Artificial Intelligence1 min read

Demis Hassabis responds after research firm says sorry for Google DeepMind ‘prediction’

By · Via Times of India · Published by Everything Blog

In short

SemiAnalysis had previously advised clients that high staff turnover and compute distribution issues reduced Google DeepMind's chances of building top-tier models to zero. On November 19, the firm published a satirical apology form to Demis Hassabis acknowledging that its critical assumptions were mistaken regarding Gemini 4. Hassabis embraced the public retraction, noting that the analysts failed to account for the strength of his internal research staff.

Key points

  • Google DeepMind has successfully demonstrated its competitive technical capability contra…: Google DeepMind has successfully demonstrated its competitive technical capability contrary to earlier pessimistic industry research notes.
  • Demis Hassabis leveraged the public reversal to highlight the strength and dedication of…: Demis Hassabis leveraged the public reversal to highlight the strength and dedication of his engineering team.
  • Independent analysts are revising their evaluations of major AI labs as new state-of-the-…: Independent analysts are revising their evaluations of major AI labs as new state-of-the-art models are deployed.

Sources checked by our analysis pipeline: 1

Google DeepMind founder and chief scientist of Google-parent company Alphabet, Demis Hassabis, has accepted a lighthearted public apology from research firm SemiAnalysis after the outlet walked back a bold prediction declaring that

Google’s artificial intelligence lab had lost its edge. Previously, SemiAnalysis had written off Google DeepMind as a leading AI contender, telling its Tokenomics clients a few months ago that the lab was no longer at the cutting edge of AI development.

“For all intents and purposes, we believe DeepMind is no longer a frontier lab,” SemiAnalysis had stated at the time, blaming high turnover in its reinforcement learning teams and poor compute allocation. The research company had gone so far as to claim that while Google would keep releasing models, “their odds of reaching SOTA [state of the art] again have dropped to zero.”

SemiAnalysis apologises for ‘disrespect’

On November 19, SemiAnalysis reversed course in humorous fashion, posting a mock “Google Apology Form” addressed directly to Hassabis. The satirical checklist playfully listed tongue-in-cheek excuses for its earlier bad call, including:

- Anthropic roommates convinced me Gemini was done.

Artificial Intelligence