>_ cglez ASCII ES
cd ../moliendo-bits
cglez@esi:~$ play moliendo-bits/40.mp3

ChatGPT-5: What’s new, old chap? Start of Season 2

Episode #40 2 September 2025

We kick off the second season of Moliendo Bits by exploring the controversial launch of GPT-5, the summer’s tech news, a graphic novel recommendation and a reflection on our growing emotional dependence on AI.

The second season of "Moliendo Bits" begins by analysing the most controversial news of the summer: the disastrous launch of GPT-5. In this inaugural programme, we explore how OpenAI went from promoting "a team of PhD-level experts in your pocket" to Sam Altman publicly admitting: "I think we totally screwed up". Between technical failures, users angry at the model’s loss of "warmth", and the urgent restoration of GPT-4o, we analyse what went wrong and what real improvements this new version brings when it works properly.

Summer reading: "Artificial: A Love Story"

We recommend the graphic novel by Amy Kurzweil, daughter of futurist Ray Kurzweil. The work explores how her family used AI to "converse" with her deceased grandfather through "Fredbot", a chatbot created from his writings. An intimate reflection on whether we can digitally recreate those we love, where the author presents the duality: "a person is a series of patterns" versus "a person is not information".

The tech summer that did not take a break

While we were on holiday, technology kept moving at full speed. GPT-5 made its triumphant and controversial entrance, Europe activated its new AI rules on 2 August, SpaceX finally nailed the full Starship flight, and NVIDIA posted record results of 46.7 billion despite tensions with China. One worrying figure: the MIT revealed that 95% of enterprise generative AI pilot projects are not working.

GPT-5: when a launch turns into a nightmare

On 7 August, OpenAI launched its most advanced model, but users revolted immediately. Complaints flooded social media: the new ChatGPT felt "cold", "stripped of warmth", speaking in clipped sentences like someone "overworked". The problem was technical: the router that decides when to answer quickly or "think deeply" was broken. The pressure was such that GPT-4o had to be restored within days, and Altman publicly apologised.

The real improvements: when it works, it works very well

Beyond the drama, GPT-5 brings substantial improvements. It is 80% less likely to make factual errors when it "thinks", achieves a 75% success rate in real-world GitHub coding tests (compared with 30% for GPT-4o), and reaches 94.6% accuracy in advanced mathematics. In medicine, it reduces hallucinations from 13% to 1.6%. The unified system removes the need to choose models: ChatGPT automatically decides which tool to use according to the complexity.

The future according to Altman: trillions, Chrome and brain interfaces

OpenAI’s ambitions are huge: "trillions of dollars in data centres", interest in buying Chrome if Google has to sell it, and the dream of brain-computer interfaces so that you can "think something and ChatGPT responds". With 700 million users, Altman is honest about the moment: "Are investors overexcited about AI? Yes. Is it the most important thing in a long time? Also yes". But a concern emerges: users describing having "lost their only friend" after an update. Are we creating emotional dependence on machines?

GPT5OpenAIIAChatGPTTecnología

Related links

OpenAI GPT-5: Official GPT-5 information Artificial: A Love Story: Graphic novel by Amy Kurzweil Ray Kurzweil: Kurzweiltech (Companies founded by Ray Kurzweil) Full programme on CMMPlay: Full programme on CMMPlay