9cc4989fc1
- 23 new institutions pages (alby, anthropic, blink-wallet, carnegie-mellon, deepmind, epoch-ai, fincept, google-cloud, hugging-face, lightning-labs, linux-foundation, mit, moonshot-ai, mozilla-foundation, openai, openai-superalignment, openrouter, polymarket, stanford-ai-lab, tu-muenchen, xai, z-ai) - institutions-ingest-runner.md script - modified concepts (agi, hardware, llm) + tools pages - wiki-lint-report update
28 lines
1.2 KiB
Markdown
28 lines
1.2 KiB
Markdown
---
|
|
created: 2026-06-24
|
|
updated: 2026-06-24
|
|
sources: [`raw/youtube/2026-06-16_deepmind-two-steps-ahead.md`, `raw/youtube/2026-06-18_deepmind-beyond-transformer.md`]
|
|
tags: [institution, deepmind]
|
|
---
|
|
|
|
# Google DeepMind
|
|
|
|
## Profil
|
|
|
|
| Feld | Wert |
|
|
|------|------|
|
|
| Typ | KI-Forschungslabor (Alphabet Inc.) |
|
|
| Fokus | Post-Transformer-Architekturen, hybride Lernmodelle, Weltmodelle (VEO) und wissenschaftliche Durchbrüche (AlphaFold). |
|
|
| Wichtigste Personen | Demis Hassabis |
|
|
| Zugehörige Tools/Modelle | Griffin, Recurrent-Gemma, Titans, Gemma 4 Edge |
|
|
| Primäre Quellen | `raw/youtube/2026-06-16_deepmind-two-steps-ahead.md`, `raw/youtube/2026-06-18_deepmind-beyond-transformer.md` |
|
|
|
|
## Fokus & Aktivitäten
|
|
|
|
Google DeepMind kombiniert Alphabets führende Forschungs-Ressourcen. Im Sommer 2026 verfolgt DeepMind eine strategische Gegenposition zu reinen Auto-Regressiven Scaling-Labs wie OpenAI/Anthropic und setzt auf vier technologische Säulen: Hybride aus Attention und Recurrence, Diffusions-LLMs, physische Weltmodelle (VEO) und Continual-Learning-Systeme.
|
|
|
|
## Cross-References
|
|
|
|
- [[../architecture/deepmind-beyond-transformer.md]]
|
|
- [[../concepts/llm/post-transformer-llm-architectures.md]]
|
|
- [[../concepts/world-models-jepa-vs-generative.md]]
|