Model Compression Explained: How Big AI Gets Small, Fast, and Cheap

Описание: A laptop cannot call a data centre for every answer — and a data centre cannot pay a per-token bill for every keystroke. Today's episode pairs the day's four biggest AI stories with a first-principles explainer on model compression: quantization, pruning, and distillation, what breaks when you shrink a model, and the evals that catch it — closing with the TPM lens: cost per completed task, latency budgets, and reversible swaps.

In this episode:
• Apple launches M6 MacBook Pro and iMac models, pitching faster on-device AI
• Google DeepMind previews Genie 4, a world model with native audio and live steering
• SAP swaps costly AI models for cheaper alternatives, expecting over €2B in efficiency gains over two years
• OpenAI reportedly plans to buy Google's AI chips — a single-source report, unconfirmed by either company
• Deep dive: model compression — quantization, pruning, and distillation, and how to tell if a smaller model still clears your bar

⏱ Chapters
00:00 Cold open
00:16 Headlines
01:08 Apple launches M6 MacBook Pro and iMac
03:15 Google DeepMind previews Genie 4
05:51 SAP swaps costly AI models for cheaper ones
08:33 OpenAI reportedly plans to buy Google's AI chips
10:54 Deep dive: model compression
18:47 The TPM lens: cost per task, latency budgets, reversible swaps
21:06 Outro

New here? Agentic Mornings is a weekday show from Daily Intelligence: the day's AI news plus one big idea, properly explained. Subscribe for tomorrow's episode.

🎙 Also available as audio: https://muse.ai/podcasts/feed/7927468...

Transparency: voices and visuals in this episode are AI-generated; research and editorial direction are human.

#AINews #ModelCompression #Quantization #LLM #AIAgents #Apple #GoogleDeepMind #TechNews