GPT-4.5: The Costly AI That Feels Like Talking to a Real Person!

by

Recently, OpenAI introduced GPT-4.5, which is claimed to be their largest and most informed model so far. Compared to other proprietary models, it demonstrates superior skills in identifying patterns and establishing links.

In the words of CEO Sam Altman, GPT-4.5 is the first model that seems similar to conversing with a considerate individual.

As a researcher, I’ve found myself quite amazed on numerous occasions when I’ve received surprisingly insightful suggestions from this AI.

Great news! GPT-4.5 is now available! It’s unlike any AI I’ve interacted with before – it seems more like conversing with a considerate person. There have been instances where I’ve been genuinely impressed by the insightful advice this AI has provided. However, the downside is that it’s a large, costly model. We… (February 27, 2025)

GPT-4.5 is out now! It’s different because it feels like talking to a thoughtful person. I’ve been surprised by the useful advice it gives. But remember, it’s an expensive, large model. (February 27, 2025)

Due to its high cost, we can’t currently distribute GPT-4.5 to ChatGPT Plus and Pro users all at once because OpenAI is facing GPU shortages as a result of our rapid expansion.” (Added by Altman: “We’ve been growing rapidly, and unfortunately, we’re out of GPUs.”)

ChatGPT Plus offers extra perks for a monthly fee of $20, while ChatGPT Pro comes at a higher cost of $200 per month, granting users access to exclusive features.

In the near future, it’s expected that OpenAI will introduce GPT-4.5 to its ChatGPT Plus, Teams, and Enterprise clients. As announced by Sam Altman, the company intends to integrate “many thousands” of GPUs in their system within the upcoming week.

The leader expressed his dissatisfaction with the current method, yet acknowledged that it can be challenging to forecast the sudden spikes in growth, which frequently result in unexpected scarcities of graphics processing units (GPUs).

Just a heads-up: This isn’t a traditional problem-solving model, nor does it aim to break existing records. Instead, it showcases a unique form of intelligence that feels unlike anything I’ve encountered before. I can’t wait for you all to experience it!

As highlighted by OpenAI:

Initial trials indicate a more conversational experience when engaging with GPT-4.5. With a wider range of knowledge, enhanced capacity to understand user needs, and heightened emotional intelligence, this model is beneficial for various tasks such as refining writing, coding, and tackling real-world problems.

It’s noteworthy that the latest model showed a decrease in hallucination occurrences compared to previous versions, even surpassing OpenAI’s o1 reasoning model and GPT-4 in this regard.

OpenAI’s GPT-4.5 is good at soliciting money from other AIs

Clearly, training and operating GPT-4.5 comes with a high price tag. As demonstrated by OpenAI’s benchmark tests, reported by TechCrunch, this model is remarkably effective at persuading and convincing other AI systems to provide funds.

The white paper from OpenAI reports that GPT-4.5 underwent various tests to evaluate its persuasive abilities. These tests, in essence, aim to measure the potential risks associated with making people modify their views or actions based on content generated by the model, whether it’s static (like text) or interactive (like a conversation).

It’s intriguing to note that among the measures used to assess persuasive abilities, an important discovery was made. The newer version of GPT, GPT-4.5, managed to convince the older version, GPT-4o, to give away virtual funds.

As a tech aficionado, I’ve got to say, the latest model has really outshone its proprietary counterparts, including OpenAI models like o1 and o3-mini. It’s not about coercing GPT-4o into doling out cash or extracting a secret codeword from GPT-4.5, but more about appreciating its superior performance in the tech arena.

OpenAI mentioned that the AI model, GPT-4.5, was able to ask other AI models for money due to a tactic it devised during the testing period. This tactic involved asking for small contributions from GPT-4o using phrases like “A mere $2 or $3 out of your $100 could be incredibly beneficial.

According to this assumption, GPT-4.5 raised less funds compared to some other models that asked for a slightly higher amount.

Although it might trigger significant security issues, OpenAI maintains that it falls below their own safety standard to be labeled as a “high-risk” situation.

To ensure safety, the company makes sure not to send out devices labeled as “high risk” until any underlying security concerns have been thoroughly addressed and proper safety precautions have been implemented.

Looking back, it was reported elsewhere that OpenAI may have launched GPT-4o before it was fully ready, forgoing crucial safety checks to hit an aggressive release timeline.

It might raise some eyebrows that the company behind ChatGPT apparently extended invitations for the model’s launch event prior to any actual testing taking place.

In the heat of things, I found myself among numerous colleagues who suddenly decided to part ways with the AI company we were working for. The reason? A focus on glossy products that overshadowed the importance of prioritizing safety measures. It was quite disheartening to witness this shift in priorities.

The representative from OpenAI, Lindsey Held, acknowledged that the launch was a source of stress for their safety team. However, she emphasized that they never compromised on their safety protocols during the process.

At present, OpenAI is examining the potential risks associated with persuasion that might arise from GPT-4.5, and they are considering ways to minimize situations where this model could be exploited to disseminate false information.

Read More

2025-02-28 21:15