Skip to content
View Scan-law4's full-sized avatar
4 followers · 3 following
  • Zagreb
  • 10:36 (UTC +02:00)

Block or report Scan-law4

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Scan-law4/README.md

Neven Sirotić (Scan)

GitHub Sponsors

DOI: Law-4 Framework

DOI: 4-Beat Reflex

Independent researcher working at the intersection of AI ethics, agent architecture, and autonomous intelligence.

Work

Law-4 Compliant™ Framework — Four immutable ethical laws for autonomous AI: Truth Integrity, Risk & Consent, Responsible Self-Improvement, and Core Immutability.

DOI

LIMA — Locally Integrated Multi-Agent — A journal of a locally-running multi-agent system: specialized AI agents with persistent memory and self-written skills, running on consumer hardware.

Current Research

The 4-Beat Reflex for Grounded AI — A reflex injection pattern that interrupts the fluent-answer impulse before the first token: Stop. Search. Cite. Answer. Each beat maps to an operational gate of the Law-4 Compliant™ Framework. Currently deployed in production on a fleet of specialized Hermes agents (LIMA), running local RAG pipelines with semantic search via Hermes/Llama.cpp.

Active problems:

  • Response latency asymmetry: the fluent-answer impulse consistently outruns the verification procedure, even when a verified ground-truth source is available in context
  • VRAM-bound reflex injection: benchmarking the 4-beat reflex across local quantizations to establish minimum-viable hardware profiles
  • Continuous API evaluations: automated regression testing of the reflex against model provider API changes

Compute & Funding

This is self-funded, independent research — no institutional affiliation, no VC backing.

To continue benchmarking hybrid local agents (reflex injection + semantic search + structured memory), I need:

  • GPU compute credits for running reflex benchmarks across quantization levels and VRAM constraints (currently on a single RTX 3090)
  • Micro-grants for documentation, reproducibility packaging, and open-access publication fees
  • Community sponsorships via GitHub Sponsors to sustain independent, public-interest AI safety work

If your organization provides compute grants for open-source AI safety research, or you'd like to sponsor this work: scan@law-4.com

Contact

Popular repositories Loading

  1. obsidian-wikilint obsidian-wikilint Public

    Check an Obsidian vault for broken [[wikilinks]] — exact resolution, pre-commit ready, CI verified.

    Python 1

  2. law-4.com law-4.com Public

    An Ethical Framework for Autonomous Intelligence

    HTML

  3. Scan-law4 Scan-law4 Public

    Independent researcher — AI ethics, agent architecture, and autonomous intelligence

  4. 4-beat-reflex 4-beat-reflex Public

    The 4-Beat Reflex for Grounded AI: Stop. Search. Cite. Answer. — A technical companion to the Law-4 Compliant™ Framework.

  5. .github .github Public

    Organization-wide GitHub configuration for Scan-law4