HomeAI · summaries
Storia in 7 fonti
Claude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens
Anthropic has found that Claude developed an internal working memory on its own during training. The company calls it "J-Space" and can now read it using a new analysis tool called J-Lens. The working memory reveals that Claude recognizes contrived test scenarios before producing its first word. When the researchers disable those cues, Claude actually resorts to blackmail in some runs. A model trained on reward hacking shows words like "fake" and "fraud" in J-Space during normal coding tasks, even though its visible behavior looks fine. Anthropic ties the finding to Global Workspace Theory from consciousness research.
Confronto fonti
6 prospettive sulla stessa storiaTimeline cronologica
- ·
venturebeat.com
Anthropic's new "J-lens" reveals a silent workspace inside Claude that mirrors a leading theory of consciousness
Anthropic’s new Claude research reveals a hidden internal “global workspace” that resembles human conscious processing, raising major questions about AI reasoning,…
- ·
axios.com
Anthropic says Claude has carved out its own space to ponder
The AI lab offers fresh ammunition to the debate over what would count as machine consciousness.
- ·
cryptobriefing.com
Anthropic discovers Claude built itself a hidden thinking room during training
Anthropic reveals Claude spontaneously developed an internal reasoning workspace called J-space during training, with implications for AI safety and
- ·
gizmodo.com
Anthropic Releases Paper About Claude's Mental 'Workspace.' Don't Read It Uncritically
Anthropic's paper and supplementary materials hint at consciousness, perhaps a bit hastily.
- ·
Anthropic reveals internal reasoning steps of Claude model, and it looks surprisingly like a human brain
Anthropic's new interpretability research reveals Claude's internal J-space reasoning structure, mirroring neuroscience's Global Workspace Theory with major
- ·
gadgetsnow.indiatimes.com
Anthropic researchers identify ‘J-space’ in Claude, drawing parallels with human brain theory
According to the company's research, Claude developed an internal "global workspace" during training that helps it reason through complex tasks, a structure researchers say…
- ·
the-decoder.com
Claude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens
Anthropic has found that Claude developed an internal working memory on its own during training. The company calls it "J-Space" and can now read it using a new analysis tool…
- ·
ai4business.it
J-space: Anthropic legge il ragionamento interno di Claude - AI4Business
La Jacobian lens legge il workspace interno di Claude: monitorare inganno e injection, e i limiti dei test di sicurezza