Wire flash
Emad Mostaque: AI models have reached human brain efficiency; Qwen 27B runs on Raspberry Pi
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
In a post on X, Emad Mostaque, a prominent AI figure, claimed that AI models 'have pretty much reached the efficiency of a human brain.' He cited a specific achievement by a team from Prism ML, which reduced the memory requirement of Alibaba's Qwen 27B model from about 60GB to just 6GB. This was accomplished by storing the model's numbers using 1.5 bits instead of the standard 16 bits, while retaining 95% of its original performance. Mostaque highlighted that at 6GB, the model can run on a Raspberry Pi, a device that consumes roughly the same amount of energy as the human brain. The post includes a link to a full video on Tom Bilyeu's YouTube channel for further discussion.
Source report
Emad Mostaque has stated that AI models "have pretty much reached the efficiency of a human brain."
Breakthrough in Model Compression
A team from Prism ML achieved a significant reduction in memory requirements for Alibaba's Qwen 27B model:
- Original memory requirement: Approximately 60GB
- Reduced memory requirement: 6GB
- Method: Storing numbers with 1.5 bits instead of 16
- Performance retained: 95%
Practical Implications
At 6GB, the model runs on a Raspberry Pi, which Mostaque notes "uses as much energy as your brain."
Full video available on the "Tom Bilyeu" YouTube channel (link in comment).
Source
rohanpaul_aiNeutral / independent