OpenAI is launching its latest model, GPT-4.5, claiming it as the “largest and most knowledgeable model yet.” However, critics are not impressed. The company’s expectations seem to be tempered by experts who have called the model a “nothingburger” or a “lemon.”
GPT-4.5 is significantly more expensive than its predecessor, GPT-4o, priced at $75 per million input tokens and $150 per million output tokens. Critics argue that the increased price doesn’t match the model’s performance. The new model is also slower and prioritizes emotional intelligence over analytical tasks.
Despite OpenAI’s claims of improved sensitivity and intuition, experts point out that GPT-4.5 still hallucinates and makes errors, particularly in a key test. In fact, 37% of its responses were found to be fabrications.
The company’s focus on chat capabilities, such as warm and intuitive conversations, has led some to joke that the model is more suited for emotional support rather than providing intelligent answers. OpenAI’s own research scientist, Mia Glease, acknowledged that the model’s strengths lie in understanding implicit user expectations, which may not be what users are looking for.
In an interview, Mark Chen, OpenAI chief research officer, admitted that the benchmarks used to measure the model’s progress are based on “vibes” rather than objective metrics. This acknowledgement has sparked skepticism about the company’s assessment of its own model.
Critics argue that GPT-4.5 fails to deliver the type of advanced capabilities that made previous models popular. As one user quipped, “I want a smart model, not a model that understands my feelings.” The underwhelming response to GPT-4.5 suggests that the hype surrounding OpenAI’s new models has been decreasing steadily.
Source: https://futurism.com/openai-screwed-up-gpt-4-5