The full timeline.

Everything in Bright, placed by date and by scale. There are two clocks, and they’re never mixed.

Published

When the paper, announcement or government report came out. This is what “recent” means.

Added to Bright

When Bright checked the source and put it in the collection. It tells you about the catalog, not the world.

A chart of all 152 records. Horizontal position is the date on the chosen clock; vertical position is the record’s physical scale, with programs and software releases in their own lane below. The six most recently published records are listed after the chart; every record is in the Index.

Every record jumps to 5 September–7 October 2026, the days Bright added them. Typical gap: about 13 months between publication and arriving in Bright. That is catalog history, not news.Each dot sits where its source was published. Yellow: the six most recent. 10 records whose source shows no date stay in their own lane instead of getting an invented one.

Published 7 October 2026 · Demonstrated

AI is helping small businesses get more of their own ideas off the ground

New Census figures show the scale of America's nonemployer economy. Individual business accounts offer a closer look at how AI can expand what an owner is able to do.

off the scale axis · U.S. Census Bureau, The Original Tamale Co.
Published 7 October 2026 · Emerging

A $1.8 billion biology coalition tackles AI’s missing experimental data

New experiments, measurement technology and shared standards could help researchers predict how cells respond to interventions and choose better laboratory experiments.

off the scale axis · Biohub, U.S. Department of Energy
Published 7 October 2026 · Emerging

What National Compute’s research preview offers

The restricted research preview describes access to eight-GPU nodes with $100 in starter credits. Its queue, billing rules and data-sharing terms show what that access…

off the scale axis · National Compute, AMP
Published 7 October 2026 · Emerging

ChatGPT is turning answers into tools you can use

OpenAI’s new interactive answers could make everyday planning and learning easier to explore. Availability depends on the model and app you use.

off the scale axis · OpenAI
Published 7 October 2026 · Deployed

Haiku 5.5 cuts API prices for small teams

Anthropic’s new small model lowers API prices sharply. Teams still need to watch prompt length, token counts and the cost of getting a usable answer.

off the scale axis · Anthropic
Published 6 October 2026 · Emerging

Google’s data center growth could put more nuclear power on the grid

A Constellation deal backs 890 megawatts of upgrades at existing plants, with the first additional output expected by 2028. The public benefit will depend on what gets…

off the scale axis · Google, Constellation

Bright’s own publishing history, in plain view.

Every one of the 152 records in this edition was added between 5 September 2026 and 7 October 2026. That burst is when Bright built its collection; it is not a burst of discoveries. Use the published clock for what is new in the world.

20added 5 September 2026
1added 6 September 2026
22added 7 September 2026
22added 9 September 2026
59added 19 September 2026
3added 25 September 2026
3added 27 September 2026
1added 28 September 2026
1added 30 September 2026
6added 2 October 2026
4added 4 October 2026
3added 5 October 2026
1added 6 October 2026
6added 7 October 2026

New to Bright, and your reading

Catalog additions and substantive changes, with source dates kept separate. Your reading position stays on this device.

BRIGHT EDITION /

A closer look at what’s possible.

One clinical study tested a second reading of scans people already receive. One research database added predicted viral protein pairs. One guide reads a release through the terms that govern its use. Each keeps the useful possibility beside the boundary of the evidence.

Three places chosen by hand. An edition is dated when it changes; there is no daily or weekly publishing promise.

  1. New to Bright

    CT scans may carry a second clue.

    EAGLE detected many esophageal cancers in external validation and one prospective hospital cohort, but performance was weaker for stage I and precancerous disease, follow-up was incomplete, and no outcome benefit has been shown.

    Added to Bright Peer-reviewed paper published

    Read the study and its limits
  2. Evidence updated

    AlphaFold’s record now includes a viral-pair research resource.

    The database added high-confidence predictions from 2,812 viral proteomes. They are research hypotheses that need experiments; the 2021 paper’s CASP14 result and Bright’s separate p53 model-v6 representation keep their own dates and claims.

    Added to Bright Database expansion announced

    Inspect the dated continuation
  3. New guide

    The weights are downloadable. The license still has a threshold.

    Liquid’s experimental 279.5-million-parameter drafter can accelerate decoding with a paired base model. Its own license requires a separate commercial agreement for organizations with at least $10 million in annual revenue.

    Added to Bright Release published

    Separate the weights, code, data and terms

Bright Weekly

Three developments, their evidence, and their limits. Read the web editions.

CONFIRMED EVENT DATES / 2026-09-18 — 2026-09-25

Events with a confirmed date in this window

Maryland makes data-center commitments public. ↗

Inspect current source reports ↗

Following and a reading bookmark stay on this device.

Catalog additions, corrections and evidence revisions

These are Bright catalog dates. Every record keeps its original source date and revision history.

2026-10-07T22:31:44.517Z · added

Haiku 5.5 cuts API prices for small teams ↗

Anthropic’s new small model lowers API prices sharply. Teams still need to watch prompt length, token counts and the cost of getting a usable answer.

Original source publication: 2026-10-07

2026-10-07T22:28:47.301Z · added

ChatGPT is turning answers into tools you can use ↗

OpenAI’s new interactive answers could make everyday planning and learning easier to explore. Availability depends on the model and app you use.

Original source publication: 2026-10-07

2026-10-07T21:18:10.783Z · added

What National Compute’s research preview offers ↗

National Compute documents a route to eight-GPU research nodes with $100 starter credits for eligible institutional users, subject to access waves and availability. The practical constraints include GPU-hour billing, ephemeral node-local storage and expanded Marshall data sharing for .edu organizations.

Original source publication: 2026-10-07

2026-10-06T00:28:36Z · revised

New York is asking hard questions about AI ↗

Added attributed evening update: Company witnesses set out access and oversight positions. Source verification and code filing clocks remain separate.

Original source publication: 2026-10-05

2026-10-06T00:28:36Z · revised

New York is asking hard questions about AI ↗

Added attributed evening update: Council seeks further written answers. Source verification and code filing clocks remain separate.

Original source publication: 2026-10-05

2026-10-06T00:28:36Z · revised

New York is asking hard questions about AI ↗

Added attributed evening update: A proposal to help New Yorkers build AI skills. Source verification and code filing clocks remain separate.

Original source publication: 2026-10-05

2026-10-05T19:27:11Z · revised

New York is asking hard questions about AI ↗

Added one AP-reported exchange on risk estimates and human control, with reporting and verification clocks kept separate.

Original source publication: 2026-10-05

2026-10-05T18:39:41Z · revised

New York is asking hard questions about AI ↗

Added attributed reporting of Coxon’s hearing remarks, BetaNYC’s preliminary written testimony and six official Council exhibits.

Original source publication: 2026-10-05

2026-09-28 · revised

NVIDIA adds an independent safety layer for AI agents. ↗

Added concise questions explaining OpenShell, Sentry and the limits of sandboxing from the existing cited sources, plus related Bright work. Original publication and source dates are unchanged.

Original source publication: 2026-09-28

2026-09-25 · added

CT scans may carry a second clue. ↗

Published “CT scans may carry a second clue.” with its source date, evidence state and material limitations explicit.

Original source publication: 2026-09-22

2026-09-25 · corrected

Seeing the shape of life. ↗

Added the September 24, 2026 AlphaFold Database viral-complex release as a dated continuation while preserving the original 2021 CASP14 result and the separate p53 model-v6 representation.

Original source publication: 2021-07-15

2026-09-19 · added

Teaching a robot from demonstrations ↗

Bright added this source-checked record on 2026-09-19. Its cited source publication date is 2024-06-13.

Original source publication: 2024-06-13

2026-09-19 · added

A local visual-inspection workbench ↗

Bright added this source-checked record on 2026-09-19. Its cited source publication date is 2022-02-16.

Original source publication: 2022-02-16

2026-09-12 · revised

A world-class forecast now fits on one machine. ↗

Re-reported around operational adoption at NOAA and ECMWF and the observing gap documented by WMO and SOFF, replacing a record based on a single vendor announcement.

Original source publication: 2025-12-17

2026-09-12 · revised

Lasers matched the herbicides. ↗

Re-reported from the peer-reviewed trial, the Western Growers audit, Rutgers extension guidance and the California hand-weeding regulation, replacing a record based on a product page.

Original source publication: 2025-06-24

2026-09-12 · revised

Twelve millimetres, or three centimetres. ↗

Re-reported around access and tumour size at detection, with the FDA-cleared indication, the Indian registry evidence and the follow-up drop-off made explicit. The supplied 'saved 24 lives' figure was traced to no source and is not carried.

Original source publication: 2025-12-11

2026-09-10 · revised

A brain interface remains a clinical question. ↗

Attached Synchron's dated NVIDIA Holoscan partnership and Chiral announcements as public evidence, and bounded the relationship to company-reported integration and stated plans.

Original source publication: 2025-03-19

2026-09-09 · added

A GPU goes into orbit. ↗

Added to Bright as a bounded source claim with visible evidence distinctions.

Original source publication: unknown

2026-09-08 · corrected

Seeing the shape of life. ↗

Clarified that Bright's p53 view uses the 2025 AlphaFold DB AF-P04637-F1 model v6 asset, while the discovery's scientific claim and publication date remain those of the 2021 AlphaFold paper. Added the model source and the limits of the predicted single-chain view.

Original source publication: 2021-07-15

2026-09-07 · added

More time to decide when to plant. ↗

A forecast becomes useful only when it reaches people in a form they can act on. This pilot joined forecasting with public delivery at unusually large scale.

Original source publication: 2025-12-02

2026-09-07 · added

A field image that asks for less tilling. ↗

It points toward agricultural decisions that aim to preserve soil and reduce runoff rather than optimize only a single output.

Original source publication: 2025-01-24

2026-09-07 · added

Finding bacteria among the crumbs. ↗

A quicker screen that can distinguish contamination from ordinary food material could help food-safety laboratories direct attention sooner.

Original source publication: 2025-11-21

2026-09-07 · added

A teacher and an AI tutor after school. ↗

The finding concerns a designed human-and-AI learning setting in a public-school context, rather than assuming chatbot access by itself creates learning.

Original source publication: 2025-05-19

2026-09-07 · added

When the answer arrives too soon. ↗

It makes the design of help visible: a system can improve assisted performance yet leave less learning when the assistance disappears.

Original source publication: 2025-06-25

2026-09-07 · added

Making room in a data center ↗

Recovering capacity can allow more tasks to run on an existing computational footprint.

Original source publication: 2025-05-14

2026-09-07 · added

Personalizing an ablation for persistent AF ↗

The study tests AI-guided targeting against a clinical comparator with a one-year patient outcome, rather than only a retrospective diagnostic metric.

Original source publication: 2025-02-14

2026-09-07 · added

A model whose full recipe is visible. ↗

People can inspect and adapt more of the system behind a model, which may make research, auditing, and specialized local development more practical.

Original source publication: 2025-03-13

2026-09-07 · added

Reasoning weights people can run themselves. ↗

Local or self-hosted use can offer people more deployment choice, including privacy-sensitive settings and adaptation to specialized work.

Original source publication: 2025-08-05

2026-09-07 · added

Choosing when a model thinks longer. ↗

Giving developers a stated control over reasoning time could make it easier to reserve slower processing for harder tasks and use faster responses for straightforward work.

Original source publication: 2025-04-29

2026-09-07 · added

A warehouse inventory check that rolls itself. ↗

If it holds up beyond the pilot, workers could spend less time on repetitive stock checking while operators gain a more current picture of what is in a warehouse.

Original source publication: 2025-03-03

2026-09-07 · added

Finding a missing line's ancient neighbors. ↗

It can give historians, students, and educators a transparent place to begin investigating fragments, while leaving interpretation with people who can weigh archaeological and historical context.

Original source publication: 2025-07-23

2026-09-07 · added

A public archive that starts with your question. ↗

A guided route through public records could make cultural and civic history more approachable for museum visitors, including people who might not begin with an archival search term.

Original source publication: 2025-11-18

2026-09-07 · added

An open reasoning model with its recipe beside it. ↗

The combination of weights, large stated data release, training environments, and recipes could let more researchers inspect, reproduce, and adapt a modern reasoning-model pipeline rather than only consume a hosted tool.

Original source publication: 2026-03-11

2026-09-06 · added

A voice in everyday life. ↗

Added the June 2026 home-use follow-up; the 2024 speech record and its measurements remain intact.

Original source publication: 2026-06-15