Observed Signal · May 22, 2026 · Analysis · Source: Astral Codex Ten · Impact: 2/5 · Sentiment: Neutral
New Paradigms Won't Save You
An essay by Scott Alexander (Astral Codex Ten) argues that expecting a radical new paradigm to prevent LLMs from reaching AGI is a weak hedge. He outlines a sequence of landmark AI milestones (from mid-20th-century neural nets through transformers and RLHF) and contends AGI will likely remain within the neural-network/deep-learning family. Using Lindy’s Law, Alexander estimates non-negligible probabilities that paradigm-level advances comparable to transformers or deep learning could arrive within years rather than decades, and gives a 25th-percentile forecast of roughly 3–5 years for such advances. He also notes that frontier labs will pursue new paradigms aggressively only if scaling of current approaches stalls. The piece engages skeptics (e.g., Yann LeCun, Gary Marcus) and discusses forecasting methods.
An analytical essay about LLM scaling and timelines for paradigm shifts offers context relevant to AI adoption and product roadmaps (conversational AI, generative models) that can influence AdTech/MarTech strategies, but it is opinion analysis rather than a platform policy or technical release.
Track Substack Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Article published on 2026-05-22.
- Author: Scott Alexander (Astral Codex Ten).
- Author lists AI milestones: 1950s neural networks; 1967 multi-layer perceptron; 2010 modern deep learning; 2017 transformer/LLM; 2022 RLHF and chatbots; 2024 chain-of-thought / test-time compute.
- Article cites skeptics Yann LeCun and Gary Marcus as questioning transformers/LLMs as the final path to AGI.
- Author applies Lindy’s Law and gives a 25th-percentile estimate that a paradigm-shifting advance comparable to LLMs could occur within ~3 years (and an advance comparable to deep learning within ~5 years).
Connected Companies & Entities
2 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
The Sigmoids Won't Save You
Opinion essay by Scott Alexander (Astral Codex Ten) arguing that the claim “all exponentials eventually become sigmoids” is technically true but can be misleading for forecasting AI capability growth. The author surveys historical examples (epidemics, airspeed records), recurring forecasting errors (UN birthrate projections, solar deployment WEO lines), and recent AI capability curves (METR). He recommends Lindy’s Law as a default prior under deep uncertainty—i.e., a trend that has persisted for a period is, absent domain knowledge, likely to continue about as long again—and shows how that yields non-trivial odds that current exponential AI growth continues for years. The piece urges critics who claim a near-term plateau to supply explicit models or acknowledge Lindy’s prior.
Neo Labs and AI Limits in Math and Science
This analytical essay (published June 30, 2026) examines whether cutting-edge AI research labs — dubbed “Neo Labs” — can materially advance scientific discovery, using a recent development in mathematics as a framing device. The author reviews macro signals (the Bank for International Settlements comparing the 2025–2026 AI investment cycle to past historical booms) and market estimates (Exponential View’s generative-AI revenue figures). The piece catalogs a set of well‑funded AI startups and research labs (Mirendil, Engram, General Intuition, Prometheus, Trajectory) and summarises their technical emphases: self‑improving researcher agents, persistent learned memory layers, world/agent models, physical-AI and continual learning platforms. The article balances optimism about accelerated R&D and automated labs with cautions about capital intensity, supply‑chain pressures (e.g., HBM memory shortages) and unresolved scientific limits, illustrated through the long‑running Erdős unit‑distance problem.
Author's AI Timelines, Risks, and Policy Views
An opinion essay (Astral Codex Ten) laying out the author's probabilistic beliefs about AI timelines, diffusion, and existential risk. The author defines and quantifies several timeline concepts (AGI, diffusion gap, superhuman gap, Bostromian superintelligence gap, and point of no return), expresses a 20% baseline probability of human extinction from AI absent strong interventions, and discusses alignment, regulatory pauses (US–China negotiations), and socio-economic outcomes (underclass, utopia). The piece gives a modal scenario of AGI around 2031 with broad diffusion through the 2030s and raises uncertainty about recursive self-improvement, compute bottlenecks, and the effectiveness of alignment strategies like interpretability and RLAIF.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.