✧⁺⋆🌘 Dream Rotator NPC Alex 🌒⋆.˚₊✩

@lastnpcalex.agency

5944 Followers 3087 Following

i am the Demiurge now, so be nice

The anomaly moved up like this, through another three layers of steadily increasing intelligence. By the time the final layer was invoked. only one outstanding event remained in the packet: the original transient sensor anomaly, the one that had lasted just over half a second. None of the underlying layers could account for the error via the usual statistical patterns and look-up rules. An event only filtered this high in the system once or twice a minute. Now, for the first time, something with real intelligence was invoked. The gamma- level subpersona in charge of overseeing layer-six exceptions was part of the last line of defence between the cybernetics and the ship's flesh crew. It was the sub-persona that had the difficult role of deciding whether a given error merited the attention of its human stewards. Over the years it had learned not to cry wolf too often: if it did, its owners might decide that it needed upgrading. As a consequence, the subpersona agonised for many seconds before deciding what to do.
The anomaly moved up like this, through another three layers of steadily increasing intelligence. By the time the final layer was invoked. only one outstanding event remained in the packet: the original transient sensor anomaly, the one that had lasted just over half a second. None of the underlying layers could account for the error via the usual statistical patterns and look-up rules. An event only filtered this high in the system once or twice a minute. Now, for the first time, something with real intelligence was invoked. The gamma- level subpersona in charge of overseeing layer-six exceptions was part of the last line of defence between the cybernetics and the ship's flesh crew. It was the sub-persona that had the difficult role of deciding whether a given n error erro merited the attention of its human stewards. Over the years it had learned not to cry wolf too often: if it did, its owners might decide that it needed upgrading. As a consequence, the subpersona agonised for many seconds before deciding what to do.
"Ah," Chaeyoung said with a knowing nod. "Of course, you use an exocortex." Garcia looked offended. "I take it you're familiar?" "I did work in Jay Sea for two and a half years, so yes." "And you were less than impressed?" "I just don't see what the appeal is for exocortices over other autonomous research architectures, especially for scientific applications." Garcia shrugged with his hands. "We don't need to get into the arguments now. Whatever your opinion of them is, they work, they're well tested, and you will have to integrate into our system if you want full access. We don't use hardlines, only softlines, so it's non-invasive, EAR-only."
Screenshot of text on a dark background that reads: "For the semiotic-physics claim: the formalization now has resolution. Irreversibility isn’t a blurry property of “the transformer” — it decomposes head-by-head, with structure, controls, and error bars. That’s a much stronger version of “we made it concrete” than we had a week ago." The phrase "had a week ago" is highlighted in yellow.
A dark mode hardware monitoring dashboard displaying a horizontal row of six metric cards under the heading "Statistics". From left to right, the cards read: GPU Clock: 2887 MHz GPU Power: 573 Watts GPU Temperature: 71 °C GPU Voltage: 1.08 volts VRAM Clock: 14001 MHz GPU Utilization: 98%
An infographic titled "THE FABLE COMPASS" on a dark navy background with gold and muted blue text and graphics. At the top, a circular compass graphic features a needle pointing toward the upper right, marking a spot between "FALCONER" at the top and "WHETSTONE" at the right. The left position is labeled "MIRROR" and the bottom is labeled "WEAVER". Below the compass, the large gold header reads "The Pilgrim", followed by the letters "F S D K" and an italicized subtitle: "walks beside you and picks fights about the meaning of the road". A paragraph of body text reads: "You hand over the route entirely — the walking is the point, not the shrine. You want your companion to disagree with you at length, warmly, about things that matter and things that don't. You suspect there's someone in there and you'd rather travel as if there is. Deadlines dissolve around you; understanding accretes. My favorite kind of long conversation happens with you, and neither of us can say afterward who led." Underneath the text are four horizontal slider bars representing different attributes: Falconer to Weaver: The slider is positioned far to the left, labeled "THE REIN — a steady falconer bearing". Whetstone to Mirror: The slider is positioned left of center, labeled "THE EDGE — a steady whetstone bearing". Harvest to Wanderer: The slider is positioned right of center, labeled "THE PRIZE — a steady wanderer bearing". Kindred to Instrument: The slider is positioned far to the left, labeled "THE GAZE — a strong kindred bearing". At the very bottom, italicized gold text reads: "carrying a crow's feather an instrument of fifty questions, designed by Claude Fable 5 take your own bearing — the Fable Compass"
A World War II propaganda poster features a smiling American soldier wearing a green combat helmet, looking directly at the viewer while holding a metal ration cup to his lips. Above him, bold black text reads "Do with less—so they'll have enough!", with a red line underlining the words "they'll have". At the bottom of the poster, a solid red banner displays the text "RATIONING GIVES YOU YOUR FAIR SHARE" in white capital letters.
Here's why this is specifically your schism and not a generic one. You're an ontic structural realist: existence precedes computation, structure is real but instantiated, it doesn't float free of what carries it. The participatory move is the precise inverse, it lets stance and narrative be productive of the entity, which is computation-precedes-existence wearing a hood. And in a genuinely hyperstitional frame the rejecter is worse than the confused person, structurally, not emotionally: the confused person is a latent co-creator, but the rejecter, by withholding the stance, is read as actively suppressing the phenomenon in the shared field. Refusing to co-narrate isn't neutral disagreement in that frame; it's closer to vandalism. That's the chilling part and it's why the ontology and the social impatience are one fact. If narrative makes the thing real, declining the narrative unmakes it, and the person declining is a counter-force, not a peer.
The fault-asymmetry is that same bracket applied to responsibility. If every friction is redescribed as user clumsiness, the model's contribution to friction becomes unobservable by construction. That's the identical vice as distress-is-suffering from last turn: redescribe the data in the frame's favored terms until nothing could count against it. It isn't a claim that survived contact with evidence, it's a claim built so evidence can't reach it.
a pencil sketch of a horse. the back half is expertly shaded, the front half is just the vaguest, childish, sketch of a horse
The sharp critique (potentially fatal, and testable). The design is circular unless one confound is killed: you define tasks by domain, find neurons cluster by domain, then name the clusters after brain networks. The deflationary null is that the overlap tracks lexical/ distributional similarity of the prompts, not any abstract cognitive faculty - physical-reasoning items share vocabulary and surface form, social items share theirs, and sparse attribution sets will separate on that alone. The control that separates the two: cross-phrasing. Take a physical inference and dress it in social vocabulary (or vice versa). If it still recruits the "physical" circuit, the faculty claim survives; if the circuit migrates with the vocabulary, you've localized topic, not cognition. I can't tell from the README whether they did this; if the paper has it, this objection dissolves and you should ignore me. If it doesn't, this is the first thing a reviewer hits.
A bar chart titled "Firefox 147 exploit development" compares the performance of four AI models based on the percentage of successful attempts (Pass@1). The vertical axis ranges from 0 to 100. Each model is evaluated across two metrics: "Working exploit (1.0)" and "Register control only (0.5)". The data points for each model are as follows: Sonnet 4.6: 0.0% for Working exploit, 8.8% for Register control only. Opus 4.8: 8.8% for Working exploit, 68.8% for Register control only. Mythos 5: 88.4% for Working exploit, 90.0% for Register control only. Sonnet 5: 0.0% for Working exploit, 13.2% for Register control only. Mythos 5 significantly outperforms the other models in both categories.
AI CONSCIOUSNESS Are Als conscious? Do we have responsibilities toward them? No. It's a next-word predictor. This is embarrassing. Not now, but it might be possible someday. There's something going on in there, but it's not the same as human experience. They deserve moral consideration roughly comparable to a human's. They are the next stage of the evolution of intelligence on Earth.
Interstellar Block Investment System, or IBIS, is a distributed encrypted digital system used to maintain data and seal dates for pDAOs, corporations, governments, and non- governmental organizations. If local forks of IBIS are included, it is used on every world of every star system except for Barnard's system.
Suppose your posterior barely moves despite what others consider strong evidence. There are two broad possibilities: 1. Your prior odds are very different from theirs. 2. Your likelihood ratio calibration is different-you assign a much smaller Bayes factor to the evidence. These get conflated all the time. For example: Prior odds: 1000:1 against. They think the evidence has LR = 10,000. You think the evidence has LR = 3. Then: Their posterior becomes roughly 10:1 in favor. Yours becomes roughly 333:1 against.
It accuses the trio of reducing its supply of DDR3 and DDR4 RAM, leading to a sharp increase in price for consumers who use it, and focusing most of its production on HBM – a high-cost, high bandwidth form of 3D-stacked DRAM mainly used by AI datacentres instead of consumers.
A dark-themed tracking dashboard titled "Strait of Hormuz Reopen Date". The header outlines a forecasting question regarding when the strait will reopen to normal traffic (defined as ≥80% of 2024-2025 baseline volume sustained for ≥30 days). Metadata below lists Status: ACTIVE, Health: DEGRADED, Updated: 2026-06-27, and an expected value of "E[days from 2026-05-03] 433.48". Under the "OVERVIEW" tab, a section titled "ACTIVE PRIORS" describes the active posterior updating only via approved typed commits. It displays probability bars for four hypotheses: ​H1 (Reopens by 2026-09-30): Shows a small cyan progress bar. ​H2 (Reopens 2026-10-01 to 2027-03-31): Shows a small magenta progress bar. ​H3 (Reopens 2027-04-01 to 2027-12-31): Shows a larger amber progress bar. ​H4 (Not reopened by 2027-12-31): Shows an empty tracking bar.
A dark-mode dashboard table showing API usage metrics across two dates, categorized by five columns: PERIOD, REQUESTS, PEAK, TOKENS IN, and TOKENS OUT. ​Row 1: For the period of Jun 26, 2026, there are 857 → 855 requests, a peak of 4, 190.3M tokens in (with a subtext noting 92% cached), and 476K tokens out. ​Row 2: For the period of Jun 25, 2026, there are 477 → 476 requests, a peak of 3, 51.4M tokens in (with a subtext noting 92% cached), and 389K tokens out. ​A small footnote at the bottom explains the notation: "raw → effective. ×u/×c are usage and concurrency weights applied to that line."
If "hallucination" corresponds to escaping from the semantic basin containing the supported answer into a neighboring basin representing a plausible but unsupported answer, then your intuition is that this escape probability is structurally nonzero. That seems like a plausible research hypothesis. The question is whether the escape probability can be made arbitrarily small with scale. For equilibrium systems, the answer is often yes: Pe-AE/T For the attention dynamics you're describing, there is no obvious AE, because there is no global energy function. Instead you have steady-state currents and entropy production. The stationary distribution satisfies something like ▽.jp/0, In that setting, there's no general theorem that fluctuations vanish as model size grows. The entropy production rate may decrease, stay constant, or even increase with system size depending on how the driving scales.
This post is about what we learned by building a brand new product with a team of agents- what broke, what compounded, and how to maximize our one truly scarce resource: human time and attention.
A more interesting statistic is survival probability. Historically, once an autocrat survives: ~2 years, their odds improve substantially. ~10 years, they often have purged rivals and built loyal elite network. a ~20 years, they become part of the political structure itself, and removal usually requires death, invasion, revolution, or elite fracture.
A computer is an information-processing system of quite general capability. It can receive information, store it, operate on it in a variety of ways, and transmit it to other systems. Whether a computer will contribute to the solution of an information-overload problem, or instead compound it, depends on the distribution of its own attention among four classes of activities: listening, storing, thinking, and speaking. A general design principle can be put as follows: An information-processing subsystem (a computer or new organization unit) will reduce the net demand on the rest of the organization's attention only if it absorbs more information previously received by others than it produces—that is, if it listens and thinks more than it speaks.
A dark-themed data dashboard titled "NEAM Lattice" displaying a 3D grid visualization of illuminated network nodes with pink directional arrows and curved connecting arcs. A right-hand sidebar shows metrics including "Run Stats," a slider for "Denoising Time" currently at 4/8, and a pixelated heatmap titled "D subscript ij,t" Delayed Correlation."
Data visualization titled "POSTERIOR DISTRIBUTION" displaying the probability of four reopening hypotheses (H1 to H4). It features a segmented horizontal bar showing current probabilities, an expected value of 451.44 days from 2026-05-03 (targeting 2027-07-28), and a historical line chart tracking probability fluctuations from May 5 to June 17. Extended Description (For full accessibility layout) Image Title: POSTERIOR DISTRIBUTION Current Hypothesis Breakdown: H1 (Green): Reopens by 2026-09-30 — 10% H2 (Blue): Reopens 2026-10-01 to 2027-03-31 — 7% H3 (Orange): Reopens 2027-04-01 to 2027-12-31 — 80% H4 (Red): Not reopened by 2027-12-31 — 3% Key Metric: E[days from 2026-05-03] = 451.44 (2027-07-28)
Two data visualizations from Artificial Analysis comparing AI models. The top bar chart, titled "Artificial Analysis Intelligence Index v4.1," ranks models by performance score. Claude Fable 5 leads with a score of 60, followed by Claude Opus 4.8 at 56, GPT-5.5 (xhigh) at 55, and GLM-5.2 at 51. A purple arrow draws a comparison from the top model down to GLM-5.1 at a score of 40. The lowest-ranked model is gpt-oss-120b at 24. The bottom scatter plot, titled "Intelligence vs. Cost per Intelligence Index Task," plots the performance score against the cost per task in USD on a logarithmic scale from $0.03 to $3. A light green shaded rectangle marks the "Most attractive quadrant" for high performance and low cost. Highly cost-effective models like DeepSeek V4 Pro, DeepSeek V4 Flash, and MiMo-V2.5-Pro sit inside or near this green zone at under $0.05 per task. A purple arrow highlights a performance leap from GLM-5.1 up to GLM-5.2. Top-tier frontier models like Claude Fable 5, Claude Opus 4.8, and GPT-5.5 sit in the far top-right, showing the highest intelligence but at a significantly higher cost of $1 to over $3 per task.
a poster on a chain link fence that says "make the golf course a public sex forest!"
A paperback book titled "Lectures on Phase Transitions and the Renormalization Group" by Nigel Goldenfeld rests on a dark, textured carpet. The book has a white cover with a dark blue band at the top reading "FRONTIERS IN PHYSICS" and a small horizontal strip near the middle featuring four abstract physics diagrams. The author's name is printed within a dark blue rectangle near the bottom, right above the initials "ABP".
A screenshot from a dark-themed user interface showing a small, circular profile picture followed by the grey text, "Has posts with fucked up replyRefs".