The hype surrounding Chat-GPT is cooling down a bit, but that also means: things are getting serious now. The technology needs to show what it can do in practice and whether the large sums of money that have been spent on it for just over a year are justified. What can Chat-GPT do, so what can generative AIs really do? Given the abundance of anecdotes, papers, blatant Twitter screenshots, studies and claims, it’s difficult to keep track. Thankfully, innovation person Johannes Klingebiel (who once worked with me at the same publishing house) has created a concise overview in his newsletter.
Above all, it shows the limitations that Chat-GPT still has: too many errors, too unreliable and fluctuating, yet inflexible for “surprising” tasks that the models are not yet familiar with. Like so many forms of automation, AI is likely to need human “minders”, which puts the idea that we are dealing with real intelligence into perspective. In addition, productivity suffered among the highly qualified – at least among management consultants.
Conclusion: GenAI is fascinating, can increase productivity – but you wouldn’t trust it with really important tasks.