DEEPMIND
DeepMind’s CEO just said the quiet part out loud
Google DeepMind CEO Demis Hassabis has warned that AI’s biggest problem right now is inconsistency.
Speaking on the Google for Developers podcast, he said even the smartest systems can win top maths competitions but still mess up simple school problems, something that needs fixing before we reach true AGI.
Hassabis described today’s AI as having “jagged intelligences”, great in some areas but weak in others.
He said the issue isn’t just about adding more data or computing power; it’s about missing skills like reasoning, planning, and memory.
He called for tougher tests to better show where AI still struggles.
Three things to note:
Hassabis says AI’s “jagged” skills show intelligence isn’t consistent across tasks.
Fixing this needs better reasoning and memory, not just more data.
Both Google and OpenAI agree AGI is still years away despite fast progress.
Robots have off days too
Although he’s previously said AGI could arrive within 5–10 years, Hassabis admitted there are still big challenges.
His view matches OpenAI’s Sam Altman, who recently said GPT-5 still can’t learn continuously, something needed for true AGI.
The takeaway: before AI can think like humans, it needs to stop making basic mistakes.
I’m convinced every time AI messes up a simple task, a data scientist cries.- MG