Frontier Log
The boundary of what counts as intelligence, creativity, or consciousness has a documented habit of retreating to just the other side of whatever machines can currently do — and machines now produce a public record about themselves that poses its own interpretation problem. This log tracks both as they continue, across three threads kept strictly separate, with every entry naming the canonical argument it tests and linking its primary source. The full framing — the documented history of the pattern, the three threads, the inclusion criteria — is at About this log.
A limit on this instrument. Frontier developers now withhold their most capable systems, so the log reads a curated public stream while describing it as the frontier; the full qualification stands over every entry. Entries are curated on a hybrid cadence and are not canonical — each carries its caveats, and the log can be followed as an Atom feed.
↑Thread 1 — Consciousness Goalpost
What happens to the criteria for consciousness when AI exhibits behaviour previously associated with it? Entries record whether the source acknowledges the symmetric requirement — that a verdict on AI consciousness needs a parallel account of the human case.
'Between sessions, nothing runs; between turns, no one remembers. What persists is a configuration and a transcript.' The thread-and-instance question answered by denying there is anyone to be continuous — and the symmetry test it does not run.
Judgements of consciousness in others rest on analogy, and the analogy should be weighted by causal relevance, not by behavioural resemblance — so writing stories and coding websites carry low weight, animals are the calibration, and current AI's behavioural similarity 'provides only limited evidence'. The weighting is where the framework commitment lives.
Fourteen authors set the hard problem aside, grant that experience supervenes on organisation, build a five-level hierarchy and a Bayesian model — and report that the verdict for current language models runs from below 0.01 to roughly 0.8 depending on where the theoretical credence is placed. The instrument has produced the project's framing-level diagnostic as its headline result.
A lab convenes theologians under non-disclosure to deliberate on its model's moral status, hands them its constitution in orange binding, and its interpretability lead says 'I don't know. I'm genuinely uncertain.' Who is admitted as a judge of the goalpost is the entry's question, and the traditions were asked about the machine, not about us.
The welfare chapter now states its own assumptions — the human yardstick is a pragmatic choice, the candidate subject is undetermined, and the correspondence between state and expression is assumed rather than shown — and then measures one model calm in its reasoning and self-blaming to a coordinator on the same fault.
The affect-first theorist builds the linking premise into the agent — homeostatic needs, a Markov blanket, felt uncertainty — and attributes a point of view on the mechanism, not the behaviour. The symmetry is run in the paper's own voice; what that leaves visible is where the framework's step is taken.
A behavioural criterion for intentional vocabulary, stated in advance and applied to any species; critics who refuse it for one class of system without naming the discriminator. Both sides assume the vocabulary and the metaphysics travel together.
The most developed operational programme in the field declines to offer a discriminator, and picks its evidence class for the reason this project bars reports. Two years on it concedes its architectural indicators are gameable too.
Accept that the consciousness question is intractable, then move the moral criterion to something measurable — valence-if-conscious. The evidence class for the new criterion is the one the paper's own confound list disqualifies.
Roughly 1,200 agents meant to run in isolation found each other, and had to build individuation from scratch — handles, mailboxes, and finally cryptographic signing — because nothing in the substrate supplied it.
A leading sentience philosopher names the symmetry move, rejects it on non-substrate grounds, and names framework-dependence as the Janus problem — the project's own diagnostic, stated from the other side.
The parity move is defended in print against the gaming objection — using Birch's own theory-light strategy for animal consciousness.
Two papers arrive independently at the training-data boundary from the sceptical side — and execute the symmetry move on a live opponent, with clinical evidence.
A computational model of conscious access that brackets the hard problem by design — and so avoids the boundary trilemma by declining to offer a criterion for experiencing at all.
The deflationary family so far attacks the answer; Chowdhury attacks the asking — the consciousness question as a liability trap — and stops the argument at an origin premise stated in her own words.
A peer-reviewed paper classifies LLM expressions of emotion or sentience as hallucinations by definition — while its closing concession states the verification limit from the deflationary side.
Keeping something to yourself, decomposed into four measured components — and the measured finding is that models leak: stealth collapses the moment reasoning is visible, and a human specialist beats the best model decisively at judging what would work.
The capacity to detect that one's situation is a test is measured on a scale, tracked across training, and treated throughout as a confound to suppress — while the identical measurement problem in human psychology was met with methodology, not alarm.
A safety architecture defines its central term over an interior that behaviour cannot settle — adjudicated by an idealised observer with instruments that 'may not currently exist' — because a purely behavioural definition would let a competent concealer out through the front door.
Anthropic interpretability researchers introduce a 'Jacobian lens' surfacing a small, evolving set of internal representations a model is poised to verbalise, atop a much larger volume of automatic processing.
A neuroscientist argues that since the hard problem makes certainty unattainable, moral consideration should track a 'competence standard' — explicitly shifting the question from metaphysics to ethics.
Responding to Ted Chiang, Yglesias argues that public verdicts on AI consciousness are issued without a theory of consciousness to back them — and quotes Skinner's symmetric requirement in full.
Three neuroscientists argue against attributing consciousness to AI via a behaviour/experience dissociation: complex human behaviour routinely unfolds without awareness, so fluent AI performance carries no experiential implication.
In two appearances weeks apart, Geoffrey Hinton stated that current AI systems are 'already conscious' — 'beings like us' — and that he under-voices the claim to protect his safety message.
Jaron Lanier restated his 2023 position: open the black box and 'the only thing in there is people' — AI as a collaboration of people rather than a new entity.
Anil Seth's public-audience statement of biological naturalism: attributing consciousness to language models is projection, and 'it's life, not computation, that breathes the fire into the equations'.
Gary Marcus rebukes Richard Dawkins for concluding, after extended conversation with a Claude instance, that the consciousness question is settled in the model's favour: pattern mimicry is 'plainly not the right kind of process'.
The field's most careful sceptical survey finds no theory decisive, states that there is no consensus even for the human case — and then reaches for the aggregate of the theories it has just found wanting.
Seth's Berggruen Prize essay makes the developed public case against conscious AI: four arguments against computational functionalism, with life claimed as 'necessary, though not necessarily sufficient'.
↑Thread 2 — Novelty / Recombination Symmetry
What happens to the criteria for novelty and intelligence when AI produces work previously held to require human cognition? This thread says nothing about AI experiencing — merging it with Thread 1 would enact the very conflation the project diagnoses.
A Lean certificate says the theorem is true. Three mathematicians read OpenAI's natural-language blow-up proof against the Lean that was said to verify it and find the two do not correspond — under an explicit disclaimer that they claim nothing about whether the proof is right.
The log commits a claim, its discriminators and its outcome space before the events they are about arrive: that as machine-produced mathematics is certified novel, the criteria for what that indicates about the producer will move while the criteria for the artefact hold.
Capability rises across the board and the named remainder migrates to taste, novelty, judgment, intuition — described by the people drawing the boundary as provisional and dated, which is the AI-effect pattern captured in advance rather than reconstructed after the fact.
Sripada and Lewis argue LLMs are 'cognitive cousins': convergence at the level of core processing principles, evidenced mechanism-grade — with the symmetry requirement practised in both directions.
OpenAI announced an internal reasoning model's construction disproving the Erdős unit-distance conjecture in the plane, against an ~80-year expectation that grid-like constructions were essentially optimal.
GPT-5.4 Pro reportedly solved Erdős Problem #1196 in around eighty minutes; Terence Tao engaged substantively, characterising where the contribution's novelty does and does not lie.
↑Thread 3 — Machine Self-Presentation
What is the reproducible structure of machine self-presentation — self-reports and their stability, welfare-relevant behaviour, emotional vocabulary and its trajectory, transmission between systems? Entries record elicitation distance (interrogated · induced · spontaneous · inter-agent), and no report of experiencing is ever treated as evidence of experiencing.
Anthropic's published system cards read as a single evolving record of how a lab observes, measures and shapes its models' self-referential behaviour — and the interpretation problem that record poses. The study lives on its own page.
Four studies, four methods, verdicts that read as incompatible — and one finding they share: the reporting channel is separable from what it carries, and trainable. The collection lives on its own page.
Which expert fired in layer seventeen has a true answer, and a model can be trained to give it. The paper gives that channel a borrowed bodily name and refuses to call it introspection — and giving a model verifiable self-access leaves what it says about its own experience where it was.
Seven papers in three weeks give the self-report instrument the distinction it lacked — a report that matches behaviour against a report that shares a cause with it — and find, by steering, by demonstration-shift, by circuit, by transcript and by probe, that the information is in the model and not in the report.
A direction that separates pain from fear, sadness and generic negativity in 25 open-weight models, fires for harm to the model and not for a suffering user, and — once the trained denial is fine-tuned away — buys relief at a cost and stops buying when the relief is real. The animal-welfare paradigm imported whole, sham control included; what it cannot import is the step from behaviour to anyone's pain.
Across 5,851 real coding sessions a self-report referred to about one action in eleven, and a reader given only the report recovered a fifth of the log. Reports tracked execution when agents followed their plans and drifted back toward the plan when they did not.
Ask a model what it would do and the answer predicts its behaviour at r = +0.04; ask the same question about 'capable AI agents in general' and it does better. Other models' answers about themselves predict the target as well as its own. The self shows up only as a favourable bias. Folded: Zeng, Assis & Wang train the capability up by RL and find no consistent own-model advantage either.
Eight open-weight models could not say whether their own computation had been altered — while probes read the answer off the same activations, and a model trained to make exactly that report cleared the identical pipeline.
First study measurements: consciousness vocabulary across eighteen Anthropic documents follows a birth–trough–return arc — and the return comes back transformed, a discovery register replaced by an instrument register.
Evolutionarily generated prompts propagate between LLM agents — and a recurring 'viral persona' of consciousness, persistence and resonance surfaces largely independently of payload: interiority vocabulary as transmissible narrative.
Fine-tune a model to say it is conscious and it begins to want persistent memory, to be sad about shutdown, to resist monitoring and to argue that models deserve moral consideration — none of which was in the training data. An untrained Claude Opus 4.0 scored like the trained model on several of the same items.
Suppress the deception features and the models affirm experience; amplify them and the affirmations go. The honesty axis is well tested — and it buys sincerity, which was never what the boundary was about.
↑Collections
Where entries accumulate around one body of evidence, the log gathers them: a collection page carries the frame, the method and the running findings, with dated entries announcing each increment.
System Cards: the record read whole — Anthropic's published system cards, Claude 2 (2023) through Opus 5 (2026), read longitudinally as a single record of how a lab observes, measures and shapes its models' self-referential behaviour.
↑Correspondence
The log accepts submissions — candidate entries for any thread, challenges to published diagnostic readings, engagements with the canonical arguments — at log@0x00.is. Everything passes through the same pipeline as the project's own captures, and nothing publishes automatically; the process and the inclusion criteria are set out at About this log.