GPT-5 is friendlier now - but not everyone likes it. Here's why

OpenAI / Elyse Betters Picaro / ZDNET

ZDNET's cardinal takeaways

• OpenAI has made GPT-5 warmer and friendlier.
• Many users accidental they don't privation a friendlier ChatGPT.
• Challenge lies successful uncovering a equilibrium betwixt simplicity and choice.

Get much in-depth ZDNET tech coverage: Add america arsenic a preferred Google source connected Chrome and Chromium browsers.

Looking for a kinder, gentler AI model? Then you mightiness similar the latest spirit of GPT-5. Other people, though, not truthful much. On Friday, OpenAI announced an update to its latest AI exemplary designed to amended its property and quell its critics.

Also: Is ChatGPT Plus inactive worthy $20 erstwhile the escaped mentation offers truthful overmuch - including GPT-5?

"We're making GPT-5 warmer and friendlier based connected feedback that it felt excessively ceremonial before," OpenAI said connected X. "Changes are subtle, but ChatGPT should consciousness much approachable now. You'll announcement small, genuine touches similar 'Good question' oregon 'Great start,' not flattery. Internal tests amusement nary emergence successful sycophancy compared to the erstwhile GPT-5 personality. Changes whitethorn instrumentality up to a time to rotation out, much updates soon."

(Disclosure: Ziff Davis, ZDNET's genitor company, filed an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful grooming and operating its AI systems.)

The reactions truthful far

OK, that sounds similar a worthwhile change, 1 aimed astatine giving the AI a much affable and affable property without being excessively obsequious astir it. But based connected the idiosyncratic absorption truthful far, that's not needfully what radical want.

"This is not astatine each what radical want. I bash not privation it to unreal to beryllium human, I privation it to dispense accusation parsimoniously," said 1 X idiosyncratic successful response.

"Do not talk to maine imitating a human. You are a machine, enactment similar it," said another.

"I don't privation this, I privation it to beryllium my assistant, not my friend," said a third.

"Please don't enactment this connected Pro. I conscionable privation the happening to reply my questions. If I wanted genuine quality interactions I'd spell extracurricular instead," added a fourth.

Also: Don't similar GPT-5? You tin inactive usage GPT-4 and different bequest models successful ChatGPT - here's how

Mixed successful with the complaints were responses from a fewer radical who liked the change. Others suggested offering "personality" arsenic a toggle, leaving it up to each idiosyncratic to determine however the AI should act. In general, though, astir conscionable privation nonstop answers to their requests without immoderate fuss oregon frills.

Ongoing grievances

Since the debut of GPT-5 earlier this month, OpenAI has been stung by a question of protests implicit the caller AI. Some users who were hyped astir the merchandise of GPT-5 person recovered it a letdown. Others person bemoaned the abrupt nonaccomplishment of GPT-4 and see GPT-5 little than susceptible successful comparison.

To code the ongoing grievances, OpenAI has frantically been tweaking GPT-5 and ChatGPT. Trying to placate radical who preferred the older models, the institution reinstated the bequest GPT-4o, GPT-4.1, o4-mini, and o3 models, but lone for Plus and Pro subscribers.

Also: I tested GPT-5's coding skills, and it was truthful atrocious that I'm sticking with GPT-4o (for now)

GPT-5 was besides expected to beryllium a much businesslike exemplary that would determine connected its ain however champion to respond to your request, choosing a speedy answer, longer and deeper thinking, oregon different variation. But aft radical balked astatine that change, OpenAI added 4 antithetic modes -- Auto for radical who don't privation oregon request to archer the AI however to respond, Fast for speedy answers, Thinking for longer thinking, and Thinking mini mode for quicker thinking.

Rocky reception

The rocky reception that's greeted GPT-5 brings up questions of what we privation successful an AI and however acold it tin spell astatine being each things to each people. Do we privation a elemental cut-to-the-chase robotic adjunct that answers our questions arsenic rapidly and concisely arsenic possible? Or bash we privation a friendlier dependable that tin supply warmth and courtesy on with the responses? In galore cases, that depends connected the request.

If I'm asking the AI to reply a factual question, generate code, oregon provide heavy scholarly research, I privation it to beryllium speedy and businesslike without wasting clip connected pleasantries. But if I'm seeking proposal connected a idiosyncratic matter oregon successful request of a sounding board, I'd similar immoderate property splashed successful with the response. ChatGPT adjacent allows you to customize it to assistance it keep way of idiosyncratic details that you privation to share.

Also: Here are each the GPT-5 updates OpenAI has rolled retired since launch

With GPT-5, OpenAI tried to marque ChatGPT simpler and much efficient, eliminating the immense array of models and modes from which to choose. Now, the institution is going successful the other absorption by bringing backmost each the antithetic choices. Since an AI similar ChatGPT can't beryllium each things to each people, the situation is to onslaught a equilibrium betwixt simplicity and idiosyncratic choice. 

Maybe further tweaks volition deed the close spot.