Ole Reissmann

About · Newsletter

THEFUTURE

The Great AI Downgrade

Newsletter sent 23.9.2025 by oler

In this issue: OpenAI gives some insight into what 2.6 billion daily requests actually look like. Why journalism has a fighting chance with the help of user needs. Cecilia Rikap on why we’re dumbing ourselves down to match AI’s limitations. Plus: How ChatGPT trained me for the Berlin Marathon and then watched me suffer in the heat.

What we’re talking about: OpenAI has published what it says is the largest study to date on ChatGPT usage. Based on real chats from their users, excluding business accounts and API usage.

The user base is growing and young—nearly half are between 18 and 25 years old. Based on names, OpenAI estimates that more requests now come from women than men. OpenAI counts 2.6 billion requests per day (Google counts around 14 billion searches).

And what are users doing with ChatGPT? A quarter of requests fall into the category of “seeking information.” If you add “practical guidance”, that adds up to almost half of all conversations. Ali Mahmood mapped the dominant ChatGPT uses to the User Needs model and arrives at three strategic questions for journalistic offerings.

His conclusion is worth reading, so just briefly: Are we just a news service, or do we help people learn and act? Our product and messaging should match whichever we choose.

What else I’ve been reading:

AI & Journalism Links

The News Industry’s GenAI Cautionary Tales: Generative AI failures have shown, among other things, the value of scrutinizing outsourced work. (Clare Spencer, Generative AI in the Newsroom)

Don’t count on counting fingers: Reporter’s Guide to Detecting AI-Generated Content (Henk van Ess, Global Investigative Journalism Network)

“Make sure your final work is yours”: Internal guidelines at Business Insider allow journalists to use AI to assist with writing first drafts without public disclosure–which looks like a sensible and practical solution. (Oliver Darcy, Status)

FOOM! This YouTube channel from Runway streams 24/7 AI-generated videos and it’s bad, but like, not all of is terrible garbage?! (via Sofie Hvitved)

And now: Time for a reality check. While some are debating whether AI will replace us, Cecilia Rikap argues we’re already replacing ourselves—downgrading our own thinking to match what machines can do. Spoiler alert: that’s not progress. Her new book, “The Rulers: Corporate Power in the Age of AI and the Cloud”, will come out with Verso in 2026.

Three Questions with Cecilia Rikap

Hands on: I let ChatGPT train me for the Berlin Marathon, hoping to break Sub-3—three minutes faster than my personal best. The plan was ambitious, with a massive training load, and I found myself wondering: Would GPT-5 break me instead?

I followed Chat’s plan religiously. After each run, I uploaded the Strava activity. I ran a spectacular half marathon in the process. I was in shape! One day before the race, Chat seemed as confident as ever:

“You’ve ticked all the boxes: speed, threshold, long-run finish, 10K sharpener, HM confirmation. We stay the course. Execute the taper, nail the logistics, and run the plan. Sub-3 is on.”

Reader, I did not break Sub-3. And somehow, it wasn’t even Chat’s fault.

Last year, Berlin saw a very nice 8°C/46°F at the starting line. This year? It was already 20°C/68°F in the morning and climbing by the minute.

While the first half went according to plan, the heat got me in the second half, adding a whopping 17 minutes to my goal time. Running in 27°C/81°F weather, I suffered.

Chat, knowing the outcome, would you have advised differently?

“The core build was right. (…) You didn’t lose to training. You lost to environment plus pacing choice for that environment. Fix the heat protocol and the ‘ego-guard’ on the first 10–15 km, and you’ll cash that fitness on the next fair-weather course.”

Chat was right all along. It just didn’t have the context. No weather information was taken into consideration. It was my fault after all.

One more thing: A very good post by Jesse Vincent on Threads:

Apple AI summaries are wild. I feel like one of these three things is supposed to be more important than the other two and also probably isn’t.

This is THEFUTURE.

Subscribe to THEFUTURE

Get this newsletter in your inbox. For free.

The previous issue is Worldviews Wrapped in Algorithms, the next issue is When Chatbots Start Asking the Questions.