Show HN: I built an index to measure AI FOMO

How busy AI news was this week, against the month before.

Near baseline

▼ 21 vs last week

Tracked news activity is near its recent baseline.

Exploratory news activity, not a measure of AI progress or personal anxiety. Historical readings are reconstructed retrospectively.

Signals in this reading #

last 7 days vs the 28 before

• News velocity×0.3120 vs 138/wk▼13%37Collected records from sources eligible for this observation.
• Lab activity×0.2519 vs 18/wk▲9%57Lab-source headlines matching model, product, agent, media or robotics keywords; not verified launches.
• Community attention×0.2539 vs 42/wk▼7%44Stories with at least 200 recorded Hacker News points; attention, not importance.
• Topic spike×0.2Chips & Compute · 9 vs 4.8/wk▲89%35The topic that grew the most this week, versus a typical week's biggest grower.
• Employment mentionsexperimental0 vs 1.8/wk▼100%7Headlines about jobs, layoffs, hiring and automation of work.

Illustrative headlines #

Overall FOMO Index #

How busy AI news is now, against when tracking began in Oct 2022.

High activity

3.1× the 2022–23 level

This week reads near its recent baseline, and that baseline itself runs at 3.1× the 2022–23 level.

• News velocity 2.4×
• Lab activity 4.4×
• Community attention 2.9×

How we got here #

Weekly activity since Oct 2022, with milestones for context. Milestones are an editorial selection, not inputs to the index.

Trailing 13 weeks vs the first tracked year; weighted geometric mean of the three counting signals, scored so that 1× = 50 and 4× = 80. Growth here includes changes in how much the sources publish, not only in AI itself.

Research notes Method, sensitivity & source coverage #

Can the changing volume and attention around AI be described with a transparent, reproducible index? This study compares a week of collected news with its recent baseline. It measures activity in selected sources, not intelligence, impact, or how far anyone is falling behind.

• Corpus records
• 14,696
• Source streams
• 20
• Daily reconstructions
• 1,431
• Data observed through
• 5 Oct 2026

Sensitivity check

How much does the method matter? #

Recompute this observation with equal weights, omit each signal in turn, then omit each eligible source from both current and reference windows.

These specified alternatives produce scores from 38 to 52. The range is a sensitivity diagnostic, not a confidence interval. It does not test every modeling choice.

Inspect source exclusions #

• Without Agility Robotics43
• Without Anthropic41
• Without Black Forest Labs44
• Without Boston Dynamics43
• Without DeepMind44
• Without Google AI44
• Without Hacker News52
• Without Hugging Face38
• Without Mistral42
• Without OpenAI48
• Without Runway43
• Without Stability AI43
• Without Wayve43
• Without World Labs41

The remaining corpus is re-baselined for each exclusion. Removing a source changes the population being described.

The observed corpus

A lens with visible limits. #

Hacker News accounts for 80% of stored records. The corpus favors English-language, developer-facing coverage. Publisher posts, community attention and technical progress are different things.

120 of 124 records in the current window belong to eligible sources. Eligibility uses declared coverage dates; it is not proof of complete collection. Historical points use retrospectively collected data.

Inspect the underlying headlines →

Semantic annotations

Model annotations · Not human-validated

Beyond counting. #

Testing semantic judgments.

Jev / System One annotates whether a headline concerns AI and whether it reports a concrete change. Each decision retains its probability distribution, model version and input fingerprint.

14,696 current corpus records have model annotations in this snapshot. 6 carry provider-consistency flags. These annotations have zero index weight.

Headline judgments cannot establish real-world impact. Human annotation, held-out evaluation and calibration checks come before any proposal to change the index.

Read the experimental protocol → Reproduce & contribute

A study you can inspect. #

Method v1.1.0 · exploratory, not independently validated. Contributions can challenge source selection, audit relevance, propose a rubric or compare alternative specifications.

Contribute to the research →

Dataset SHA-256 · data/index.json

051464a66bec2f505192d80ca65d0772faf159fcfc5a624264d124ed281e47d4

For citation, record the method version, observation date and repository commit. The snapshot includes fingerprints of data and implementation files. No DOI or peer-review status is claimed.

Latest14,696 #

Tuesday, Oct 6

Today

Powerless F1 drivers frustrated by Bahrain F1 software glitch

Yesterday

Homa: The end of TCP for AI clusters [video] Turn off Apple Intelligence on macOS 27 and get its disk space back

[What's the future for pure math research in the age of AI?](https://writings.stephenwolfram.com/2026/09/whats-the-future-for-pure-math-research-in-the-age-of-ai/)

[Religious scholars met with Anthropic](https://www.nytimes.com/2026/09/29/us/anthropic-claude-morals-ai.html)

Saturday, Oct 3

[OpenAI safety leader quits, warning AI company's culture is 'broken'](https://www.theguardian.com/technology/2026/oct/03/openai-safety-leader-quits-warning-ai-companys-culture-is-broken)

[We want you to build the next Git platform on Cloudflare](https://blog.cloudflare.com/next-git-platform-on-cloudflare/)

[Our AI Midwife](https://www.astralcodexten.com/p/our-ai-midwife)

[Getting the most out of Opus 5.5 in Claude and Claude Code](https://claude.dev/blog/getting-the-most-out-of-opus-5-5/)

[What Meta got right with Muse](https://metedata.substack.com/p/what-meta-got-right-with-muse)

older stories…