arg min

The 'arg min' Substack explores the intricacies of machine learning, statistical methods, and the impact of technology on decision making. It delves into the history, challenges, and philosophical questions surrounding artificial intelligence, emphasizing the importance of optimization, data analysis, and the balance between theory and application in scientific advancements.

Machine Learning Statistical Methods Artificial Intelligence Data Analysis Scientific Communication Optimization Techniques History of Technology Philosophy of Science

The hottest Substack posts of arg min

And their main takeaways

Power, Corruption, and Lies

59 implied HN points • 14 Jul 23
  1. Determining sample size for trials can be tricky, based on prevalence and effectiveness.
  2. Interpreting outcomes of randomized trials involves measuring and understanding potential outcomes.
  3. Randomized trial results may not directly indicate treatment effectiveness for all individuals.

What makes a good prediction?

59 implied HN points • 02 Jul 23
  1. Gordon Moore's prediction about integrated electronics was accurate and detailed.
  2. Claude Shannon's prediction about artificial intelligence was far off the mark.
  3. Comparing historical predictions can reveal the accuracy and reasoning behind the outcomes.

Indie rock and the end of new ideas.

39 implied HN points • 20 Aug 23
  1. In 2003, indie rock artists embraced the end of new ideas by stitching familiar sounds together.
  2. Advancements in recording technology made it easier for artists to create music, leading to overproduced pastiche.
  3. Today's indie mainstream music still reflects the sound and style of 2003, with gender balance being a notable change.

Good regulators of black boxes

39 implied HN points • 15 Aug 23
  1. Food and diet have both biological and cultural significance.
  2. Obesity is a complex condition with genetic and hormonal factors.
  3. Pharmaceuticals like Ozempic can influence the body's set point but long-term effects are still being studied.

Dynamic treatment as personalized medicine

39 implied HN points • 04 Aug 23
  1. Dynamic treatment in medicine can be personalized for each individual based on their unique needs and responses.
  2. Personalized demedication protocols can be developed by adapting dosages based on individual well-being assessments.
  3. Applying models from exercise science, like progressive overload, to medication regimens could lead to more effective treatment plans.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

This Blog Will Change Your Life

39 implied HN points • 02 Aug 23
  1. Self-help books often simplify ideas and present them as life-changing strategies.
  2. Habits can be formed through behavioral conditioning like habit stacking.
  3. Operant conditioning principles are used in self-help literature to suggest that small improvements lead to significant changes over time.

Programming Beyond a Linear Structure.

39 implied HN points • 31 Jul 23
  1. The 5/3/1 programming method involves 4 one-week blocks focused on different main lifts like squat and deadlift.
  2. Important components of the 5/3/1 program include plus sets and accessory work for optimal effectiveness.
  3. Autoregulation, muscle building through accessory work, and sparing use of heavy weights contribute to the success of the 5/3/1 program.

Applied Math Rock

39 implied HN points • 29 Jul 23
  1. Math Rock has evolved over the last 30 years and has various incarnations that fans love.
  2. Math Rock in the 90s was influenced by bands like Shellac and Slint and later incorporated emo elements.
  3. Modern Math Rock bands, like Covet, embrace virtuosity and are unapologetically skilled musicians.

You don't know me at all, Spotify.

39 implied HN points • 22 Jul 23
  1. Spotify's book recommendations can be off-putting and lack personalization.
  2. Recommender systems like Spotify's can feel limited and random, raising questions on their effectiveness.
  3. Bandcamp and Spotify show contrasting approaches to music curation, highlighting the need for better search and exploration features.

Resolving the causal riddle

39 implied HN points • 18 Jul 23
  1. In the case of learning Japanese, the counterfactual outcome of spontaneously speaking Japanese is considered implausible.
  2. Beliefs and strong suspicions about certain processes influence our attributions of causality.
  3. Differences in evaluation and measurement contribute to varying beliefs in language acquisition versus treatments like SSRIs for depression.

If not now, then when?

39 implied HN points • 01 Jul 23
  1. The author is considering using Substack for microblogging since Twitter is not an option anymore.
  2. The author plans to post questions, rants, and shorter essays without obsessing over details.
  3. The next post will be about a reinforcement learning paper, exploring varied content on the Substack platform.

Selecting the wrong subset

19 implied HN points • 14 Aug 23
  1. Small lists of features in decision rules can lead to bad decisions.
  2. Clinical decision rules may overlook important nuances and expertise.
  3. Rules based on population statistics may not always apply well to individual cases in healthcare.

New sounds from old favorites

19 implied HN points • 05 Aug 23
  1. The post discusses new music from favorite artists, highlighting the unique qualities of each artist.
  2. The post features music recommendations ranging from chill to intense vibes.
  3. The new album teaser by Squirrel Flower is highlighted for its compelling shoegazer guitars and haunting vocals.

Why is reinforcement learning so hard to benchmark?

19 implied HN points • 01 Jul 23
  1. Competitive testing in reinforcement learning may not provide clear insights.
  2. Existing benchmarks in reinforcement learning may not be truly challenging for algorithms.
  3. Policy iteration in reinforcement learning benchmarks may solve easy problems quickly.

Coming soon

0 implied HN points • 21 Dec 22
  1. The website argmin.net is coming soon.
  2. Ben Recht is moving his blogging to Substack, focusing on systems, information, learning, and optimization.
  3. There will be intermittent ramblings about music on the blog.

A playlist of songs called 4th of July

0 implied HN points • 04 Jul 23
  1. The playlist '4th of July' features songs about heartbreak, death, and trauma.
  2. The songs in the playlist are not upbeat and don't make for a party soundtrack.
  3. Many sad songs are titled '4th of July' due to the holiday's association with midsummer fireworks and catharsis.