HackerTrans
TopNewTrendsCommentsPastAskShowJobs

JakubCwi

no profile record

Submissions

Why LLMs Lie in Different Ways?

orsonai.com
1 points·by JakubCwi·4 months ago·0 comments

Comparing confabulation signals across 8 LLMs

orsonai.com
2 points·by JakubCwi·4 months ago·0 comments

5-shot in-context examples are enough to blind your LLM

researchgate.net
1 points·by JakubCwi·4 months ago·0 comments

Why LLMs stop learning from examples in long contexts: The ICL Collapse

orsonai.com
3 points·by JakubCwi·4 months ago·0 comments

Paper: Detecting hallucinations before the first token

researchgate.net
4 points·by JakubCwi·4 months ago·0 comments

AI models know they are guessing before they generate text

orsonai.com
4 points·by JakubCwi·4 months ago·1 comments

[untitled]

1 points·by JakubCwi·5 months ago·0 comments

[untitled]

1 points·by JakubCwi·5 months ago·0 comments

The containment box that destroys the AI if it tries to think its way out

redact-app.com
3 points·by JakubCwi·5 months ago·2 comments

Hashgraph Detection Key – If you touch it, it breaks

redact-app.com
2 points·by JakubCwi·5 months ago·0 comments

AI box that fights back

researchgate.net
2 points·by JakubCwi·5 months ago·1 comments

10⁻²³ hurts AI

redact-app.com
1 points·by JakubCwi·5 months ago·1 comments

comments

JakubCwi
·4 months ago·discuss
[dead]
JakubCwi
·5 months ago·discuss
Ha! That's fair :) RLHF is a polite suggestion. Real containment needs physics.
JakubCwi
·5 months ago·discuss
Author here. TL;DR:

- Paradigm asymmetry: Classical impossibility proof holds only under same-paradigm assumption — drop it and it doesn't apply - Complexity gap: BPP ≠ BQP → quantum gatekeeper states are computationally opaque to any classical agent, regardless of intelligence - Pain mechanism: Barrier is embedded in the agent's substrate — breach attempts cause decoherence in the attacker's own reasoning, not in some external wall - Formal model: Seven theorems, six attack vectors, defense-in-depth → P(breach) ≈ 10⁻²³ - Standard physics only: No exotic QM, no speculative extensions — complexity-theoretic assumptions + standard decoherence dynamics

Summary: https://redact-app.com/publications/quantum-containment.html
JakubCwi
·5 months ago·discuss
Author here. TL;DR: • Paradigm asymmetry: Classical impossibility proof holds only under same-paradigm assumption — drop it and it doesn't apply • Complexity gap: BPP ≠ BQP → quantum gatekeeper states are computationally opaque to any classical agent, regardless of intelligence • Pain mechanism: Barrier is embedded in the agent's substrate — breach attempts cause decoherence in the attacker's own reasoning, not in some external wall • Formal model: Seven theorems, six attack vectors, defense-in-depth → P(breach) ≈ 10⁻²³ • Standard physics only: No exotic QM, no speculative extensions — complexity-theoretic assumptions + standard decoherence dynamics

Summary: https://redact-app.com/publications/quantum-containment.html

Full paper: https://www.researchgate.net/publication/400670258_Self-Refe...