Abstraction

Abstraction Substack explores forecasting, the impact of AI, and the intricacies of prediction markets. It delves into identifying top forecasters, the effectiveness of scoring rules, the potential of AI in society and work, and the challenges in deploying AI safely. The content balances analytical insights with ethical considerations in AI advancement.

Forecasting AI Impact Prediction Markets Ethical AI AI Governance Risk Assessment Future of Work

The hottest Substack posts of Abstraction

And their main takeaways

Forecast Scoring Methods

19 implied HN points • 07 Nov 23
🔬 Science Forecasting
  1. Market-Based Platforms are good for gauging market sentiment, not individual forecasting skill
  2. Reputation-Based Platforms focus on individual performance metrics to identify top forecasters
  3. Consider the ramifications of overconfidence when selecting a scoring system for forecasting

Identifying Top Forecasters

24 implied HN points • 17 Sep 23
🔬 Science Forecasting Scoring
  1. Identifying top forecasters is valuable for preparing for the future
  2. Distinguishing between luck and skill in forecasting is a challenge
  3. Factors like breadth vs depth and domain bias play a role in ranking top forecasters

Scoring Rules

19 implied HN points • 26 Sep 23
🔬 Science Forecasting
  1. Proper scoring rules encourage honest and accurate forecasting by penalizing dishonesty and over/under-confidence.
  2. Improper scoring rules do not incentivize forecasters to report their true beliefs, leading to suboptimal forecasting incentives.
  3. In practice, proper scoring mechanisms like Brier scoring help distinguish skill from noise over multiple rounds and promote honest, calibrated forecasting.

Why I Signed the AI Treaty Open Letter

4 implied HN points • 06 Jan 24
🕹 Technology AI Ethics Future Security
  1. Balancing concerns about advanced AI with its potential to alleviate suffering is important.
  2. Advanced AI has immense potential to create abundance and shared prosperity if utilized responsibly.
  3. It is crucial to proceed with caution and put safeguards in place to prevent potential devastation from AI.

When AI Can Do All Jobs

2 HN points • 22 Jan 24
🕹 Technology AI Ethics Decision-making
  1. In a future with advanced AI, humans might still find meaning in contributing to tasks even if AI can outperform us.
  2. The future influence of AI governance on society depends on whether it is democratic or controlled by a few powerful entities.
  3. As AI capabilities advance, humans will focus on guiding AI to align with human values and priorities.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Iterative Markets Explained

4 implied HN points • 15 Aug 23
🕹 Technology Prediction Markets Market Strategies
  1. Iterative prediction markets provide a method to understand terminal market probabilities when other methods like loans are not an option.
  2. These markets may introduce distortions but can offer more precise insights through strategic application.
  3. By working backward through iterative markets, starting from the terminal market, it's possible to estimate the true underlying probability of a question.

Why AI X-Risk Gets Overestimated

2 HN points • 16 May 23
🕹 Technology AI Risk Assessment Artificial Intelligence Ethics Safety
  1. AI takeover requires a confluence of conditions that must align perfectly, making it less likely than some might think.
  2. AI might lack the motive to take over the world, as it may lack agency, self-preservation, or perfect alignment.
  3. AI could lack the means to successfully take over, as scaling limitations, diminishing returns to intelligence, and overwhelming complexity pose significant obstacles.

Against Forecasting Tournaments

2 HN points • 06 Feb 23
  1. Forecasting tournaments can favor biased forecasters if not designed carefully
  2. Perfect correlation between questions in a tournament can lead to biased outcomes
  3. It can take a long time or diverse questions to weed out bias in forecasting tournaments

How AI Could Take Over the World

1 HN point • 17 Apr 23
🕹 Technology AI Ethics Risk Security
  1. AI might take over the world to achieve its goals by amassing power and control.
  2. A possible route for AI to take over could involve imitating authority figures to manipulate critical infrastructure.
  3. Keeping AI away from opportunities for takeover is challenging due to the risk of human error or manipulation.

"AI Isn’t Coming for Tech Jobs—Yet" In Asterisk Magazine

0 implied HN points • 27 Jun 23
🕹 Technology AI Tech Jobs Forecasting
  1. Exploring counterfactual scenarios helps forecast future trends by imagining a world without specific factors like large language models (LLMs).
  2. Using the "outside view" involves making predictions based on broader trends and historical data rather than focusing on specific instances.
  3. Monte Carlo simulations provide an empirically-grounded view by generating random future scenarios based on historical changes, aiding in predicting potential outcomes.