ADI
An AI system architecture design philosophy whose core principle is the strict separation of a language model's probabilistic reasoning from deterministic operational responsibilities—such as memory access, identity management, permissions, execution rights, and verification—with the latter governed by an independent governance architecture.
Timeline (last 90 days)
Researcher Branden Laskowski (Cortex Agentics Global) published the first research paper on Artificial Deterministic Intelligence (ADI).
The core claim of ADI is to completely separate probabilistic reasoning from externally enforced deterministic authority, letting language models focus on reasoning generation while delegating memory access, identity, permissions, execution rights, verification, and publishing to an independent governance architecture.
ADI experiments were conducted on synthetic historical memory tasks, with 30 task types designed, each configuration repeated three times, for a total of 90 runs.
Under the constrained context configuration, the 8B model achieved exactly the same perfect score as the 35B model (both 90/90).
The 35B model achieved a pass rate of 90/90 under the constrained context, while the 4B model achieved 72/90.
When the 35B model switched to a wider context, its pass rate plummeted to 46/90, with 43 responses truncated due to a fixed output budget.
The experiment is a first-party synthetic study, designed and conducted by the author's team, has not yet been independently replicated, and proprietary implementation details have not been made public.
The preprint of the first ADI paper has been published via Zenodo, with DOI 10.5281/zenodo.23265200.
The author plans to advance the ADI research series at a pace of one evidence-reviewed paper per week.
The related documentation and complete research series are hosted in a GitHub repository.
1 more timeline events
All Facts (11)
The open engineering question posed by the author is: how much operational responsibility can be migrated from a reasoning model to a governed architecture while maintaining the required task performance.
50%UnverifiedThe core claim of ADI is to completely separate probabilistic reasoning from externally enforced deterministic authority, letting language models focus on reasoning generation while delegating memory access, identity, permissions, execution rights, verification, and publishing to an independent governance architecture.
50%UnverifiedADI experiments were conducted on synthetic historical memory tasks, with 30 task types designed, each configuration repeated three times, for a total of 90 runs.
50%UnverifiedUnder the constrained context configuration, the 8B model achieved exactly the same perfect score as the 35B model (both 90/90).
50%UnverifiedThe 35B model achieved a pass rate of 90/90 under the constrained context, while the 4B model achieved 72/90.
50%UnverifiedWhen the 35B model switched to a wider context, its pass rate plummeted to 46/90, with 43 responses truncated due to a fixed output budget.
50%UnverifiedThe experiment is a first-party synthetic study, designed and conducted by the author's team, has not yet been independently replicated, and proprietary implementation details have not been made public.
50%UnverifiedThe related documentation and complete research series are hosted in a GitHub repository.
50%UnverifiedThe author plans to advance the ADI research series at a pace of one evidence-reviewed paper per week.
50%UnverifiedResearcher Branden Laskowski (Cortex Agentics Global) published the first research paper on Artificial Deterministic Intelligence (ADI).
50%UnverifiedThe preprint of the first ADI paper has been published via Zenodo, with DOI 10.5281/zenodo.23265200.
50%