THE WORLD IS NOT STANDING STILL.RSS
BIG CHANGE.
BIG CHANGE, one tap away
BIG CHANGE, one tap away

Make room for the big picture.

Add BIG CHANGE to your home screen or desktop. It opens in its own window, with the latest stories when you’re online.

iPhone & iPad
Open bigchange.ai in Safari. Tap Share, then Add to Home Screen. Turn on Open as Web App if shown, then tap Add.
Android
Open bigchange.ai in Chrome. In the browser menu, choose Install app or Add to Home screen.
Desktop
In Chrome or Edge, use the install icon in the address bar or the browser’s app menu. In Safari on Mac, choose File → Add to Dock.
All briefings

Follow the change ·

The day,
in perspective.

What happened in AI and robotics — and the change behind it. A reading path through 50 stories from BIG CHANGE.

Ordered using RADAR attention, story opens and likes. Excerpts from our published reporting; open a story for the full context and sources. Dates in UTC.

guide

Measure six ways to cut a scraping agent’s HTML input without losing card fields

Compare six local controls on three permitted HTML listing pages, count serialized input with a pinned tokenizer, and check every extracted card against an independent reference.

The big change

What changed: The comparison starts with raw HTML, then filters the listing, extracts fixed fields, serializes structured JSON, batches pages under a token cap, skips unchanged pages and reuses versioned deterministic results. Every candidate includes its instruction and source metadata.

Why it matters: On the saved October 24 corpus, the comparable cold-run input fell from 27,863 tokens for raw pages to 2,083 for one bounded JSON batch, while all 24 card records matched an independent six-field reference. This is a local cl100k_base input count, not an API bill or model result.

What to watch: The warm run needed zero new extraction input for three unchanged pages. Reading cached records in a later agent conversation still consumes context. Current live pages may contain different rows, and changed markup requires a checked parser revision.

Read the full story
—
Back to the briefing
guide

Take a software milestone through GSD Core phases and acceptance

Use current GSD Core commands to carry an owned software milestone from requirements and phase plans through fresh-context execution, handoff, project tests and human acceptance.

The big change

What changed: The archived Get Shit Done repository points to GSD Core as its active successor. The current package has its own version line and requires Node.js 24 or newer and npm 10 or newer.

Why it matters: A milestone can carry numbered requirements through phase context, plans, execution summaries, verification and human acceptance across fresh Claude Code sessions, giving the owner concrete records to inspect.

What to watch: Generated PASS reports do not establish that your actual software works. Check project tests and user acceptance yourself, and keep pushing, PR creation and milestone completion behind separate owner decisions.

Read the full story
—
Back to the briefing
guide

Run a Claude Code evidence brief through three reviewed specialists

Build a source-grounded Claude Code brief with three fresh specialists, sealed handoffs and owner-reviewed rule trials. Includes the complete copyable workflow.

The big change

What changed: Project instructions, skills and custom specialists can be composed into one reviewed brief process. In this design, the specialists return bounded evidence, draft and evaluation artifacts tied to the same owner-approved rules; the assembled process has not been run in an account.

Why it matters: The owner can inspect where a claim came from and stop a brief when a source, rule or saved stage has changed. The same record makes a proposed rule improvement reviewable before it reaches production work.

What to watch: The next decision is whether the actual Claude Code session exposes the selected skill and agent-definition origins. Hidden or mismatched origins hold the run; only real owner-reviewed cases and a separate promotion decision can justify changing shared rules.

Read the full story
—
Back to the briefing
analysis

NVIDIA Flora separates robot-video annotation into three tasks

NVIDIA Flora labels robot video with time segments, action instructions and reward scores. Researchers need to validate each output against recordings.

The big change

What changed: Flora's separate outputs let researchers trace an annotation through timing, language and scoring. A team can examine each result before deciding whether a recording belongs in its dataset. That makes the source of an annotation error easier to locate.

Why it matters: Video labels become useful for learning only when the words, time span and assessment agree with the footage and the researcher's task. The model output is a starting point for that check. A convincing caption alone cannot establish that the selected interval or outcome score is right.

What to watch: The test is whether these labels remain accurate on the team's own camera views and tasks, including difficult boundaries and incomplete attempts. The noncommercial license confines this release to research and evaluation; commercial and production use are excluded.

Read the full story
—
Back to the briefing
analysis

LitPivot ties proposal revisions to paper rankings

LitPivot connects proposal edits with literature.

The big change

What changed: LitPivot links proposal edits to chosen papers. It flags affected passages and lets researchers inspect evidence before reranking the existing set.

Why it matters: A ranked paper set can omit an earlier contribution. The Paper Finder maintainers describe model-judged retrieval and a public code snapshot with fewer data sources than the live service. It helps choose what to read; its boundaries still matter when a contribution claims to fill a gap.

What to watch: A separate study of automated novelty judges found verdicts sensitive to small instruction changes. An originality claim therefore needs its comparison papers and judging method examined, alongside citations in the draft. A favorable score cannot supply a missing predecessor or show that the proposed difference would survive wider reading.

Read the full story
—
Back to the briefing
analysis

What a Trainium model-support PASS establishes

A Trainium PASS supports a first checkpoint trial; serving suitability needs separate tests.

The big change

What changed: Hugging Face’s October 19 TorchNeuron announcement described a native PyTorch backend for Trainium. The later lookup attaches a tested configuration and script to selected checkpoint results.

Why it matters: The documented run gives one controlled starting point. The application still needs its own checks of answer quality and serving behavior with the prompts and operating load it will use.

What to watch: Define what a useful answer looks like for the intended prompts, then compare outputs with that criterion. BIG CHANGE reviewed documentation; we did not run Trainium or inspect a live result row.

Read the full story
—
Back to the briefing
analysis

IORA separates geometry from holding strength

Robot grasp planners need to check both contact geometry and the forces a hand can actually produce.

The big change

What changed: IORA's refinement reduced simulated penetration but also reduced grasp success. It therefore tests refined contacts under simulated loads.

Why it matters: A hand can reach useful contact positions without being able to generate the forces needed there. For engineers, checking geometry alone leaves an essential part of grasp quality unanswered.

What to watch: Environmental support helps only while the supporting contact behaves as intended. Changes in contact or slipping can undermine a hold, so transferring a grasp requires attention to the surface as well as the hand.

Read the full story
—
Back to the briefing
analysis

Real PubMed IDs can still point to the wrong evidence

A PubMed ID identifies a citation, not evidence for a claim. A biomedical agent study shows why a resolving link still needs a check of the cited abstract and verdict.

The big change

What changed: A working PMID identifies a PubMed citation, not whether that paper bears on the sentence beside it. Treating a resolving link as evidence stops at the record; the claim and cited text still need comparison.

Why it matters: Biomedical literature teams can use an ID lookup to catch broken references, but claim verification needs the cited text. The reader must establish which finding the abstract reports and whether it supports, refutes or leaves the particular claim open. A plausible citation can otherwise lend a verdict authority it has not earned.

What to watch: Evaluation of literature agents turns on separate measurements of retrieval, evidence relevance and verdict accuracy. In this experiment, supplying the designated abstracts produced higher accuracy than the realistic search setups. That comparison identifies an evidence-access problem in this test; it does not establish a clinical result or the best way to search PubMed.

Read the full story
—
Back to the briefing