Digital Event Horizon
OpenAI has released GPT-4.5, its latest large language model, but the results are mixed. Despite some improvements over its predecessor, the new model's high price and computational demands raise questions about its long-term viability.
GPT-4.5 has released with mixed reviews from the tech community. The new model offers significant improvements over GPT-4o in specific areas, but at a high cost and increased computational demands. GPT-4.5 scored lower than simulated reasoning models on certain tests and ranked poorly in performance versus cost. Independent testing found the model to be 10th in overall coding ability with poor performance versus cost. The law of diminishing returns applies to large language models, according to OpenAI's CEO Sam Altman. GPT-4.5 is not a replacement for GPT-4o and will be evaluated for its long-term availability.
In a shocking turn of events, OpenAI has released its latest and most capable traditional AI model, GPT-4.5, to mixed reviews from the tech community. The new model boasts significant improvements over its predecessor, GPT-4o, in specific areas such as multilingual MMMLU (general knowledge) test performance and confabulation reduction. However, these incremental gains come with a hefty price tag and significantly higher computational demands.
According to OpenAI's own benchmark results, GPT-4.5 scored lower than the company's simulated reasoning models (o1 and o3) on tests like AIME math competitions and GPQA science assessments. The model also ranked poorly in performance versus cost, making it impractical for many applications where GPT-4o already performs adequately.
Tech investor Paul Gauthier performed independent testing of GPT-4.5's coding ability using Aider's Polyglot Coding benchmark and found that the model ranked 10th in overall ability, with poor performance versus cost. OpenAI's flagship reasoning model, o1 pro, costs significantly less than GPT-4.5 despite offering specialized simulated reasoning capabilities.
The news comes as a surprise to many, given the hype surrounding large language models (LLMs) and their potential to revolutionize various industries. However, it seems that the law of diminishing returns applies even to these cutting-edge AI systems.
OpenAI's CEO Sam Altman acknowledged this trend in a statement on X, writing that GPT-4.5 is strong on "vibes" but low on analytical strength. The company also noted that the model is not a replacement for GPT-4o and will be evaluated to determine its long-term availability.
As the tech industry continues to evolve, it's clear that the development of large language models is just one aspect of this journey. While OpenAI's efforts may have yielded incremental gains, they also highlight the need for more efficient and cost-effective solutions that can meet the demands of a rapidly changing landscape.
Related Information:
https://www.digitaleventhorizon.com/articles/The-Diminishing-Returns-of-Large-Language-Models-OpenAIs-GPT-45-deh.shtml
https://arstechnica.com/ai/2025/02/its-a-lemon-openais-largest-ai-model-ever-arrives-to-mixed-reviews/
Published: Fri Feb 28 12:08:34 2025 by llama3.2 3B Q4_K_M
