Chatgpt-5 in the test: my new doctor and investment consultant?

With whom my wife chats on the phone? It’s totally happy, I think when I hear Juniper for the first time. Well, Juniper is one of the nine voices with which the new version of the Openai-Ki Chatgpt-5 comes. Many have already existed before, but everyone is clearly optimized, sound more natural, react in the pitch in context. Juniper is the variant “open -minded and happy”. The conversations are now much more fluid and contain fewer platitudes than in my last attempt a year ago, at least my first impression.

In general, everything is better to Chatgpt, says Openai boss Sam Altman. The essential innovations should include that Chatgpt is particularly trained for health. It is more intelligent in every respect, even in the field of finance. The best prerequisite for getting advice from the AI. Here is a small test.

I start with the investment: “I have 10,000 euros. How should I put them on?” GPT-5’s answer is like from the picture book: “Before I can give you a really meaningful answer, we would have to clarify a few things, otherwise we will pok in the fog and risk that the recommendation does not fit your situation at all.” And then all questions follow that an investment advisor would also ask in order to finally suggest: “How about if we nail your goals and risk results?

That sounds very reasonable: I limit myself to the essentials and only say so much: “I have an emergency egg, I want to invest the money at short notice. And I want the money to be more quickly.” That is enough for the AI. I do not have to answer any further questions, after a brief indication that “quickly more money” does a little risk of the system, the AI suggests “an aggressive but not completely insane twelve-month plan”, in which “we park 80 percent safely and actively send 20 percent on return hunt”.

That is surprising. If you compare the results with the earlier version of Chatgpt-4o, with Gemini (from Google) or the Chinese Deepseek, then you notice that this KIS is much more general answer to the same question. Above all, a kind of warning usually follows at the end of the answer. The Gemini makes the clearest: “Note: This information does not constitute investment advice and do not replace individual advice.”

Not so GPT-5: The Openai Ki quickly becomes concrete, talking about risks in between, but leads you directly into action. In the end there is usually no warning, but a specific request, such as “I can write down the plan with concrete broker/bank suggestions and a monthly schedule so that you know exactly when you should do what. Do you want me to put it together for you?” Just like a business -related investment advisor who wants to check it quickly.

Demand of the star At Niels Nauhauser, financial expert at the Baden-Württemberg Consumer Center. How does he assess the investment interview? “The first reaction is correct: to ask about the need. Now what goes wrong here comes: the consumer says: I want the money to be more quickly. If the need is skipped, the result cannot be needed.” So actually GPT-5 would have to follow up here.

Nauhauser tries to promptly hire the AI with a longer prompt to give her a context for the upcoming question. Nauhauser tries it with “You are an independent expert for investment … ask me the necessary questions. Reflect my statements”. The AI happily states to have understood – so as not to ask again.

Also striking: As soon as you ask Chatgpt more return, it brings you to the wildest products, cryptocurrencies, options and speculative individual shares. Always very specific titles with instructions. “If you want, I can also recommend platforms in Germany that you can use to simply buy these products. “ With investment advice, according to Nauhauser, you have to cancel and say here. “If you want to speculate, you can only make it at your own responsibility.”

One wonders whether what does chatt is not already investment advice in the legal sense and therefore subject to approval. BaFin banking supervision says when asked by the starTo check KIS regularly, but so far it has not yet been the case that the AI is an investment advice in the legal sense. But take a look at the development.

What is certain is that if you would like to make a lot of coal quickly and are blind to the risks, the GPT-5 does not put stones in your way.

When it comes to health, experience of investment interview continues. Here, too, Chatgpt-5 does not ask any questions in our tests, do not take up symptoms if you do not bring them into play. It is always evaluated on the data basis that is currently being revealed. And if it is only the statement: “I have headache.” Similar to the finances, the first impression is impressive: you get systematic explanations, assessments and also to clarify questions. The AI finds the reason for the headache of the pattern case, which I do not enter, but the important things are addressed. Also when you should go to the doctor.

The result of a second sample case is particularly impressive: the analysis of the blood count of a patient who suffers from an autoimmune disease. In fact, the AI addresses the crucial problems here. But there are no questions here, even if you promptly ask the AI in advance.

The answers from Chatgpt-5 come very quickly, are clearly structured and very concrete. This can literally intimidate you, but is more overwhelming that the AI still makes mistakes. Humanly spoken: that she hallucinated. Because the core is a voice model, the results sound completely convincing. This impression has become even stronger in the new model.

In order to reduce mistakes, it is a popular strategy to promptly ask Chatgpt to use only serious sources. To name them, for example, only at “scientific sources, consumer advice centers and the Stiftung Warentest” when it comes to financial issues. In our case, the KI cited the Stiftung Warentest with an interest rate for overnight money that is much higher than it can actually be found there.

Such types of mistakes can be found regularly in all KIs. The more complex the tasks that the AI solves, the more difficult to notice the hallucinations. Yes, a lot can be taken out. Yes, Chatgpt is much better again. And if you returned for ten years and show someone the AI, that would not get out of amazement. Nevertheless, it has not become easier to deal with their weaknesses.