Archive

All stories.
Every piece, in one place.

The full catalogue of scrollytelling pieces, visual essays, and data reports from Vizmaya Labs — geopolitics, technology, demographics, and the slow variables that explain the fast headlines.

35 stories in AI & Technology
#01Nov 9, 2026

Robotaxis: Tesla's 2M miles vs Waymo's 4M a week

Tesla has driven about 2 million unsupervised miles ever. Waymo drives about 4 million every week. The market still pays 160x profit for Tesla.

Read →
#02Nov 7, 2026

SwarmEvolve: how small AI swarms out-search big models

A small AI swarm tried 734 programs in an hour vs 158 for its best rival, and its 8B model scored 0.804 vs 0.774 for rivals on a 32B model.

Read →
#03Nov 6, 2026

TimeArch: why AI-built software needs a paper trail

A 2026 paper adds four steps and human sign-offs to AI software design. Its own test found different scores on the same input.

Read →
#04Nov 5, 2026

Plain AI science summaries: more understood, less smart

In a 2024 PNAS Nexus study, readers of plain GPT-4 science summaries explained the research better (0.92 vs 0.72) but rated the scientist as less smart.

Read →
#05Nov 5, 2026

AI-simplified science: readers learn more, trust is mixed

In a 2024 PNAS Nexus study, readers explained research better after plain GPT-4 summaries (0.92 vs 0.72), but trust gains were mixed.

Read →
#06Nov 5, 2026

Will androids outnumber humans? The 2050 and 2100 math

Humans lead robots 9.66 billion to 1.0 billion in 2050, even on a bullish forecast. A new paper shows what it would take for robots to catch up.

Read →
#07Nov 5, 2026

AIRoA-0: a robot taught by 16,151 hours of human hands

AIRoA-0 learned two-handed robot tasks from 16,151 hours of handheld human demos and scored 78% vs 27% for the best rival on ten tasks at an unseen site.

Read →
#08Nov 4, 2026

DepthGraft: reusing memory made an AI model score higher

On a 16B model, MMLU-Pro rose from 23.86 to 30.86 after its top layers reused earlier layers' notes, with a third less cache memory.

Read →
#09Nov 4, 2026

Why AI forgets | Vizmaya

Read →
#10Nov 4, 2026

GPU utilization: what the busy number really means

The GPU busy number only says a task was running, not how much of the chip it used. A new paper times the hidden steps. What holds up, and what does not.

Read →
#11Nov 3, 2026

SpecOPD: how AI guessers learn from their own mistakes

A 2026 paper trains a small AI guesser on its own guesses. Training ran 16.09x faster and guesses lasted longer, but gains over the best rival were small.

Read →
#12Nov 3, 2026

AI trip planning: 10% right alone, 100% valid with rules

A standard AI model wrote a valid trip plan for 10% of problems. Put the rules inside the decoder and every plan was valid, up to 1.43x faster.

Read →
#13Nov 3, 2026

SpecPilot: how an AI loop sped up LLM decoding 2.13x

An AI agent loop took one H100 from 184 to 416 tokens per second without changing answers, but one trick made batches 13% slower.

Read →
#14Nov 3, 2026

CUA-Master: a small AI, a terminal, and a 6.6 to 43.9 jump

Authors report a 27B AI scores 80.2 on a computer-use test. Its real edge: a terminal took one hard task from 6.6 to 43.9 in fewer steps.

Read →
#15Nov 3, 2026

PerceptRL: why an AI magnifying glass missed, and the fix

A preprint reports an AI that learns where to point its magnifying glass scores 71.6% on seven vision tests. Unreplicated, no code yet.

Read →
#16Nov 3, 2026

Tsinghua preprint: teaching vision AI where to look

A Tsinghua preprint reports that four attention heads decide where a small vision AI looks. Training them lifts accuracy 74.1% to 76.6%.

Read →
#17Nov 3, 2026

PDScope: why home AI reads fast but writes slow

A new paper says home AI stalls because reading and writing want opposite things. Its own test credits plain plumbing: up to 2.24x in decode.

Read →
#18Nov 3, 2026

Rimuru: the GPU memory wall that capped AI chats at 34

A new paper took down a fixed memory wall inside a GPU and ran 90 chats at once instead of 34. What it shows, what it costs, and what is unproven.

Read →
#19Nov 3, 2026

Diffusion AI agents waste compute after tool calls

A new study found AI agents waste compute after a tool call. A small controller cut 24.6% to 43.1% of the work with accuracy about level.

Read →
#20Nov 3, 2026

Apple SynthAPI: why AI agents learn from variety

At the same 32,768 training tasks, Apple's SynthAPI recipe with more apps, more worlds and more clutter beat a 'more tasks' recipe in 12 of 12 comparisons.

Read →
#21Nov 3, 2026

AI buildout news gives nearly half the world 2% of coverage

Across 42 editions of AI-buildout news, countries with 25.8% of world GDP and 46.2% of people got 2% of country tags. The US alone got 48.9%.

Read →
#22Nov 2, 2026

Evolving Agent Teams: how AI coders get unstuck

A lone AI coder stalled at 0.51× expert speed on a GPU task. A team that rebuilt itself reached 0.74×, then 1.21×, says a Berkeley preprint.

Read →
#23Nov 2, 2026

How AI search learns from its own wrong turns

By step 3 of a hard question, 70% of an AI search tool's notes hold its own wrong passage. Training on its own mistakes lifts full answers from 44.6 to 50.5.

Read →
#24Nov 2, 2026

KARMA: the ad AI that improved without retraining

Alibaba's KARMA lets an AI ad agent edit its own rulebook, one tested rule at a time. Average score on unseen cases rose from 30.19 to 39.15 of 100.

Read →
#25Nov 2, 2026

Phone AI memory: why flash is 1,333x too slow

Phone flash answers in 40 µs, memory in 30 ns. A new design, tested only in simulation, runs an AI agent team 21.4x faster. How it works and where it breaks.

Read →
#26Nov 2, 2026

Speculative decoding energy: when the speed trick backfires

A speed trick makes AI answer up to 4.3 times faster for one user, but the energy saving fades as users join and can turn into a loss.

Read →
#27Oct 26, 2026

Google's Finland data centres: forest cut, wolves hyped

Google's Finnish subsidiary felled about 525 hectares (1,300 acres) before its environmental assessments. The wolf 'invasion' rests on one villager's quote.

Read →
#28Oct 23, 2026

Would AI Have Told Alexander Graham Bell No?

Bell, Curie and Einstein each beat a written record an AI would have trusted. Why judgment, not prompting, is the skill that matters.

Read →
#29Oct 23, 2026

Entry-level jobs and AI: one lab's numbers vs the market

Anthropic says juniors fell from 27% to 9% of its tech hires. Fed and BLS data show no national break yet: ages 20 to 24 unemployment is 8.0% vs 9.2% a year ago.

Read →
#30Oct 16, 2026

Your AI agent bought the wrong thing. Guess who gets the box.

Amex will back your shopping agent’s mistakes, right after you’ve tried returning them to the merchant. Meanwhile, agent checkout is still measured in basis points.

Read →
#31Oct 9, 2026

The biggest seller of AI memory lost 8 points of share in a quarter. Its sales rose by half.

SK hynix gave up 8 points of AI memory share in one quarter while its total sales rose 51%. Here is how the market works, from the ground up.

Read →
#32Mar 1, 2026

The Quiet Inflation

How a strait reprices a GPU hour.

Read →
#33May 4, 2026

Stargate's Real Constraint

OpenAI is building 9 gigawatts of AI infrastructure across seven US sites. Three years in, only one is running. The map of where Stargate is being built is really a map of where you can plug in fastest.

Read →
#34Sep 21, 2026

The Domestic Stack

Memory, carbide, EDA and AI clusters. China is announcing all of them at once.

Read →
#35Sep 22, 2026

The Site-Selection Race

Materials, power, packaging, cooling and land. Nine bets on where AI infrastructure gets built.

Read →