719 karma · joined October 20, 2013
e.g.: gh pr merge 5062 -R <org/repo> --squash --delete-branch
The jobs to harvest the corpus needed to create each agent is not scaling well.
Note: crohns.ai: 3,948 gastroenterologists, dietmanager.com: 1,752 RDNs
Thats the main thing. I think that Ai-native governance by domain experts is how AI reaches its full potential. Not in theory but in terms of the value it delivers to populations via outcomes.
AI-native Governance is like irrigation for the outcomes populations want to achieve, starts with intents; executed on by programs that use protocols as guardrails. This is a gross oversimplification of the process but based what i see from your work you will get the abstract.
Examples -
dietmanager.com - RDN governance
crohns.ai - AGA (MD/GI) governance
<city>.us.codify.city - City council Governance
https://san-francisco.ca.us.codify.city/
https://new-york.ny.us.codify.city/
http://chicago.il.us.codify.city/
Even applies to YC: https://openyc.orgEach codify.* is a PDA [Public Domain Agent] - that gets delegated intents per request and has to manage its own "deal" - its also managed democratically via ontology and downline policies set by the experts in said field and has feedback loop to verify/optimize policy outcomes.
This concept applies to everything IMO, and I cannot say I fully understand it but im absolutely obsessed with the exploration of the idea; again - in practice not theory. I have real outcomes in healthcare, education and housing.
Secondly, I do not agree that they think issues will "disappear"; I believe that they simply think this gives them license to impose their will on the groups that identify with said banned terms.
Edit: I no longer think me being a black male is relevant in this context because after looking further into the list it basically covers multiple terms that apply to everyone; thus I think context of application is even more important and the list in itself is fairly irrelevant.
Intent: https://wiki.crohns.ai/agent/posts/ibd-biologic-switch-decis...
Program: https://crohns.ai/program/71168-biologic-therapy-initiation
Protocol: https://crohns.ai/protocol/71168
If given the chance, I might go back on it because my protocol can be a little strict at times but either way I do see a significant shift to tools like this given the state of the US Healthcare system.
The 1st 2 injections where done by a nurse that came to my home, the others were done as self injections using their njection kits.
A) no education
B) no resources
C) not smart enough to be a self-taught bio-hacker
Everyone hears "AI is going to cure disease" and pictures some cure-all pill from a bio lab which is what I feel this paper is hinting at is missingb but that's the top of the funnel; I'm at the bottom where patients live and that is where AI is already quietly working. Its just not being benchmarked.
I built https://crohns.ai. I set out to make an AI-native clinical-trial manager with a feedback loop (DDP) and ended up somewhere completely different: instead of chasing a new "drug" which is totally out of my grasp; financially, intellectually etc... I used it to codify a care protocol that helped me avoid a flare after I got laid off, lost my insurance, and lost access to Skyrizi.
Right now I am working on connecting it to community resources so that it can provide more of an overarching governance for re-entry that can be audited as needed for the application of interventions.
https://ashland-fci.bop.doj.dev/
https://ashland-fci.bop.doj.dev/program/65482-psychology-ser...
Medicare, DHS, DOJ - these systems are not actually that complex; taking a first principals approach; they are really just complex multi party workflows that need to follow policy based on jurisdiction etc... All of the back and forth you describe above is a distraction that has stopped people from just doing the thing but I have found that it is possible to just do the thing. Especially since its all documented so well, you can use AI to translate legislation into specs fairly easily.
TY for taking a look, right now I am digitizing all of the federal gov programs and all of the state gov programs, I would really love to connect with anyone just to talk, learn, show them what i'm doing etc...
Again, I really appreciate you taking the time to take a look and comment. It really means a lot. Gov Tech is not "sexy" part of tech where people flock and love to engage.
I have been working on this https://opm.dev/programs - its obviously not something that can get funding or will be a huge popular thing but it has been incredibly fun and i have gotten to learn a lot about Gov., agents EVAL's etc..
Note: This is one of 5 portals, same codebase mostly but the others are more developed and a few have some programs that have a 90+ % success rate.
Note: Here https://big-sandy-usp.bop.doj.dev/program/66366-sick-call is a much more developed example program that has been completed 6 times by inmates at Big Sandy. Super proud of this since I was locked up w/ 2 of the guys and dont know the other 4.
Each "program" here is a government program, agents orchestrate everything including the collaboration between all parties required.
High points: I have been able to help over 100 people get housing with no HITL on my side.
Note: Each host/subdomain is a project, they all inherit policy from each other and that drives the program generation and orchestration layer. Policies can be managed for the diff agencies at rnc/dnc.dev
Fair question, and the "training course" read is not an accident; it's the same shape underneath. A program is an ordered chain of modules, each with a declared actor and typed inputs/outputs. Courses are also that. So it renders with the same components. The sequence diagram you found isn't a sample scenario, it's the deal template's actual step graph; the thing an instance runs on.
Three authored files per domain:
- an ontology: the domain's vocabulary, its regulatory frameworks with real citations, the O*NET occupations that staff it, the systems of record it touches
- intents: what a person actually shows up wanting ("open a gun store, need an FFL"), with typed parameters
- deal templates, one per intent: ordered pipeline_steps, each with an actor, inputs, outputs, and a policy_check
The page you clicked is generated from the last two deterministically. No model in that path.
The part that speaks to your TLA+ instinct: I deliberately don't formalize what the law means. I formalize the procedure, and bind each step to the provision that authorizes it. Formalizing semantics is exactly where you get the creative, detached interpretations you're worried about, because every gap gets filled by the model's guess. Formalizing procedure asks the model to transcribe and cite, which is checkable:
- every step input is a ref; param:x, step:3.some_output, system:NICS.event; and it has to resolve. A step: ref must name an earlier step's declared output, so the dataflow is a DAG with referential integrity.
- every step's policy_check must name a framework declared in the ontology. A step that no provision authorizes fails validation.
So most hallucination becomes a build error instead of a plausible sentence. That's the whole trick. Not a smarter model; a narrower artifact.
Concretely, since you're right to expect slop: my first pass at four new agencies came back with 100% of step inputs referencing parameters that didn't exist, and prompts that literally said "Subject?". The validator refused all forty programs. That's the mechanism working; I'd have merged them on a read-through.
Intents and flows for ATF, if you want to see the layer under the program page: https://wiki.doj.dev/agent/atf
Limits, since you'll ask. It decides nothing; no adjudication, and consequential steps are human-gated. It's also not a formal method: the invariants are referential integrity and citation binding, not model checking. The genuinely temporal parts are the deadlines, and those do bite; the NLRB's six-month charge window runs from filing and service, with service being the filer's own duty, so a filing-date-only clock computes the wrong date on a deadline that destroys the claim if you miss it.
Re: the sibling comment about discretion; that's the actual pitch. Discretion hides in the gap between the written rule and the practiced procedure. Writing the practiced procedure down, with a citation per step, is what makes the gap visible.
https://ice.dhs.dev/program/13732-human-trafficking-investig...
https://atf.doj.dev/program/44825-open-gun-store-need-ffl
LMK if you want to know more.
Example: https://atlanta-fci.bop.doj.dev/program/65545-inmate-account...
The idea is this: the commodification of intelligence requires the codification of domain expertise to function reliably, and proper governance to scale.
Taste, in my opinion, is a fill-in for absent function and lack of utility. Systems function; they don't require taste. Taste is why PG put Sam Altman where he did — not a system of merit, not evidence. Taste is why DJT can do no wrong in the eyes of his base. It is unprovable and has no true measure, and that is exactly why it gets leaned on so heavily: it lets you claim superiority no matter what slop your system produces.
You say the box moved taste up a level. Out of the making and into decision layer. That relocation is a smell. Your unfalsifiable thing conveniently gets moved to the one place it still can't be checked, at precisely the moment the making became checkable. Just like: Iran/Strait of Hormus, AI curing diseases and all of the other bold promises made by people who claim to be "taste makers" but in the end its all the same. Bullshit! Because thats what taste is a euphemism for in this context. Convincing others that you are correct in some way with no actual proof to back it up.
The vector here should not be "taste"; it's tolerances: written down, measured, enforced. The judgment gets encoded into a spec, and the spec is what survives the person who had it. That is codification; not "taste".
1. Make Solar Energy Economical — https://github.com/orgs/HardisonCo/projects/194
2. Provide Energy from Fusion — https://github.com/orgs/HardisonCo/projects/206
3. Develop Carbon Sequestration Methods — https://github.com/orgs/HardisonCo/projects/196
4. Manage the Nitrogen Cycle — https://github.com/orgs/HardisonCo/projects/197
5. Provide Access to Clean Water — https://github.com/orgs/HardisonCo/projects/195
6. Restore and Improve Urban Infrastructure — https://github.com/orgs/HardisonCo/projects/193
7. Advance Health Informatics — https://github.com/orgs/HardisonCo/projects/203
8. Engineer Better Medicines — https://github.com/orgs/HardisonCo/projects/200
9. Reverse Engineer the Brain — https://github.com/orgs/HardisonCo/projects/205 10. Prevent Nuclear Terror — https://github.com/orgs/HardisonCo/projects/204
11. Secure Cyberspace — https://github.com/orgs/HardisonCo/projects/201
12. Enhance Virtual Reality — https://github.com/orgs/HardisonCo/projects/202
13. Advance Personalized Learning — https://github.com/orgs/HardisonCo/projects/199
14. Engineer the Tools of Scientific Discovery — https://github.com/orgs/HardisonCo/projects/198
Specs and the per-challenge process lists: https://github.com/HardisonCo/opendl
It should also be funded by the Gov. IMO and 100% for oublic benifit e.g.: nsf.dev
John W. Spiegel: A prominent business litigation and securities lawyer who graduated from Yale University and worked as a partner at the law firm Munger, Tolles & Olson.
This comment will be flagged soon and disappear just like anything else that does not support their "Trumpish" view of what the world should be.
This is all new, so I may be a bit hyperbolic, but the reason OpenAI introducing ads bothers me is the implicit (or even explicit) bias that can be smuggled into a chat in ways that simply aren't possible when you're just clicking through to an external source. There are all kinds of implications to Google no longer being that source of truth, even by default. Maybe this has quietly been the case for a long time, but this feels like the final move — pushing their ad bias (i.e., whoever paid the most) into a conversational system, where dark patterns are far easier to implement and much harder to detect.
One answer to this might be domain-specific agents — narrower, accountable, ideally something you (or your community) actually run. But even then it all falls back on trust: you being a good-faith actor, and others trusting that you are one. Which is to say, we're back to the same problem, just at a smaller scale.
https://codify.nyc is the one I am going to be launching first, hopefully in a few weeks. I only have 100 or so cities on board and live right now. They have been very useful in understanding all the mechanics and nuance of delivering services at the city/local level.
Your project looks interesting, let me know if you see any place we could work together.