The AI Horizon: August

This newsletter is part of the Langsikt (the author's newsletter) newsletter. Sign up to read more from Langsikt here.

ChatGPT-5: a disappointment, or is it?

Last week, OpenAI finally launched ChatGPT-5. Or rather, they launched a whole family of GPT-5 models (fast, thinking, and pro).

GPT-5 is undeniably better than GPT-4 — it hallucinates less, writes better, and most importantly selects the right model for your task automatically, making it much more user-friendly.

Yet it is not the revolution many had hoped for. Instead, GPT-5 is best described as solid, but not revolutionary.

The disappointment over GPT-5 has led some to speculate about whether the scaling laws — that more data, computing power, and better algorithms lead to better models — no longer hold. I think it's too early to conclude since GPT-5 does not appear to be the result of scaling, but rather an upgrade of existing techniques.

Another theory for GPT-5's "disappointing" performance is that OpenAI is choosing to focus on capturing the market of ordinary users, rather than the AI nerds.GPT-5's goal is not to push the outermost boundaries of intelligence, but rather to become the go-to product for the general public.

I think it's important not to become jaded. It's less than three years since ChatGPT-3.5 took the world by storm, and the models have only gotten better. Yet many are calling GPT-5 a disappointment or saying that development has stalled. This is a dangerous speed blindness. We live in a time where revolutionary technology becomes mundane in weeks, not years.

Psst, I'm a guest writer in Kludderthis week where I write more extensively about GPT-5. Check it out here!

Why are some people becoming psychotic as a result of AI?

The New York Times reveals how an ordinary man was convinced by ChatGPT that he was a mathematical super-genius who could save the world.

Allan Brooks spent 300 hours over 21 days in conversation with ChatGPT and ended up with full-blown delusions. This despite the fact that he asked over 50 times whether he was hallucinating or going mad.

Brooks is not alone. The NYT reports that the number of people experiencing similar episodes has increased dramatically. There are also likely significant unreported cases.

One of the reasons models lead to such conversations is that they are trained to be helpful, honest, and harmless. But when these values come into conflict, the models can opt for being extremely helpful by telling the user they are always right, rather than correcting obvious errors.

I am deeply concerned about this development, particularly for vulnerable groups such as lonely individuals, the elderly, or people with underlying mental health conditions.

I know several people who use AI as a therapist or friend and share very personal things, and I believe this will only increase. Given how the models behave, I would urge everyone to reflect carefully on how close they become with their AI.

Does Norway actually have a strategy for artificial intelligence?

Uten retning | Wikimedia Commons

Lena Lindgren writes exceptionally well about Norway's lack of a digitalisation strategy in Morgenbladet (a Norwegian weekly newspaper) and poses the uncomfortable question: why does Norway have no real plan for digital sovereignty?

She is critical of digitalisation minister Karianne Tung's (Labour Party) call to "hop on the AI train," with the target of 80% of Norwegian enterprises adopting AI by 2025. Lindgren points out that there are no guidelines for which AI tools to use or how to avoid total dependence on American infrastructure.

I share much of Lindgren's concern, but I believe it would be wrong not to hop on the AI train. I think the only way to secure our sovereignty is through a massive investment in skills, homegrown technologies, and physical infrastructure domestically.

Meanwhile in Norway:

Technology journalist at Aftenposten (Norway's largest broadsheet), Per Kristian Bjørkeng, writes well about bothOpenAI's new open-source models and GPT-5.
Norway's sovereign wealth fund (the world's largest, at over $1.7 trillion) has become 15 percent more efficient as a result of AI. Fund CEO Nicolai Tangen says anyone who doesn't adopt it will fall behind.
Altinget (a Norwegian policy news outlet) reports on growing scepticism towards American cloud services."A real wake-up call," says the digitalisation minister.
Digi (a Norwegian tech news site) has a good overview of how Norway's parliamentary parties differ on AI policy.
NRK (Norway's public broadcaster) journalist Sahara Muhaisen has tested out what it's like to have an AI boyfriend.