Archive · 13 Aug 2025
Google DeepMind CEO: AGI Still Years Away, AI Makes Simple Mistakes
The head of Google DeepMind, Demis Hassabis, thinks we're still far from AI that's as smart as humans.
Even though AI like Gemini can do amazing things, like get a gold medal in a math competition, it can still make silly mistakes.
Hassabis says just making AI bigger isn't enough.
We need to make AI better at things like thinking, remembering, and understanding the world around them.
He also wants harder tests to find any weaknesses.
Google DeepMind CEO Demis Hassabis says AGI is still years away.
Current AI, despite achievements, makes simple errors.
Hassabis cited Gemini, which excelled in the IMO but falters in basic math.
He argues scaling alone won't lead to AGI; stronger reasoning and memory are needed.
New, tougher benchmarks for AI are necessary, including physics and safety tests.
- Who
- Demis Hassabis, Google DeepMind CEO.
- What
- Google DeepMind CEO Demis Hassabis discusses the timeline of Artificial General Intelligence.
- When
- Recently, as of this writing.
- Where
- N/A
- Why
- To address current limitations in AI development and provide insights on the path to AGI.
Current AI Progress
AGI Timeline
Development Strategy
Current AI Progress
Current AI can be quickly advanced by simply scaling existing models.
AGI Timeline
AI requires more than scaling; fundamental capabilities need improvement.
- Key Figure
- Demis Hassabis
- Company
- Google DeepMind
- AGI Outlook
- Years away
- AI Weakness
- Simple Errors
- Current AI Model
- Gemini
- Focus Areas
- Reasoning, Memory, Safety