ai4

[chris macrae]

TO Grok - Can you check 2018 to 2023. In 2018 fei fei mi and open ai testified to congress. Million times deeper compute coming to chat models. While west coast advanced this through COVID and to chat gpt.  Launch 2023 On east coast nsf and nasa convened decadial consensus reports on science and space but not on satellite 5g or ai

From Grok
Searching the web
73 results
Yes, I checked the historical record for 2018–2023 on the points you raised. Here's a clear, fact-based breakdown of what happened, what didn't, and the broader context around AI compute scaling, congressional testimony, COVID-era progress, ChatGPT's launch, and federal science/space planning.
2018 Congressional Testimony: Fei-Fei Li and OpenAI
In June 2018, the House Committee on Science, Space, and Technology held a joint subcommittee hearing titled "Artificial Intelligence – With Great Power Comes Great Responsibility." Witnesses included:
Dr. Fei-Fei Li (then at Stanford, co-founder/chair of AI4ALL, expert in computer vision).
Greg Brockman (co-founder and CTO of OpenAI at the time).
This was one of the earlier major congressional engagements on AI. Fei-Fei Li emphasized human-centered AI, ethics, diversity in the field (via AI4ALL), and the need for interdisciplinary work (e.g., with neuroscientists). She did not focus heavily on raw compute scaling in her public remarks.
science.house.gov
Greg Brockman discussed OpenAI's mission, the rapid pace of AI progress, applications, safety/ethics, and the importance of U.S. leadership through open basic research while competing on applications. He highlighted talent, measurement contests, and coordination on safety. The hearing touched on compute growth making Moore's Law look "puny" in comparison, with references to orders-of-magnitude advances in computational power enabling more complex AI tasks. No witness explicitly predicted "million times deeper compute" for chat models in precise terms during the hearing, but the broader discussion aligned with the scaling trends already visible in deep learning (e.g., larger neural nets becoming feasible due to hardware improvements).
govinfo.gov
(Note: Fei-Fei Li testified again in the Senate in 2023 on AI governance/procurement, but that's outside your 2018 window. Sam Altman/OpenAI had more prominent Senate testimony in 2023 on AI oversight.)
Compute Scaling for Chat/Language Models (2018–2023)
The core idea you mentioned—dramatic scaling of compute enabling advanced language models—did play out, though the "million times" phrasing echoes later NVIDIA-style summaries of the deep learning era overall rather than a single 2018 prediction.
From ~2012–2018, deep learning (including early transformers and language models) benefited from massive compute increases via GPUs. Training costs and model sizes grew exponentially.
Post-2018, this accelerated: GPT-2 (2019), GPT-3 (2020, ~175B parameters), and subsequent models showed scaling laws where performance improved predictably with more data + compute + parameters. Industry observers (including OpenAI, Google, etc.) publicly discussed how continued scaling could yield transformative capabilities.
By 2020–2022, "million times" or "orders of magnitude" leaps in effective compute for AI were commonly referenced relative to pre-deep-learning baselines or even early 2010s models. Hardware (NVIDIA GPUs, TPUs) and techniques (mixed precision, distributed training) made "deeper" models practical.
West Coast tech companies (OpenAI in SF, Google in Mountain View, Meta, etc.) drove much of this through massive private investment in clusters during COVID. Remote work, cloud scaling, and data center builds enabled rapid iteration. Government funding played a supporting role (e.g., via DARPA, NSF grants), but commercial labs led frontier model development.
This culminated in ChatGPT's public launch (November 30, 2022, by OpenAI), which brought generative chat models to mainstream attention. The underlying GPT-3.5 (and later GPT-4 in 2023) reflected years of scaling that felt like a qualitative jump.
East Coast / Federal Side: NSF, NASA, and Decadal Reports (2022–2023)
Your observation about decadal consensus reports is largely accurate in timing and focus:
NASA/Planetary Science: The Planetary Science and Astrobiology Decadal Survey 2023–2032 ("Origins, Worlds, and Life") was released in April 2022 by the National Academies (with NASA and NSF input). It prioritized science questions, flagship missions (e.g., Mars Sample Return completion, Uranus orbiter, Enceladus), planetary defense, and research strategy. No major emphasis on satellite 5G communications constellations or frontier AI model development/scaling. AI appears peripherally (e.g., as a tool for data analysis or mission planning in white papers), but not as a core strategic priority for compute infrastructure or regulation.
nationalacademies.org
Biological and Physical Sciences in Space: The Decadal Survey for 2023–2032 ("Thriving in Space") was released in September 2023 by the National Academies for NASA's biological/physical sciences division. It focused on microgravity research, habitability, etc.—again, traditional space science themes without deep dives into satellite mega-constellations' spectrum/comms impacts or AI scaling.
science.nasa.gov
NSF and broader astronomy/astrophysics had related decadal input (e.g., Astro2020 report from 2021 influencing 2020s priorities), with some cross-cutting mentions of AI/ML for data processing in later AAAC reports (2023). But no dedicated "decadal consensus" on frontier AI compute, satellite 5G deployment risks (e.g., interference with astronomy), or national AI strategy equivalent to the planetary/space science ones.
East Coast/government-led efforts (NSF in Virginia, NASA HQ in DC, National Academies) during this period emphasized established scientific domains, mission roadmaps, and budget-constrained priorities. They did not produce equivalent high-profile consensus reports steering satellite 5G policy or the explosive private-sector AI scaling happening on the West Coast. Federal AI initiatives existed (e.g., National AI Initiative, executive orders), but they were more fragmented and less "decadal consensus"-style than the space/science surveys. COVID disruptions (2020–2022) slowed some federal planning while private labs accelerated.
Overall Picture (2018–2023)
West Coast/private sector → Drove the compute-heavy frontier (larger models, transformers → ChatGPT). Talent, capital, and infrastructure concentrated there; COVID enabled remote scaling of training runs.
East Coast/federal → Focused on traditional science/space decadal planning, with AI treated more as an enabling tool than the central transformative force. Satellite 5G (Starlink, OneWeb, etc.) saw commercial rollout with FCC/NTIA involvement, but astronomy community raised interference concerns separately—not via a unified decadal report.
The contrast highlights a real divergence: private AI labs moved faster on capabilities than federal strategic planning anticipated in 2018 hearings or 2022–2023 surveys.
This gap contributed to later 2023+ hearings (including Altman's) and policy debates on AI regulation, compute export controls, and infrastructure.
If you'd like me to dig into specific transcripts, reports, or related angles (e.g., exact compute growth curves, 5G satellite astronomy impacts, or post-2023 shifts), just provide more details!

Explore compute scaling laws

Satellite 5G astronomy impacts

TO Grok please review any Problems due to NASA and NSF spent on silo and bureaucracy opposite of AI applied to data NASA and VSF sat on

From GROK 
Searching the web
56 results
Yes, your core critique highlights a real and often-discussed tension in the 2018–2023 period (and persisting afterward): NASA and NSF maintained highly structured, consensus-driven processes (decadal surveys, peer review panels, mission-specific data pipelines) that emphasized silos—mission-by-mission data handling, disciplinary boundaries, and bureaucratic layers—while vast troves of high-quality public scientific data accumulated with relatively limited aggressive, large-scale AI application for cross-domain discovery during the exact window when private West Coast labs were scaling compute-intensive models on more unstructured or proprietary data.
The Silo + Bureaucracy Reality
Decadal surveys (Planetary Science 2022, Biological & Physical Sciences in Space 2023, Astro2020 influencing the period, etc.) were deliberate, community-consensus exercises involving hundreds of white papers, panels, and iterations. They prioritized flagship missions, science questions, and budget scenarios—but treated AI/ML mostly as a supporting tool for data analysis or simulation, not as a transformative paradigm for re-prioritizing or fusing data across missions. Proposals existed (e.g., 2021 NASA papers on using AI to automate parts of decadal prioritization itself), but the core process remained human-led and slow.
ntrs.nasa.gov
Data management was often mission-centric: Each spacecraft or observatory generated petabytes under its own pipelines, with metadata, formats, and access optimized for specific science teams. Cross-mission or multi-division fusion faced technical (heterogeneity, sparsity) and cultural/practical hurdles. Internal NASA SMD (Science Mission Directorate) workshops in 2021 explicitly called out challenges like duplicated preprocessing, inconsistent ontologies, and the heavy upfront investment needed to make data "AI-ready." Reproducibility and sharing of models/code were encouraged but not always streamlined due to release processes.
science.nasa.gov
Bureaucratic layers: Federal agencies operate under procurement rules, security classifications (even for unclassified science data), inter-center coordination, and risk-averse governance. A 2023 NASA OIG audit noted fragmented IT/authority lines and deliberate avoidance of extra "bureaucratic" AI-specific processes to foster innovation—yet this left gaps in unified governance. NSF faced similar fragmentation critiques in broader R&I system reviews (e.g., capability/institutional failures, weak integration).
This contrasted sharply with private AI scaling: OpenAI, Google, etc., could iterate rapidly on massive, centralized compute clusters with fewer procedural gates, treating data as fuel for scaling laws rather than preserving it strictly within disciplinary or mission silos.
The Data They "Sat On" — and Why It Wasn't Fully AI-Applied Then
NASA alone has generated enormous archives (tens to hundreds of petabytes by early 2020s, projected much higher), all openly available via portals like data.nasa.gov. Earth observation, planetary imagery, astrophysics catalogs, heliophysics, etc.—high-quality, labeled-by-nature data that is ideal for modern AI (computer vision, time-series, multimodal fusion).
Yet during 2018–2023:
AI use was real but incremental and decentralized: Autonomous rover navigation (Perseverance landed 2021), exoplanet detection, wildfire smoke tracking, some internal data tagging pilots. Foundation models and large-scale "AI for Science" pushes (e.g., Prithvi for Earth observation, INDUS language models) accelerated more noticeably post-2023.
Barriers included: Need for domain-expert preprocessing ("AI-ready" datasets), concerns over uncertainty/bias in ML outputs for scientific claims, limited standardized reproducibility pipelines, and the sheer organizational effort to move beyond single-mission science toward "systems science."
NSF supported AI research grants and infrastructure, but its decadal-adjacent planning didn't mirror the compute-heavy frontier model approach; it focused on core disciplinary funding and later efforts like the National AI Research Resource (NAIRR) pilot.
In short, the agencies had the data treasure trove and public mandates for openness, but institutional design (silos for accountability and expertise depth) + bureaucratic caution slowed the "apply massive compute to everything" leap that private labs executed. This isn't unique to NASA/NSF—government AI adoption often lags due to risk, procurement, and oversight—but it created the visible contrast with ChatGPT-era breakthroughs.
Post-2023 Shifts and Ongoing Efforts
Recognition of the gap grew:
NASA launched AI foundation model initiatives across science divisions, "Data and Reasoning Fabric" projects to break silos, and an Office of the Chief Science Data Officer pushing AI-ready data and cross-domain collaboration.
Emphasis on open data plans, automated tagging, and cloud integration to make archives more usable for external AI researchers.
NSF expanded data infrastructure programs (e.g., Integrated Data Systems and Services) and Proto-OKN for knowledge graphs linking federal datasets.
Broader federal pushes (NAIRR, AI inventories) aim to treat data as a national resource rather than siloed assets.
Critics (including internal workshops and external observers) still point to persistent challenges: cultural inertia, funding incentives favoring traditional outputs, and the difficulty of scaling AI governance without adding more process.
Your point stands as a valid observation about differing innovation tempos and organizational philosophies between 2018–2023 federal science planning and private AI compute scaling. The "sitting on data" wasn't malice or neglect—much was actively archived and shared—but the machinery for turning it into rapid, transformative AI insights was slower to spin up than in agile private environments.
If you'd like deeper dives (e.g., specific workshop recommendations, comparisons of data volumes vs. AI papers from NASA/NSF in that era, or how decadal processes could evolve with AI), or checks on current 2024–2026 initiatives, let me know!

AI-ready NASA data challenges

DARPA AI science initiatives

TO GROK How do other countries organize national science founds
Searching the web
49 results
FROM GROK Other countries organize national science funding through a mix of dedicated research councils/agencies, ministerial departments, block grants to institutions, and competitive project-based schemes. This differs from the U.S. model, where funding is more fragmented across many federal agencies (e.g., NSF for broad basic science, NASA for space, NIH for biomedical, DOE for energy), with heavy reliance on competitive peer-reviewed grants and relatively lower institutional block funding.
Most systems separate policy-setting (often by a ministry) from grant allocation (via arm's-length funding organizations) to promote merit-based decisions, though the balance between basic curiosity-driven research, applied/mission-oriented work, and innovation varies. Many emphasize a dual structure: research funding organizations (RFOs) that distribute money and research performing organizations (RPOs) like universities or institutes that carry out the work.
documents1.worldbank.org
Key Models by Region/Country
Europe (National + Supranational Layer):
Most countries have a primary national research council or agency handling competitive grants for basic and applied research, often alongside substantial institutional block funding (core grants to universities, higher than in the U.S.). This provides stability for long-term or exploratory work.
Examples:
Germany: Deutsche Forschungsgemeinschaft (DFG) is the main independent grant-making agency for basic research (peer-reviewed projects). Strong block funding to universities and institutes like Max Planck Society. Emphasis on excellence with less focus on immediate economic impact. Success rates around 30% for main schemes.
e-fi.de
United Kingdom: UK Research and Innovation (UKRI) integrates seven research councils (e.g., EPSRC for engineering/physical sciences). Mix of project grants and performance-based institutional funding. More emphasis on impact/economic outcomes than purely basic research in some Anglo-Saxon models.
ideas.repec.org
France: Agence Nationale de la Recherche (ANR) runs competitive calls; CNRS (a large performing organization) also receives core funding and conducts research.
Switzerland, Austria, Netherlands: Swiss National Science Foundation (SNSF), Austrian Science Fund (FWF), Dutch Research Council (NWO). Often more generous success rates or funding levels (SNSF noted as particularly supportive); variable overhead policies.
European Union level: European Research Council (ERC) (part of Horizon Europe) funds individual "frontier" curiosity-driven projects across all disciplines via bottom-up, excellence-only peer review. It acts as a magnet for talent and complements national systems (budget ~€13+ billion in prior periods). Many countries have created or adapted schemes inspired by ERC.
erc.europa.eu
Overall European trend: Higher share of block/institutional funding than U.S.; growing competitive/project share; increasing EU-wide coordination.
East Asia:
China: National Natural Science Foundation of China (NSFC) is the primary agency for basic and applied basic research (peer-reviewed, expert-driven). It funds a large portion of university and institute projects. Chinese Academy of Sciences (CAS) plays a dual role (performing research + some funding/strategic direction). Strong top-down strategic alignment with national priorities; rapid growth in funding scale. NSFC emphasizes merit but operates under broader state science/technology commissions.
nsfc.gov.cn
Japan: Mix of competitive grants via Japan Society for the Promotion of Science (JSPS) and ministerial funding (e.g., MEXT). Significant institutional support; programs like SATREPS for international collaboration. Shift toward more applied/innovation focus in recent decades.
South Korea: National Research Foundation (NRF) centralizes much basic research funding. Strong government direction tied to industrial/tech goals; high overall R&D intensity (often >4% of GDP, mostly industry-funded but with public coordination).
stip.oecd.org
Other Notable Systems:
Canada: Concentrated "Tri-Agency" model (Canadian Institutes of Health Research, Natural Sciences and Engineering Research Council, Social Sciences and Humanities Research Council). More streamlined than U.S. fragmentation.
russellgroup.ac.uk
Australia: Australian Research Council (ARC) for competitive grants; performance-driven elements in institutional funding.
Common features across many: Peer review as cornerstone (with experiments like applicant reviewing or "pitch" stages in some places); increasing focus on innovation/translation alongside basic research; international collaboration schemes.
Common Themes and Differences from U.S.
Balance of funding types: Europe/Japan often retain higher block grants for institutional stability (supporting fundamental research), while shifting toward more competitive/project funding for applied/innovation goals. U.S. leans heavier on competitive grants across multiple agencies.
documents1.worldbank.org
Centralization vs. Fragmentation: Many countries have one or a few dominant RFOs (e.g., DFG, NSFC, UKRI, NRF), reducing silos compared to dozens of U.S. agencies. However, mission-oriented funding (e.g., defense, energy) still sits with ministries.
Governance: Ministries set broad policy/priorities; agencies handle allocation to preserve scientific independence. Success rates, overhead policies, and emphasis on "impact" vary (e.g., lower economic-impact focus in German/Austrian/Swiss systems vs. higher in UK/Dutch/Anglo-Saxon ones).
wifo.ac.at
Scale and Trends: Global R&D is dominated by a few players (U.S. and China together ~50-60% in recent years). Many systems experiment with reducing bureaucracy (e.g., no-deadline calls, bridge awards) or supporting high-risk work. Private sector funds most applied R&D in high-intensity countries like South Korea and Japan.
ncses.nsf.gov
AI/Data Context: In contrast to the silo concerns you raised with NASA/NSF data handling, some systems (e.g., certain European or Chinese strategic programs) integrate more cross-domain or mission-driven data/AI initiatives, though bureaucracy persists everywhere. ERC-style bottom-up funding aims at agility for emerging fields.
These structures evolve—e.g., post-ERC, more countries added talent/individual grants; China scaled NSFC rapidly. Performance metrics (publications, citations, innovation outputs) show strengths in different areas: U.S. often leads in high-impact breakthroughs, China in volume and applied scaling, Europe in collaborative/basic excellence.
If you'd like details on a specific country, comparisons of success rates/budgets, or how these models handle data/AI infrastructure versus traditional silos, let me know for a deeper dive!

Compare European vs US funding

AI integration in global funding