LAST WEEK OpenAI CEO Sam Altman unveiled the AI start-up’s much-anticipated GPT-5 model saying it would make AI assistant ChatGPT feel like talking to a “PhD-level expert in anything”. Within minutes of the flagship model’s release posts appeared on social media picking holes in those claims.
How many Bs in ‘blueberry’? Three, said GPT-5: there’s one at the start, one in ‘blue’, and one in ‘berry’. And how many US states have an R in them? More than you might realise: PhD geographer GPT-5 included Indiana, Illinois and Texas plus the well-known states NESKALIA, NEW JEFSET and MICANN.
OpenAI promised a model that hallucinated less. That doesn’t appear to have happened. So what went wrong, and what are the implications for the future of generative AI and its large language models (LLMs)?
To find out, Charting Gen AI turned to Maria Sukhareva, principal AI expert at Siemens who writes in a personal capacity on Substack as the AI Realist. Maria has been working in natural language processing (NLP) and AI since 2008. “I love AI, but I’m allergic to hype, empty promises, and fraudulent sales pitches,” says Maria who last week shared examples of GPT-5 making a dog’s dinner of European maps and getting basic maths problems horribly wrong. She tells Charting: “this is one of the most spectacular disappointments in AI.”
In Charting’s interview with Maria you will learn:
Why she believes OpenAI chose to make GPT-5 an “ensemble of models” with a smaller “underthinking” model, prone to errors, and a larger “overthinking model”, which takes more time to produce outputs. Spoiler: it’s all about the money.
How GPT-5’s problems might now spell disaster for the much-hyped ‘agentic AI’ future.
How LLMs are sucking the oxygen out of research into alternative approaches to potentially more reliable forms of AI.
Why OpenAI is now likely to ditch the GPT brand. “I don’t think there will be a GPT-6. It’s over.”
Her message to politicians and business owners who will now be hearing there are problems with generative AI and looking for advice.
Whether she thinks negative publicity for OpenAI might be good for creators around the world.
Why there is now a risk that investors lose faith in AI, leading to another ‘AI winter’ when funding dries up and AI developments are put on hold. “Winter is coming,” warns Maria.
◼️ More reaction on the GPT-5 fiasco in Friday’s Weekly Newsletter. If you’re receiving this post but not the main Weekly Newsletter then please check your spam folder or get in touch.
