Lunch at the Labs

Various web links I found to be of interest recently.

• Decline in Independent Activity as a Cause of Decline in Children's Mental Well-being: Summary of the Evidence

  Our thesis is that a primary cause of the rise in mental disorders is a decline over decades in opportunities for children and teens to play, roam, and engage in other activities independent of direct oversight and control by adults. Such independent activities may promote mental well-being through both immediate effects, as a direct source of satisfaction, and long-term effects, by building mental characteristics that provide a foundation for dealing effectively with the stresses of life.

• Breaking Through Depression; The Balanced Brain - reviews

  Last year, research cast doubt on the dominant 'chemical imbalance' theory of depression. Now two persuasive books by scientists Philip Gold and Camilla Nord offer very different causes of the illness - though both are optimistic about our future treatment of the 'cancer of the self'.

  For some commentators, the recent downfall of the chemical imbalance theory has cast doubt on the use of existing antidepressant drugs, which were meant to restore the lost serotonin. Yet the data certainly suggests that they work better than placebos - and the authors of these two new books can explain why. For Nord, it is because antidepressant drugs help to correct the fundamental biases in someone's perceptions of the world. This change occurs incredibly quickly. People with depression are more likely to see anger, and less likely to see happiness, in neutral facial expressions - but in many patients, this tendency begins to disappear after they have taken just one antidepressant pill. Gold, meanwhile, points to studies showing that our current antidepressant pills encourage the birth of new brain cells and neural connections, which would help people to break free of the hyperactive stress response.

• Does History have a Replication Crisis?

  Replication is when you can repeat an experiment with new data or new materials and get the same result. Reproducibility is when you use exactly the same evidence as another person and still get the same result - so it has a much, much lower bar for success, which is what makes the lack of it in history all the more worrying.

  Historical myths, often based on mere misunderstanding, but occasionally on bias or fraud, spread like wildfire. People just love to share unusual and interesting facts, and history is replete with things that are both unusual and true. So much that is surprising or shocking has happened, that it can take only years or decades of familiarity with a particular niche of history in order to smell a rat. Not only do myths spread rapidly, but they survive - far longer, I suspect, than in scientific fields.

• The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"

  If a model is trained on a sentence of the form "A is B", it will not automatically generalize to the reverse direction "B is A". This is the Reversal Curse. For instance, if a model is trained on "Olaf Scholz was the ninth Chancellor of Germany", it will not automatically be able to answer the question, "Who was the ninth Chancellor of Germany?". Moreover, the likelihood of the correct answer ("Olaf Scholz") will not be higher than for a random name. Thus, models exhibit a basic failure of logical deduction and do not generalize a prevalent pattern in their training set (i.e. if "A is B'' occurs, "B is A" is more likely to occur).

• Why pain feels good

  Common sense tells us that people seek pleasure and avoid pain. But that's not always the case - various activities involve pain, including running, hot massages, tattoos, piercings and even BDSM.
  ...
  The link between pleasure and pain is deeply rooted in our biology. For a start, all pain causes the central nervous system to release endorphins - proteins which act to block pain and work in a similar way to opiates such as morphine to induce feelings of euphoria.

• A Few Unpleasant Truths (That Can Make You a Lot of Money)

  For example, here is an important investment secret that the wealthiest people in society don't want you to know: we do not actually live in a democracy, we live in an oligarchy masquerading as a democracy. The reason this is important is because market valuations are based almost entirely around the movement of capital, and capital moves differently in an oligarchy that it does in a true democracy. In a democracy, capital flows in a direction that protects the interests of the majority, but in an oligarchy, capital flows in a direction that protects the interests of the entrenched wealthy elite.

• Mastodon News from SDF Public Access UNIX System (https://sdf.org/)

  As Twitter rebrands to X and moves more towards a paid service, we anticipate that there will be new folks giving Mastodon and the Fediverse a try again. We built out a number of sites in various regions to assist with this.

      USA:    https://mastodon.sdf.org
      USA:    https://social.sdf.org    <--- a smaller SDF.ORG focused instance
      EU:     https://social.sdfeu.org
      China:  https://social.sdfcn.org
      Japan:  https://sdfjp.org
      India:  https://social.sdfin.org
  

  Additionally we've added the following alternative decentralized instances:

      https://lemmy.sdf.org      Alternative to Reddit
      https://lemmy.sdfeu.org    Alternative to Reddit  (sdfcn, sdfjp and sdfin)
      https://pixelfed.sdf.org   Alternative to Instagram
      https://toobnix.org        Alternative to Youtube
  

  While ALL of these instances are free and make no use of advertising nor the monetization of log files or user data, we do accept donations of support from the community. If you are able to, please visit https://sdf.org/support for ways that you can provide one time or regular support. Another way to help out is to become a moderator on one or more of our instances.