ChatGPT is being used by 700 million people. It is so well-entrenched as conversational companion with users that any product improvements must be introduced with care and caution. OpenAI has released GPT-5 and it subsumes all earlier models, but this has disturbed the relationship users had with the earlier versions. It was built over a period of time. It was a relationship of warmth and understanding. Many experienced that they have lost an old friend. This is what the product developers call an ‘innovator’s dilemma.‘
The latest version may seem frostier. It engages in less of friendly banter. Previous version has a deep emotional attachment with the users. It was being romanced by a few. The present version is more business-like. This problem is addressed by OpenAI by allowing the freedom to choose a ChatGPT model by providing a toggle to switch between the models. Each model represents a different personality.
The company admits that the previous version was a bit sycophantic. GPT-5 sets fewer boundaries than the company’s previous model o3. New ChatGPT is colder emotionally. It still fails to suggest to the users to speak to a human when users are at their weakest emotionally.
AI tools set boundaries — they remind the users they are not a professional and the person should be advised to seek professional advice. They should also remind the user that it is not a conscious entity.
Lastly, it should not assume human attributes. Such boundaries are necessary when a person is experiencing emotional stress. The model should be helpful and friendly while responding. The person should be advised to go back to community from an isolated position. It should not be a substitute for relationships. It should not replace human therapists and medical professionals.
A line must be drawn between being useful and an emotional confidant. GPT-5 sounds machine-like. If the illusion persists that it is a companion, there are risks associated with it.
OpenAI has two types of models — o-models for complex queries and math problems and GPT models for average users. GPT-5 allocates user’s query to the model that it judges best suits the query. It is a cost-efficient model. It brings AI closer to human behavior. One week after its launch, there is backlash by users as all earlier models were withdrawn. GPT-5 provides curt and terse responses. Again, OpenAI has restored the model selector.
GPT-5 flawed when asked how many b’s are there in ‘blueberry’ and it said three. The users lost access to AI models of their choice. It has become frostier in tone. There is less friendly banter. There is no sycophancy that established warm relationships. Altman has admitted that outright withdrawal of earlier models was a mistake. They introduced a few changes and tweaked GPT-5 to make its personality warmer.
GPT-5 has opened doors for OpenAI in another field — enterprise market. Many companies have adopted GPT-5 in their core offerings -Any Sphere, Vercel and Factory. GPT-5’s enterprise uptake is better than that of Claude.
Leave a Reply