The promise of an AI utopia is crumbling before our eyes

Telegraph

18 hours ago

When your company is valued at half a trillion dollars and promises an imminent utopia, you must not flop.
OpenAI discovered this the hard way last week. It had kept the world waiting two and a half years for a new large-language model, GPT-5, and unveiled it to great fanfare on Thursday.
But the reaction was so underwhelming, the company was immediately on the defensive.
'GPT-5 is the first time that it feels like talking to an expert in any topic, a PhD-level expert,' Sam Altman, OpenAI's messianic chief executive, had promised us. Alas, GPT-5 turns out to be better at some things – OpenAI claims that computer programming and maths have improved – but only some of the time, and that was not enough.
New 'foundation' models like GPT-5 are released every few weeks, and each one is a little more capable than its competitors in some way. Since ChatGPT is the best-known brand in AI – like Xerox or Google it has become a verb – the disappointment was far deeper felt.
'It doesn't feel like a new GPT whatsoever,' complained one user. 'It's telling that the actual user reception is almost universally negative,' wrote another.
Each ChatGPT update has been worse, wrote another user, and the endemic problems aren't getting fixed. Your chatbot still forgets what it is doing, contradicts itself and makes stuff up – generating what are called hallucinations.
GPT-5 remains as prone as ever to oafish stupidity, too. A notorious error where the chatbot insists there are two occurrences of the letter 'r' in the word strawberry has been patched up. But ask how many 'bs' are in blueberry? GPT-5 maintains that there are three: 'One in blue, two in berry'.
OpenAI also annoyed customers by removing the option of using its older models, prompting cancellations. It quickly reversed course, but the damage has been done. Confidence in OpenAI on the prediction markets – online forums where punters place bets for the question 'Which company has the best model at the end of August?' fell from 75pc to 8pc overnight.
'The model-building race is basically over,' concluded one strategic IT consultant.
Not all is gloom. One experienced computer scientist building a business around the capabilities of large-language models for software development just shrugs. These splashy new models are for the press and investors, he told me: coaxing something useful out of an existing model remains a largely unexplored field.
But that's not the narrative we have been given.
The most utopian AI advocates call themselves 'accelerationists', some using the abbreviation e/acc to signpost their enthusiasm.
But the conceit of accelerationism is that things are supposed to be getting faster, and not slowing down. By Friday, social media wits had turned the familiar ascending curve used by futurists upside down to illustrate how AI has plateaued.
Talk of 'superintelligence' now looks very silly.
So what does it mean? Generative AI is too useful to disappear, but a plateau has huge implications for OpenAI, the stock market, investors and policymakers. If AI isn't getting rapidly better, then businesses can wait until it gets cheaper, and it will. Only fools will rush into AI now.
OpenAI still loses money on every user. It will spend $13bn (£9.6bn) training its models this year, which is just about on a par with the $12bn annualised revenue it reportedly reached at the end of July, but then billions more dollars are spent hosting the service.
Markets may now ask why developing these AI models needs to be so expensive, when competitors can do so the same more cheaply. On Friday, Altman vowed to stick to the strategy – not something he would normally have had to defend. Next, it will be Nvidia's turn.
Compare OpenAI to ASML, the Dutch semiconductor fabrication company. If ASML disappeared overnight, the world's computing, telecommunications businesses would be set back by several years, for only ASML has cracked the science of extreme ultraviolet (EUV) lithography. But if OpenAI disappeared, we probably wouldn't even notice.
This isn't like the early days of steamships or aviation, where bigger and better engines led to larger vehicles that could go longer distances and carry more people. The 'race' metaphor journalists love doesn't really fit either.
Today, companies like AI can spend billions on model training and chips, and find their designs copied within weeks – or incorporated into royalty-free, open-source models. Chinese researchers, who are keen to make AI useful in their manufactured goods, have proven how easy it is to compete at a fraction of the cost.
In AI, the tiny mammals may well beat the lumbering dinosaurs.
When Altman was interviewed last week, he did something that he usually pulls off very well: deflecting scrutiny. With his big wet eyes turning to the horizon, he began to speculate about GPT-6, GPT-5's successor.
'What does it mean to discover new science?' he mused. 'Maybe we won't deliver that, but it feels …' he always pauses here for dramatic effect '… within grasp'.
But this time, he looked as tired as a beaten dog. Maybe Altman knows the game is up. For big-spending AI, it looks like it is almost over.