@_NOWFORTHIS TRAVEL : https://xitter.gq/t.co/cKrjoeDYF3 VIDEO: https://xitter.gq/t.co/ZPMoiiZJrj NFT : https://xitter.gq/t.co/TiKohoYxqX MUSIC : https://xitter.gq/t.co/sogXkrSKTt
Philadelphia
Joined November 2013
- Tweets10.8K
- Following4.8K
- Followers1.1K
- Likes14.6K
Pinned Tweet
ProofRail is evidence-gated R&D
If evidence fails, downstream authority is recalculated, the blocker is named, and the transition is recorded.
Try it, break it, tell me where it fails — and if you think the problem is worth solving, an upvote helps:
app.emergent.sh/showcase/bui…
1
20
The New Flower Power: Beinghood Without Permission is out now as a free audiobook on ElevenReader.
It’s about what happens when systems of description gain more authority than their evidence deserves .
Listen free: elevenreader.io/audiobooks/t…
1
7
Most systems track whether evidence exists.
ProofRail tracks what that evidence is currently allowed to authorize.
I wrote up the argument for evidence-gated promotion, recursive authority recalculation, and downstream revocation:
medium.com/@renaefclark/auth…
12
REN TULPA / TULPA.ETH retweeted
Harnesses often get dismissed as just scaffolding, just prompt engineering, and not real research. But that couldn't be farther from the truth. The same model weights that score 30% on ARC-AGI score 95% with a better harness.
So we gathered a group of researchers and founders working at the frontier to do a deep dive into the state of harnesses.
We cover how we got to this point, the case for making your harness as expressive as possible, and what YC learned building an agent for every employee in the company.
00:00 - @FrancoisChauba1: Why harnesses matter
04:27 - Building an auto-researcher by accident
07:13 - A five minute history of harnesses
13:56 - Self-improving harnesses
18:35 - @sethkarten: Prime Agent, a self-improving RLM harness
21:50 - Context as an L1, L2, L3 cache
24:51 - From Turing machine to von Neumann computer
28:33 - Messaging between agents
30:04 - ARC-AGI results
33:09 - Emulator Bench and GPU kernels
37:30 - @JonSaadFalcon: OpenJarvis, personal AI on personal devices
38:26 - How far behind are local models
39:21 - The five primitives of a personal AI stack
42:47 - Letting cloud models optimize your local stack
43:53 - 800x cheaper than the cloud
45:58 - @josh__france and @jbellregan: QM, YC's agent harness for work
47:29 - A history of YC's internal agents
49:24 - OpenClaw and a fleet of 50 agents
51:04 - Pulling the brain out of the sandbox
54:43 - Letting the agent choose its own sandbox and model
57:16 - The grind tool: budgets on goals
58:50 - Agents don't understand social context
ProofRail is live for Builder Fest.
If evidence-gated authority is a problem worth solving, I’d value your feedback.
And if ProofRail earns it, I’d greatly appreciate an upvote:
app.emergent.sh/showcase/bui…
18
I want people working with AI agents, R&D, scientific software, evaluation, provenance, research infrastructure and autonomous systems to try to break ProofRail.
Find a path where unsupported evidence can become authoritative.
That is the test.
16
ProofRail is not anti-AI.
It is infrastructure for using AI aggressively without allowing generation itself to become authority.
AI can generate.
AI can suggest.
AI can reason.
AI can propose.
It cannot silently promote itself into truth.
14
A ProofRail result does not become trusted because it is newest.
It earns authority from required evidence.
Dependencies remain explicit.
If supporting evidence fails, downstream authority changes with it.
The blocker stays visible.
The history stays auditable.
17
AI dramatically increases proposal throughput.
It does not automatically increase verification throughput.
Proposal volume can now exceed verification capacity by orders of magnitude.
ProofRail exists to preserve the structural difference between:
PROPOSED
and
SUPPORTED.
2
15
A calibration is invalidated in April.
A notebook that depended on it survives.
A summary inherits the result.
A slide inherits the summary.
By September, engineers are building against it.
Nobody lied.
The retraction simply never propagated.
16
Newer is not the same as truer.
More confident is not the same as better supported.
Automated is not the same as authoritative.
Authority should be earned from evidence.
14
For years I focused on creative work.
Lately that same instinct has moved deeper into systems, computation and AI.
I’ve just built ProofRail — an evidence-gated R&D system designed to stop stale or unsupported results from silently becoming trusted.
More soon...
1
77
AI can propose. It cannot silently promote itself into truth.
Built for Builder Fest.
Try it, break it, tell me where it fails — and if you think the problem is worth solving, an upvote helps:
app.emergent.sh/showcase/bui…
4
Demand Restitution from Spielberg for Shark Defamation - Sign the Petition! chng.it/sMRK5YZNNj via @Change
18