelligence is integrated into society.
O1: In the Early Stages of Development
Sam Altman says ‘OpenAI o1-preview is deeply flawed’, reflecting the company’s understanding of its shortcomings. Altman compared the new O1-preview to GPT-2, suggesting that the model is still in its early development phase. While the model already shows potential, Altman acknowledged that there is much more room for growth. He assured users that significant improvements would be made in the coming months, as OpenAI moves from O1-preview to the full release.
The O1-preview model has notable limitations. It currently lacks features like image support, browsing the web, and file uploads. As it stands, users must input text-based prompts for the model to generate responses. This means that while O1-preview can solve complex math problems, it cannot process visual inputs, a feature available in other advanced models.
O1 vs. GPT-4
Comparisons between O1-preview and GPT-4 have emerged in the media, but Altman highlighted that the two models serve different purposes. O1 is particularly aimed at users who require advanced reasoning capabilities for solving complex problems in areas like coding, mathematics, and scientific research. In contrast, GPT-4 is still more capable in common, everyday tasks.
Another drawback of the O1-preview model is its slower response time. OpenAI stated that the model is designed to “think before answering,” mirroring the human process of carefully considering a problem before offering a solution. While this may improve the quality of responses, it leads to slower performance compared to other models.
Looking Ahead
As AI development progresses, Sam Altman says ‘OpenAI o1-preview is deeply flawed’. But, OpenAI has made it clear that the O1-preview model is just the beginning. With future updates promising enhanced performance and more features, users can expect a much-improved version in the coming months. Though still a work in progress, O1 marks a significant step toward the company’s goal of building AI systems that can solve complex problems and propel humanity forward.
Alongside O1-preview, OpenAI has introduced O1-mini, a smaller and more affordable version of the model. O1-mini is 80% cheaper and is designed for developers needing advanced coding capabilities but not requiring the broader reasoning skills of O1-preview. Both O1-preview and O1-mini are now available to ChatGPT Plus and Team users, with message limits set for each model.
Pricing Details
OpenAI’s O1-preview comes at a higher price than GPT-4, with the cost per 1 million input tokens set at $15, compared to GPT-4’s $5. This pricing difference is reflective of O1’s specialized reasoning capabilities. In India, the cost to access O1-preview stands at around INR 1,260 for 1 million input tokens, compared to INR 420 for GPT-4.
The ChatGPT Plus subscription, which grants access to these models, is priced at $20 per month, or approximately INR 1,650 in India.
Also Read: Visa Will Acquire Featurespace, Expanding Its Fraud Protection Tools.
