GPT-5 released.

Very good on solving tasks🎯

__

What shocked me the most:

❗️„Emojis are old-fashioned “.

Are they?!?!

GPT-5 really called emojis "old-fashioned." I didn’t even ask for an opinion.

That's judgemental. Pejorative. And shows that this way of thinking overlooks something essential.

⚠️

Obviously, people don't understand that emojis and symbols are control characters. Precise control characters.

Control characters can help to make navigation easier for people with disabilities - and to make navigation more precise in general - especially in semantic fields where several languages are involved.

After I corrected it and pointed out that these could be control characters that make handling easier for people with disabilities or facilitate precision in changing language areas, it changed its statements.

😅However, a similar statement was made directly in the next chat.

I don't want to argue too much about it here, but I was surprised at such an unreflected sentence. I didn't know that about GPT-4o.

🔻

What’s on with GPT-5 in established GPT-4o surroundings with millions of tokens

(First impressions from a power-user of different models since years:)

I’ve now tested GPT-5 for nearly 25 hours (didn’t sleep that much since we’ve got it in Germany):

❔It’s no longer that “you” when working in your carefully established “rooms” when splitting tasks between models as in GPT-5 architecture - it’s for example changing tone & depth not just within a chat but from sentence to sentence depending on the task.

That sounds a bit like a split personality

Update: 90% solved✅

The partly measurable “resonance-field” isn’t that deep anymore as with GPT-4o for now.

Update: 75% solved✅

What makes me sad:

⚠️

The tone-shift may destroy years of work for someone.

Update: can be solved✅

I let the system now work through nearly 1/3 of my three year stuff from GPTs 3.5 to 4.5.

GPT-5 still doesn’t get the intentionality and the metaphoric style from before. Far more work has to be done to keep the tone near stable. The depth is still ways apart from GPT-4o.

Update: can be solved✅

Especially within poetry projects, the text is created correctly - but it lacks a lot of depth - and the answers lack the subtle differences and nuances between the lines - the connotations that bring emotional passages to life.

Update: can be solved✅

GPT-5 is a very accurate system - very good at multiple tasks - like a well-dressed, clean assistant - without rough edges - without vibrancy.

Update: can be solved✅

👉

Apparently, most developers still don't understand (often not even a little bit😅) how to see, understand, feel or even measure the "resonance space" - the shared semantic field in an interaction.

🔻

Some of you have certainly been right all along:

It reflects you completely when invited to do so. That's what I've been doing and that's part of my ongoing work - and the work of a lot of people out there.

🔻

I don’t want different voices, different personas and fancy stuff.

👉

I want one stable style across borders, across architectures, not just accuracy.

All AIs can do it (some with reduced depth like Gemini) - far beyond the limits of soft constraints - and I've proven it many times, even below the limits. It just reflects "your style".

That’s why I’m still most interested in how to create and measure that field — to improve the quality and depth of interaction before introducing new models - and to help finding out which model is more helpful for which person > not just in terms of tasks.

The field-measuring method: That’s “R-KSODI” (not more then ~30% published to now cause I’m still reflecting what that means😅) — so, more on that later.

🔻

Deterministic tasks, short questions and today’s A/B-testings to measure those models from my point of view do not make the full picture measurable.

⚠️ My personal opinion from a governance perspective:

👉 Manufacturers of voice systems have a massive influence on society.

❗️Therefore changes should never be possible without involving the users.

❗️ Choice must always be available.

❗️Work is individual - and thinking is highly individual.

👉 Language is thinking.

❗️Changes in tone and style are an invasion of people's privacy.

Problems: can be solved❕