Intro

Having embarked on two spontaneous short day-trips mostly due to the inspiration from the novel On the Road, I realised I share their trait of finding solace and peace in journeys, but I don’t need to cross the whole continent multiple times just for that.

This week should be my last of buffer vacation between my conclusion of studies and my official commencement of working full time, so I can still have the luxury of enjoying a very diverse selection of reading. I am also having a great time reading Oscar Wilde’s The Picture of Dorian Gray, and I cannot recall the last time I came across such beautiful English proses.

I am, however, still saving all the writing till the deadline of the newsletters set by myself, instead of writing down my thoughts on the fly right after finishing on an article. This is not necessarily damaging to the quality of the contents, because I can have some time to digest and brood over the readings, while I would also have a second chance to finetune them when I read them over again just before I write a few days later. BUT it does create a sizeable backlog on my agenda every week, so I shall try to shift my strategy just a bit in the coming weeks and see if it makes any difference to my time management difficulties.

With the relentless heat of England beating in my veins, I hope this issue can keep you cool wherever you are :))

Inbox

Chinese Businessmen: Superstition Doesn’t Count

• As a follow-up to Cedric’s last article on the analysis of successful Chinese businessmen, this piece further explored the role of superstition in their businesses, and why it is so pervasive yet it doesn’t appear to be a burden at all.
• The author posited that there are two common types of rationality: epistemic rationality and instrumental rationality, where the former concerns how we know that our beliefs are true, and the latter about how we make better decisions to achieve our goals. While epistemic rationality has attracted many scholars, and many research articles have been produced on the topic, it is instrumental rationality that “dominates when it comes to success in business and life”. He further argued that the “trial-and-error” practice common in most successful Chinese businessmen is one way of improving their instrumental rationality.
• Nonetheless, two alternative explanations were also offered:
  1. These businessmen don’t really examine their superstitions as much as they do their business models, and the beliefs can serve as a relief of psychological stress.
  2. The beliefs are held for pragmatic and social reasons, so that they can be accepted as a member of the in-group.

Introduction to Diffusion Models

• Although my own studies and work in ML have nothing to do with diffusion models, I have seen them trending all over my social media feeds, and I cannot help but look it up and catch the hype (the tail of it at least). This blog article presented a very clear introduction to me with a fine balance between visual explanations and maths.
• One sentence summary: it is a generative model trained by destroying the training data step-wise with Gaussian noises and then recovering the original image by reversing the process, and generation is done by passing random Gaussian noises into the denoising model.

Do-nothing scripting

• This article proposes a very practical and efficient way of approaching automation from zero to one. A main problem with automation is that we can only write the script to partially automate the procedure, and this might merely result in a new procedure of how to use the automation script. The do-nothing scripting, however, encodes the instructions of a procedure, encapsulating each step in a function, without actually performing the task. It prompts the user one step at a time for them to complete manually. The author then proposed three not-so-obvious merits of this approach:
  1. It’s less likely to lose our focus and skip a step.
  2. Each step is encapsulated in a function, which is possible to replace the instruction with actual codes performing it.
  3. Over time, an elaborate automation library can be developed, making future automation much easier.

Is medicine overrated?

• Growing up in a family with deep faiths in TCM(Traditional Chinese Medicine), and having been treated with it so many times, I hold a very complex yet ambivalent attitude towards both TCM and modern medicine. On the one hand, TCM was always the first-resort of my family to any nonlethal illnesses (which cover almost all common diseases), especially the chronic ones, but they almost never worked for me despite all the wondrous effects they claimed beforehand, and they often taste nasty. On the other hand, modern medicine, effective as it is, usually produces strong and unwanted side-effects, sometimes so serious that I might even wonder I should’ve just stuck to TCM. As I grew to the age where I can take my own medicine, as much as I am well aware that TCM would be more or less a placebo, I still willingly take it at times for minor conditions before having to resort to modern medicine.
• This blog article by Scientific American discusses the issues of modern medicine. Although the piece was centred around therapeutic nihilism, the author was by no means (nor am I) against science and modern medicine - rather, the nihilist sentiment was rooted in the hope of improving medicine when the current climate is more or less despairing. Here’s a summary of the key points:
  1. Medical research is slanted towards positive results
  2. The rigor of research on medical treatments is inversely proportional to the benefits found
  3. Drugs' harmful effects are underreported
  4. Healthcare providers engage in “disease-mongering”
  5. Screening doesn’t save lives
  6. Modern medicine is overrated
• In my own words, I think one salient point that drives all factors above would be “intricately vested interests”, be it commercial success, academic reputation, or personal financial gains.

Thoughts

Academic frauds

Two weeks ago when I first read the article blots on a field, a strong sense of sympathy came to me, as the ML community, unfortunately, is also permeated with academic fraud and disintegrity, and it’s just as challenging, if not more, to expose these misconducts. There are several very common fraudulent behaviours in ML, which can be easily generalised to the whole experimental academia:

1. Plagiarism - as this is the misconduct that’s the easiest to detect, it’s perhaps not as threatening as the other more implicit but undermining behaviours
2. Collusion rings - a clique of “renowned scholars” that use their influence as reviewers to sway the decisions of paper acceptance to their benefits.
3. Fake authors and paper riders - The group of people who claim authorship and credit on projects to which they make practically no contributions.
4. Blatant falsifications - yes, the researchers are so desperate that they have to fabricate data and results to boost their studies/career with fancy papers that “advance the field”. This, alas, is far from uncommon in the field, which leads to a big collection of unreproducible papers .
5. Truth but no the whole truth - This is perhaps the most common type of fraud that the vast majority in ML community commits. Some examples include cherry-picking datasets, seeds, hyperparameters, and results. However, they are so common and mistakable with heartless slips that the community pretty much acquiesced and everyone has to be equipped with the highest level of alert whenever they see better-than-SOTA results.

And I am particularly troubled by the last one on the list - not necessarily because the omission of the whole truth is the most detrimental, but the endless condoning and universal complicity almost strip the last shred of optimism away from me. I firmly believe that there indeed are respectable scholars who keep their works up to the academic rigour, but the overall environment would inevitably marginalize their influence should they not collude with the rest.

The term “career researchers” was defined in this article as those who are “motivated largely by the power and prestige that rewards those who excel in the academic system, rather than idealistic pursuit of scientific truth”. And I shudder to estimate the proportion of them in the contemporary ML community. When the motivations are corrupted at the start, there won’t be enough checks to stall the growing trend of subtle misconducts that could culminate in complete meaninglessness in research papers.

Even so, I believe certain countermeasures can still help thwart the momentum and greatly reduce the motivation of the frauds, subtle or not:

1. Metric shift - shifting the focus from minimal improvement in SOTA to less universal but more verifiable metrics.
2. Academic assessment system reform - scrapping the convenient assessment standards based on number of publications and a simplistic view of influence indices
3. Publication system revolution - as I have contended in the second issue, a multi-media publication system can address many existing problems of the existing paper-based journals, which can significantly block the attempts at blatant and subtle frauds.

These measures are by no means exhaustive, and each of them would be confronted with very strong protests from the academia, not because of their lack of efficacy, but rather the opposite. One way or another, I genuinely hope this ivory tower will not be left with a mere good-looking shell.

Brag documents and cringe culture (Part 2)

• Reputation saved (refer to my last issue)! And the fact that I managed to accomplish this sequel without bragging it on social media is already a testament to my sincerity of this piece of opinion. Picking up from where I left off last time - yes, that social media would be LinkedIn, no doubt. Surely we have all seen some humblebrags there with boastful claims shabbily disguised under a stunning facade of humility, where in effect the function of the modesty is to amplify their achievements they are trying to flaunt. And that’s just the noob-level cringe. If you have the luck of never seeing them, please have a look at this page - it’s never short of examples. Or, if you prefer, here’s a ranking of the most painful few.
• Why is LinkedIn like this? I was not the first person to wonder about it, no doubt. I love the perspective of this article with opportune illustrations all over the place. It suggests three main factors that contributed to the cringe:
  • The personality - the career-focused nature of LinkedIn pushes people to take on a radically different persona than their own identity, and they have to reinforce it with exaggerated wording and super-achieving outlook.
  • The customer - humblebrags are often incentivized as they can attract prospective employers.
  • The algorithm - the content we see is a result of interesting developments to drive the users to “derive value from the newsfeed when they feel heard, can start conversations and are able to swap advice”. This has been accomplished by new site designs, rise of “broetry”, optimising different metrics (such as prioritizing connections, dwell time, and real engagement), etc.
• I agree with all explanations provided by the author, and I have two points to add myself:
  • Demographics: in my own experience, there’s quite a strong correlation between real-life and online expressiveness in terms of bragging and sharing “inspirational” content, and I think LinkedIn alone usually cannot completely alter one’s identify by that much. However, since LinkedIn has already evolved to be the cringe-tolerant platform with many ready to share their achievements, it’s more likely that those who tend to flaunt a bit in real life would be more encouraged to do so on LinkedIn, and the platform will also, in turn, attract users of the same kind even more, thereby leading to an imbalanced active user demographics.
  • It is us who are the “cringe”: From my angle, the cringey LinkedIn users are awkward and insufferable, but I may well be a part of the minority. After all, the platform does intend to pander to the majority of users who may heartfully welcome the soft-skills and self-motivational posts that could, in a way, lighten themselves up and gain moral support among the exaggeration. I cannot even assert with 100% certainty that I might feel the same when I am more involved in businesses in the future. It would be truly comical if I start a post with “I am humbled to announce” and end it with “agree?” in my 40s, and this post is dug up alongside.

Miscellaneous

Less Wrong

• Less Wrong is a community dedicated to

 improving our reasoning and decision-making.

They seek to

 hold true beliefs and to be effective at accomplishing our goals. More generally, we work to develop and practice the art of human rationality.

• It is a rather large site (which surprised me even more given my complete ignorance of this forum), and I haven’t got the chance of reading through the most popular posts there, but frankly the refreshing and minimalist look of the website itself already captivated me. The post quality is also generally high with weekly highlights available too. There’s also a strong presence of ML discussions among many other intellectual topics.

As a companion to my discovery of this wonderful community, here’s the dessert - A list of common cognitive biases.

Penguin Great Ideas

• I have seen this series of books - Penguin Great Ideas many times in my local Waterstones, and I have always wanted to try some of them, especially all are very lightweight (with beautifully designed covers, I must say I am a thoroughly visual creature). And I think my first pick will be On the Shortness of Life by Seneca. Hailed as a must-read for stoicism, the book could hopefully enlighten me a bit on the topic, as I don’t feel I have truly understood stoicism.

Every Noise at Once

• An “awesome” collection of ALL music genres, subgenres, signature pieces, and everything in between. They adopt a word-cloud approach of presenting their comprehensive record of the musical genres and artists, alongside the traditional lists. For each genre they also proffer a number of lists (full playlist, intro, new, favorites) available on spotify and linked in the website. I was truly amazed by the finesse of this project and the sheer volume of the workload put into collating all the playlists. This will be my go-to place whenever I feel a bit adventurous in my music taste.

On the Road

• On the Road by Jack Kerouac opened up a new path for me in terms of reading experience as well as life experience. Refer to my book review for more information!