06 // The writing

Look past the claim.

Ten dated essays from Aska Digital on what AI is actually changing, and on what has to be built around it for the change to last.

Most writing about AI argues about how clever the machines are. These essays ask a plainer question: what has to be true around a machine for its work to count? A proof counts only when someone can check it. A tool counts only when someone owns the outcome it is meant to improve. A public good lasts only when someone can pay for it without owning it.

Analysis · not peer reviewed 10 essays · 13 Aug – 7 Oct 2026 data/writing.json

The essays One claim, drawn The theses map Claims with a date How to read them

FeaturedDay-one review · 7 Oct 2026

Math, faster

On 6 October 2026 OpenAI published 722 mathematics manuscripts written by one of its AI models. Aska read them on day one, ranked the headline claims and checked what stands behind each. The finding: checking, not discovering, is now the slow part.

722
manuscripts in one commit, grouped into 372 families of results
README @ adc7f12
0
independently confirmed when the essay appeared
the essay, 7 Oct
235/372
families with a Lean scope document on day one: a plan for machine-checking, not a finished check
CONTENTS.md @ adc7f12
3
manuscripts withdrawn the next day, after a sign error
history.md @ fd4aeeb

UpdateThe essay is a day-one review. On 7 October OpenAI withdrew three manuscripts, revised fourteen and added formal proofs; when we checked on 14 October the essay had not been updated. Our page After the 722 follows what people and their AI agents did next.

Read the essay After the 722 Standalone edition

01//The essays

Aska Digital b.log · newest first

Ten essays, one thread.

Each is dated, names its sources and says how it could be wrong. Filter by the Foundation0 theme an essay serves, or search the titles, summaries and notes.

Showing all 10 essays.

7 Oct 2026AI / Mathematics
Day-one reviewFeatured
Math, faster
OpenAI published 722 machine-written mathematics manuscripts in 372 result families on 2026-10-06; Aska Digital read the corpus on day one, ranked the headline claims and checked the evidence behind each. Nothing was independently confirmed, a minority had machine-checked proofs, and the real shift is that checking, not discovering, is now the bottleneck.
Why it matters hereThis is the kind of advance the Ledger exists to record: a machine wrote the proofs, and only checking by people and proof-checking software can say which are true. It also shaped how this site labels a claim, from 'claimed' up to 'formally verified', and why withdrawals stay on the record.
- Frontier
Written on day one; predates OpenAI's 7 Oct withdrawals
3 Oct 2026Crypto Markets
Field guide · dated
Crypto Industry Lowdown
A guide for readers returning to crypto: prices fell, but stablecoins, ETFs, custody, perpetuals and agent-payment rails became infrastructure. The security threat moved from contract code to keys and suppliers.
Why it matters hereA side thread for us: AI agents are starting to hold wallets and pay each other, and AI now helps audit code, though the essay is clear it cannot fix how keys are kept.
- Markets
Not investment advice
3 Oct 2026AI / Machine Learning
Practical guide
Post-Training Your Own Models
A practical guide to teaching an open-weight model a task, a workflow or your writing voice: methods, memory math, prices, tools, evaluation and three playbooks. Its first rule is to try a better prompt, then retrieval, and only then training.
Why it matters hereIt shows that one person with one graphics card can now teach an open AI model a new task, which is what our Instruments work is for: tools that put lab-grade methods in an independent researcher's hands.
- Instruments
Prices are web-research snapshots, not live
18 Sep 2026Agentic Systems
Plain-words taxonomy
The Naming Decides Who Will Run the Machines
Aska Digital's IR-Taxonomy gives non-engineers five plain names for agentic systems, so a buyer knows what kind of machine they own before a demo decides for them. Each name comes with one question the owner is entitled to have answered in writing.
Why it matters hereIf the people who deploy AI systems can name what kind of machine they own, they stay answerable for what it does. Shared plain words are part of working with machines rather than under them.
- Symbiosis
- Instruments
Says itself it is not a standard
10 Sep 2026Featured essay
Traced argument
Subtractive Fragility
Long calm periods make systems fragile less by piling up risk than by quietly removing capacities that only pay off in a crisis. Those capacities get rebuilt only when someone can bill for the rebuild; the essay makes seven dated, falsifiable predictions (2027–2035).
Why it matters hereIts seventh prediction names a precise risk for people working with AI: if AI removes junior jobs, nobody is trained to become the next experts. Its rule for funding what must last is a model for a public-benefit foundation.
- Symbiosis
- Public good
7 dated predictions, none scored yet
8 Sep 2026Prediction Markets
Market snapshot · dated
Prediction Market Outlook
A snapshot of what public prediction markets priced on 2026-09-08, read as scenario weights rather than forecasts. Includes a small regression on what makes a market's probability lopsided.
Why it matters hereMostly a dated snapshot. Worth reading for how plainly it states its own limits: 23 markets, not a random sample, and a fit that explains how lopsided prices are, not how accurate.
- Markets
Prices as of 8 Sep 2026; not betting advice
2 Sep 2026Agent Operations
Working example
The Agents Are Replaceable. The Operating System Is Not.
A one-person consultancy runs like a ten-person firm on a team of AI agents. The agents are interchangeable; what makes the system work is the discipline around them: handoffs, blocking gates, checks on the real output, and memory written down outside any one session.
Why it matters hereAska's own working example of a person and a team of AI agents cooperating under written rules, and honest about its limit: nothing is proven until a stranger takes the kit and runs a real team on it.
- Symbiosis
- Instruments
Says itself its value is asserted, not yet shown
27 Aug 2026Institutional Adoption
Argument
The Institution Is the Bottleneck
Healthcare, education and government keep failing to adopt working technologies in the same way. A technology takes on the incentives of the institution it lands in, so it sticks only where an outcome has an owner, a measure and a priced consequence.
Why it matters hereIt explains why free tools such as MedCode and Nutrak need someone who owns the outcome, a way to measure it and a way to reach people, not just good software.
- Public good
Sources named by publisher and year
19 Aug 2026AI & Public Good
Argument
The Money Skips the Public Good
Canada's 'AI for All' strategy names the public, but every funding channel it built pays only for AI that can bill for its own value. Public-good AI is left with the slow, constrained charitable route, so the legal form chosen on day one decides whether the mission survives.
Why it matters hereIt is about how a public-benefit AI organisation, a foundation, has to be set up and funded so that its mission survives the money, and so that its free tools stay free.
- Public good
Canadian funding as of Aug 2026
13 Aug 2026Decentralization
Argument
No Coup, Just Comfort
Decentralized systems are rarely defeated by enemies; they are given away one convenient default at a time. The industry cannot see it because its metrics count what adoption adds and nothing that decentralization loses.
Why it matters hereThe same drift threatens people working with AI: help becomes habit, and habit becomes dependence. Its rule, measure what is being taken away and not only what is added, is one we work by.
- Symbiosis
About decentralised networks

Source: data/writing.json (titles, dates, summaries, themes, from digital.askaconsult.com/blog); the type, status and “why it matters here” lines are ours, in writing/annotations.json.

02//One claim, drawn

From Math, faster · family 087

A shape times its shadow.

Math, faster rated one result the best-evidenced in OpenAI’s corpus: a proof of Mahler’s conjecture in every dimension. In two dimensions you can watch the whole idea.

Take a shape K centred on a point. Its polar, written K°, is a kind of shadow: the points y with x·y ≤ 1 for every point x of the shape (x·y, the dot product, multiplies matching coordinates and adds them). A big shape casts a small polar; a square casts a diamond, and a circle casts a circle. In 1939 Kurt Mahler asked how small the area of a shape times the area of its polar can be. The round shape gives the largest product. Mahler guessed that boxy shapes give the smallest, and proved that in the plane; in higher dimensions it has stayed largely open.

Shape and shadow

How to read: gold is the shape K, the set of points with |x|^p + |y|^p ≤ 1; cyan is its polar. Slide p from a diamond (1) through a circle (2) to a square (∞) and watch the product of the two areas: it never drops below 8. Then try a shape that is not symmetric.

Claimed · Lean scope document + 10 check programs · not refereed

p 1

Areas computed in your browser: a p-shape has area 4Γ(1+1/p)²/Γ(1+2/p), and its polar is the q-shape with 1/p + 1/q = 1. The bounds, 4ⁿ/n! for symmetric shapes (8 in the plane) and (n+1)ⁿ⁺¹/(n!)² in general (6.75 in the plane), are the statements in OpenAI’s Lean scope document for family 087. Math, faster called the evidence the strongest in the corpus, and the result “provisional, not cleared”.

Show the numbers

Area of the shape, of its polar, and their product
Shape	Area of K	Area of K°	Product

03//The theses map

Our reading, not the essays’ own labels

One argument, from ten angles.

Read together, the essays make one claim: progress now outruns the structures that check it, fund it and remember it. The map links each essay to the themes it serves, and each theme to the pages on this site where it becomes work.

Essays → themes → pages on this site

How to read: each line says “this essay serves this theme” (left) or “this theme becomes work on this page” (right). Point at or select any node to trace its links. With the keyboard, Tab into the map, move with the arrow keys, press Enter to pin a node and Escape to clear.on a narrow screen the map is a list. Each theme shows the essays that serve it, then the pages on this site where it becomes work.

Essays 10

Themes 5

On this site 6

Source: themes from data/writing.json; the links to pages, and the reasons, are ours, in writing/annotations.json.

Show as a list

Every link in the map
Essay	Themes it serves	Pages on this site, and why
Math, faster	Frontier	After the 722 (via Frontier) The Ledger (via Frontier) Research (via Frontier)
Post-Training Your Own Models	Instruments	Instruments (via Instruments)
The Agents Are Replaceable. The Operating System Is Not.	Symbiosis, Instruments	Charter (via Symbiosis) Instruments (via Instruments)
The Naming Decides Who Will Run the Machines	Symbiosis, Instruments	Charter (via Symbiosis) Instruments (via Instruments)
No Coup, Just Comfort	Symbiosis	Charter (via Symbiosis)
Subtractive Fragility	Symbiosis, Public good	Charter (via Symbiosis) Charter (via Public good)
The Money Skips the Public Good	Public good	Commons (via Public good) Charter (via Public good)
The Institution Is the Bottleneck	Public good	Commons (via Public good)
Crypto Industry Lowdown	Markets	None: a dated snapshot
Prediction Market Outlook	Markets	None: a dated snapshot

04//Claims with a date

Status as of 14 Oct 2026

Seven predictions, each with a date and a way to fail.

In Asimov’s Foundation, Hari Seldon’s recorded image appears in a Time Vault at moments he fixed in advance, each time to speak to the crisis he had predicted for that moment. Subtractive Fragility does the same in plain sight: seven claims, each with a scoring date and a stated condition under which it fails. If they fail, the essay says, its thesis is wrong.

Seven predictions, waiting for their dates

How to read: each row runs from the day the essay was published, 10 September 2026, to the date its prediction is scored (the ring). The solid part is time already passed; the vertical line is today. Select a row to read the claim, how it fails and where it started.

No date set · from Math, faster

“What would confirm the thesis”: the essay names three events and gives them no date. We list them as open.

OpenA second lab reproduces the pipeline.
OpenLean coverage grows on a visible schedule.
OpenOne falsifiable core gets killed or confirmed in public.

Source: the tracker in Subtractive Fragility (10 Sep 2026), checked 14 Oct 2026; claims restated in plain words in writing/annotations.json. Statuses are the essay’s own. Next to be scored: no. 1, Dec 2027, in 1 year 3 months. The two that carry the thesis, 5 and 7, are scored in 2032 and 2035.

Show the numbers

The seven predictions, as of 14 Oct 2026
No.	Scored	Claim	Fails if	Started from	Confidence	Status
1	Dec 2027	US lead times for large power transformers are still above 80 weeks, despite announced factory expansions.	Lead times drop below 80 weeks.	128 weeks on average in 2026 (IndustrialSage)	Moderate	Pending
2	Dec 2028	US output of heavy-duty gas turbines has grown by less than 40% compared with 2025.	Output rises more than 40%.	GE Vernova backlog of 116 GW in 2025 (company filings)	High	Pending
3	Through 2030	US electricity demand grows between 1.5% and 3.5% a year, every year through 2030.	Growth exceeds 4% in any year, or falls below 1% for two years running.	2.1% a year, five-year average to 2025 (EIA)	High	Pending
4	Nov 2030	The price of electricity becomes a first-rank issue in a US federal election, with datacenter siting or who pays for the grid as explicit policy planks.	It stays a matter for state regulators.	57 to 70% oppose a local datacenter (Gallup, Brookings, 2026)	High	Pending
5	Dec 2032	Even if AI spending falls by more than half from its 2026 peak, US grid capacity ends up higher than in 2026 and gets used, not stranded. (thesis)	An AI spending collapse is followed by measurable stranded capacity.	2,600 GW waiting in interconnection queues (FERC, 2026)	Moderate	Pending
6	Dec 2032	No US scheme that pays to keep new antibiotics available, comparable to the PASTEUR Act, has passed and paid out at scale.	One passes and pays out.	PASTEUR Act reintroduced Feb 2026, still unpassed (Congress.gov)	High	Pending
7	Dec 2035	At least one high-skill profession shows a measurable shortage of senior experts that hiring cannot fix, because the people who would now be senior were never trained as juniors. (thesis)	Senior pipelines refill normally, or senior roles are restructured rather than lost.	Junior employment down about 8% since early 2023 at firms adopting generative AI (Harvard working paper)	Moderate	Pending

05//How to read them

Arguments, not peer-reviewed research.

The house method

Each essay opens with its method, gives a section that states the strongest case against itself, closes with a compressed version of the argument, and lists its sources.

Who wrote them

The bylines are ASKA Research or The ASKA Team, at ASKA Business Solutions Inc. No post names an individual author, and none says, post by post, what AI did in the research or the writing.

Dated pieces

Market pieces carry their dates and say they are not investment or betting advice. Prices and counts in them are as reported on those dates, and many are labelled by the essays themselves as index-level figures that were not checked live.

Corrections

Math, faster predates OpenAI’s 7 October withdrawals; we link the update above. The statuses in Subtractive Fragility are set by its authors; we show them as they stood on 14 October 2026. Found an error on this page? Report it.