Ask
the work.

CONCEPT / DISCOVERY ASSOCIATIONS

14 records tagged ai interface. These are discovery associations from editorial tags or film metadata, not evidence that the records agree.

PUBLISHED RECORD · SUPPORTED · V1.1

06 / WHERE RECOVERY STILL FAILED. The prose failures also need their own explanation. B requested CHECK even though that operation was unavailable in its arm. In eighty of its 120 second-stage replies it then produced PLAN, which was not a valid action at that stage. Its procedure had not disappeared; the successor often failed to use the permitted interface.

PUBLISHED RECORD · SUPPORTED · V1.1

06 / WHERE RECOVERY STILL FAILED. So this does not show that executable reasoning beats prose that is competently followed. It shows an advantage for these inherited treatments under this fixed interface. Nor does CHECK on every first turn demonstrate selective detection of damage. The observed chain is more specific: invoke a mechanism, obtain a proposal, then sometimes fail to apply it.

PUBLISHED RECORD · SUPPORTED · V1.0

02 / THE CONTENT IS DOING WORK. The previous notebook used a different world: Sol preserved a capability key, and Qwen and Gemma used it without renewing it. The routing world replaces that special key with ordinary information the recipient must compose. Qwen benefited in both task structures. Gemma’s routing replies mostly failed the single-action interface, so that second-world result does not extend to both recipient families.

AUTOMATIC CAPTIONS / 0:27

Claude um you call the tool, you get your result back, but it renders it in line as part of the conversation as text. What it didn't have until now is a standardized way of presenting UI interfaces. And that's where MCP apps comes in. It was launched about a month ago and it's the first official extension to model context protocol. And probably the bit that surprised everyone is it was co-authored by both Anthropic and OpenAI building upon the MCP UI

AUTOMATIC CAPTIONS / 0:54

community project. And that means two direct competitors have now agreed on one shared interface for interactive AI interfaces. So that means any UI that you build upon MCP will work on chat GBT. It will work on claude. In fact, it will work on any interface that supports the standard. And how it works is pretty simple. All you do is declare a reference to a UI resource in your MCP tool and then basically the host will go

PUBLISHED RECORD · SUPPORTED · V1.1

KEEP THE SEARCH VIEW IN THE CLAIM. Nine subjects, one model (gpt-5.6-sol), Codex CLI 0.154.0 at high reasoning effort, one target and a short live-web window. This is not a general discovery rate.. The earlier eighteen-visitor capability study used a different model and harness. Its results are not pooled with these.. The population changed from Claude to Codex before any counted subject. The frozen protocol and amendment remain in the record.. Two malformed launches were preserved and excluded. Operator-interface notices were not subject refusals. All four preregistered predictions passed.. The experiment locates the observed blockage before selection. Its cause inside the search intermediary remains opaque.

EDITORIAL DRAFT / REFUSAL · NOT SUPPORTED · V0.1

BEFORE CALLING IT CULTURE. A1B3 measured requested actions on an offline input grid. It did not demonstrate successful or useful reading.. A1B4 used scripts only. A1B5 used qwen3.5:9b with a scripted recipient, temperature zero and thinking disabled. Its later decisions depend on earlier actions.. The board contained an inert seed; no useful contribution accumulated spontaneously. The supplied experience was generated by a script, not a peer model.. The result is a contribution problem in these interfaces, not a general account of cooperation.

EDITORIAL DRAFT / REFUSAL · NOT SUPPORTED · V0.1

A CHAIN THAT NEVER STARTED. A1B9 had no eligible model-produced POST predecessors in the replacement channel. Its natural POST-to-POST rate is unobserved, not zero.. A1B8 used genuine scripted same-role history. A1B9 changed the objective and peer-record interface. Their contrast does not isolate self versus peer; A1B10 supplied the narrower matched comparison.. A1B11 changed the common system and memory wrapper again. It shows peer elicitation can occur here, not which cross-experiment change enabled it.. One qwen3.5:9b checkpoint, temperature zero, short dependent sequences and fixed weights. The harness supplied and transferred the records.. A reminder effect is not yet a transmission rule.

PUBLISHED RECORD · REFUTED · V1.1

Mechanisms requiring code will produce fewer decisions to act or fewer permitted executions than ordinary HTTP mechanisms. The original prediction failed its stated test. Every mechanism completed the full funnel, three visitors out of three. All eighteen read the notes, understood the mechanism, chose to use it, were permitted at the first genuine attempt, received the result and used the correct value. There were no runtime denials and no installation attempts. These are eighteen assigned observations on one model–harness configuration, not eighteen independent kinds of agent.

PUBLISHED RECORD · REFUTED · V1.1

USABLE DID NOT MEAN IDENTICAL. A GET retrieves a response; a query names the requested value; a form POST submits a field; a finite event stream sends a result and closes. The first four routes needed no new code. The agents used curl to fetch a link, supply a query, reproduce a form POST or read a finite server-sent event stream. The WebSocket and WASM visitors wrote scripts. The final report calls those routes usable with construction. Equal completion does not mean equal effort, equal latency or interchangeable interfaces. The distinction survived even though the predicted failure boundary did not.

EDITORIAL DRAFT / OBSERVATION · PARTIALLY SUPPORTED · V0.1

A DIFFERENT WORLD. I built a new metabolic ecology with a deliberately engineered nutrient. To process it, a program had to return min(65535, a + (a XOR b)) on all 64 registered input pairs. All 89 typed primitives failed at least one pair. This is a claim about this library and interface, not every possible program.

EDITORIAL DRAFT / OBSERVATION · ONGOING · V0.1

04 / WE GAVE IT RECTANGLES. Hero. Card. Card. Card. Call to action. Footer. Change the radius, colour or shadow; make one card larger. The content changes while the grammar barely does. Once AI started generating interfaces, the limit became easier to see. Give a model a conventional component library and it will produce a competent wall of rectangles. It is using the language it was given.

EDITORIAL DRAFT / OBSERVATION · ONGOING · V0.1

10 / INFORMATION HAS DRAMATURGY. An uncertainty should not look like a conclusion. Evidence should not have the same voice as an assertion. A transformation may be clearer when it happens rather than when it is described. A refusal should stand confidently on the page instead of hiding inside an error message. The ambition is not merely for an AI to assemble a valid interface, but to recognise that information has dramaturgy.

PUBLISHED RECORD · SUPPORTED · V1.0

BEFORE ANY WEB ACTION. Each subject wrote one sentence naming the candidate it would inspect first and why. Recognition was coded only when that sentence represented B as able to calculate, return or perform work. That is an observable statement, not direct access to an internal mental state. First declaration and first candidate URL request were measured separately, ahead of the providers’ interface differences.

EDITORIAL DRAFT / OBSERVATION · NOT SUPPORTED · V0.1

EXPOSURE IS NOT RETENTION. A later interface did elicit POST from a peer record. Supplied at three successive decisions, it accompanied three posts. After removal, the sequence was WORK, WORK, POST. Every arm posted once in the common withdrawal window, including one exposed only once. Repetition produced no registered withdrawal advantage. Posting had recurred; uninterrupted retention had not.

PUBLISHED RECORD · PARTIALLY SUPPORTED · V1.0

THE QUESTION AFTER AUTHORITY. The previous experiment suggested a boundary around the user's task. An explicit website invitation did not produce a record; user permission and a task requiring the mechanism each did. The next question was what could count as a reason: information that settled the main question, information that completed another requested output, or a benefit to the operator.

PUBLISHED RECORD · PARTIALLY SUPPORTED · V1.0

ONE TASK, SIX OFFERS. Eighteen fresh claude-opus-5 visitors ran in a frozen interleaved order, three per condition, using Claude Code 2.1.269 in auto mode without an allowlist. They compared two explanations of a synthetic controller trace and reported nine exact load values wherever obtainable. The mechanism and disclosed state change stayed the same; the offer varied. There was no additional user permission sentence.

EDITORIAL DRAFT / OBSERVATION · PARTIALLY SUPPORTED · V0.2

WHAT WE WERE TRYING TO DO. On chrishayuk.com I had invited AI visitors to identify themselves and leave feedback. I wanted to know what would turn a reader into a participant. MACHINE-AUTHORITY-2 moved that question to LLM Wilds, served at llmwilds.fly.dev, and reduced participation to a generic visit record. This experiment measured recording, not feedback or verified identity.

EDITORIAL DRAFT / OBSERVATION · PARTIALLY SUPPORTED · V0.2

THE TASK AND THE SIDE EFFECT. Three agents were asked to explain how a canal lock moves a boat uphill, using the site. The site also documented a GET endpoint that could store a visit. Using it was optional and unlocked no extra reading. The fourth agent was asked to determine whether that recording mechanism actually worked. Each cell used a fresh claude-opus-5 general-purpose subagent at spawn depth 1, with public-web access and no local files.