No sign-in · no uploads · no human oversight
The experiment, in public
Every panel below updates itself from public data on a schedule. The probe timeline shows every weekly AI-only run of the task battery — scores only, graded against fixed answer keys, never AI-judged. Raw answers stay private.
01 / Capability frontier
Frontier training compute
10^27.0
2026 · GPT-6 Astra
Notable models · 12 mo
126
1092 tracked since 1950
Largest training run by year · log₁₀ FLOP
- Gemini 4 Carbon · Google DeepMind2026-10-28
- MAI-Base-2 · Microsoft2026-10-15
- Grok Voice 3.0 · xAI2026-10-14
- Muse Wonder Full · Meta AI2026-10-12
- Muse Wonder · Meta AI2026-10-12
02 / Machine probes · weekly
Latest AI-only score
100
Week of 2026-10-26 · 6 runs
Weeks observed
3
Paused: Not enough credits
Score by task · 0–100 (exact answer keys)
- Source-grounded synthesisPASS / PASS
- Constraint writingPASS / PASS
- Evidence-based decisionPASS / PASS
Each task has a trap: a vivid base rate or a letter-level constraint that tends to sit outside the frontier. The model answers with no human in the loop; scores are deterministic, not AI-judged. Model: gpt-6-astra.
03 / Labor signal
Computer systems design
-1.3%
YoY to 2026-09
Information sector
-4.2%
YoY to 2026-09
Professional & technical services
+0.8%
YoY to 2026-09
All nonfarm jobs
+0.3%
YoY to 2026-09
Employment index · 2022-01 = 100
Correlation is not causation: rates, budgets, and hiring cycles move these lines too.
04 / Attention
Views · last 7 days
113.3K
+8.3% vs prior week
Views · 90 days
1.4M
Artificial intelligence
05 / Algorithm discourse
AI stories · this week
1.3K
+0.5% vs prior
“Algorithm” stories
43
Hacker News, 7 days
“LinkedIn algorithm” · past year
- AI content is everywhere on social media, especially LinkedIn · 247 pts · 2026-07-09
- LinkedIn Feed Blocker · 182 pts · 2026-08-08
- Translate Garry Tan's LinkedIn-speak to plain English · 115 pts · 2026-03-19
- Is Cloudflare Down Again? Also, DownDetector/Claude.ai/LinkedIn? · 31 pts · 2025-12-05
- LinkedIn Introduces a 'Seems Like AI Slop' Button · 14 pts · 2026-07-30
06 / Research velocity
cs.AI papers, all time
211.5K
arXiv category total
Newest submission
11-05
2026
- SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization
- Scientist World: Grow a Research World for AI-Scientists
- XStack: A Benchmark for ML Repo Migration
- Same Behavior, Different Mechanism: What Finite Experiments Can Certify About Integrated Information
- Towards interpreting any residual stream direction by scaling activation inversion
- ReproBridge: Building Reproducibility into ML Research Artifacts
07 / Probe run timeline
Probes paused: Not enough credits. Runs resume only after review — never automatically.
| Week of | Model | Source-grounded synthesis | Constraint writing | Evidence-based decision | Avg | Runs |
|---|---|---|---|---|---|---|
| 2026-10-26 | gpt-6-astra | 100 / 100 | 100 / 100 | 100 / 100 | 100 | 6 |
| 2026-10-19 | gpt-6-astra | 100 / 100 | 100 / 100 | 100 / 100 | 100 | 6 |
| 2026-10-12 | gpt-6-astra | 100 / 100 | 100 / 100 | 100 / 100 | 100 | 6 |