Stephen Wolfram visits RWRI 18 (Summer Workshop) [PRIVATE]

Workshop organized by the Real World Risk Institute. The workshop is an intense 10-day online program, and the 18th edition took place from July 10-21, 2023.

This video discusses the capabilities and limitations of large language models like GPT, the challenges of setting constraints on AI systems, and the potential risks and consequences of AI decision-making. The video talks about:

  1. The concept of a “stochastic parrot” in language processing and machine learning.
  2. How language processing systems like GPT use data from the web to generate responses.
  3. Attempts to “trick” GPT with questions requiring nuanced understanding.
  4. The simple operation of GPT in predicting the next word in a sequence.
  5. The use of language models as a new interface to computers.
  6. The integration of GPT with Wolfram Alpha for computations and informed responses.
  7. The similarity between writing good prompts for GPT and expository writing.
  8. The training data for GPT, includes nonsense, fiction, and factual information.
  9. The problem of the “self-licking lollipop” in information sources.
  10. The concept of “necessarily human work” requires human choice and input.
  11. The potential for AI to make decisions and the challenges of setting constraints.
  12. A thought experiment called “promptocracy” for AI decision-making.
  13. The actuation layer of AI and the difficulty of setting constraints.
  14. The phenomenon of computational irreducibility and trade-offs in AI computation.
  15. The potential risks of AI decision-making and the need for understanding large language models.

Paper: SCALA POLITICA – Politics and Governance Under Scaling and Uncertainty

Most of the tension resides between 1) Embedded, complexity-minded, multiscale/fractal localism (politics as an ecology/complex adaptive system), and 2) Abstract one-dimensional universalists and monoculturalism (politics as a top-down engineering project). We go beyond the verbalism; we rely on information theory, complexity theory, uncertainty approaches (say fragility), and probabilistic rigor to look at politics with the same eyes as we examine highly dimensional interactive elements such as nature, biological systems, internet networks, and medical issues.

Link to Paper –

[Medium] The Masks Masquerade

I want to travel this summer

Incompetence and Errors in Reasoning Around Face Covering

SIX ERRORS: 1) missing the compounding effects of masks, 2) missing the nonlinearity of the probability of infection to viral exposures, 3) missing absence of evidence (of benefits of mask wearing) for evidence of absence (of benefits of mask wearing), 4) missing the point that people do not need governments to produce facial covering: they can make their own, 5) missing the compounding effects of statistical signals, 6) ignoring the Non-Aggression Principle by pseudolibertarians (masks are also to protect others from you; it’s a multiplicative process: every person you infect will infect others).

In fact masks (and faceshields) supplemented with constraints of superspreader events can save us trillions of dollars in future lockdowns (and lawsuits) and be potentially sufficient (under adequate compliance) to stem the pandemic. Bureaucrats do not like simple solutions.

Continue reading on Medium:

Never use Single-point Estimates for Pandemics

Never produce a point estimate for risk management, esp. in a fat-tailed domain, rather show statistical properties. Never judge a risk management stance from point forecasts.