D-Machine

Who cares? So much books are just slop (pulp), or, these days, over 70% filler. Even if they are "decent" fiction or non-fiction, the majority still are almost entirely shallow entertainment, whether explicitly or implicitly as simplistic "edutainment" of some kind. There are blogs, articles, technical papers, heck, even user discussions that contain far greater depth and/or demand far greater attention and mental work than the vast majority of "books" out there today.

Just thinking "reading books" is something good or impressive borders on anti-intellectual in the world of the internet. A much better indicator of real intelligence is e.g., does a person read actual scientific papers or technical documents, or sites like ACX, HN, SeriousEats (or any other site which dives into any hobby or art with research and with long-form articles), do they know about e.g. SciHub and LibGen and Anna's Archive, do they know about people like James Hoffmann if they are into coffee, or Kenji Lopez if into home cooking, and I'm sure hundreds of other careful and obscure podcasts and individuals, discussion forums, and other digital textual sources.

Yes, please have read some serious books and works in your life, at some point (preferably some classic and modern literature and philosophy, but anything with real depth is good). But worship of "books" simpliciter is pure midwit in 2025 (and was so already in 2010, at bare minimum).

It isn't remotely true even in the US, anyone claiming this doesn't know how to cook anything.

Anyone who believes something like this you can be 100% sure doesn't know how to cook even the most basic of staple foods. Cooking your own food is nearly an order of magnitude cheaper and, with a few cheap spices and seasonings, almost always tastier. The only valid argument is prep time here, and that too even only applies to certain kinds of foods.

Heritability IMO falls into the same bin as "standardized/relative effect sizes" (e.g. correlation coefficients, Cohen's d, odds-ratios, "explained variance", relative risk, etc), in that a division / re-scaling is introduced to supposedly increase interpretability, but, in reality, this has precisely the opposite effect.

Heritability is a bit worse though because the variance is partitioned into three giant piles of mush, at least two of which piles are very poorly measured / controlled at all.

Lifespan is a quite skewed distribution, so the SD looks large because it is in fact a poor summary of the bulk of the distribution. The actual part we care about for age-related mortality is narrower than such an SD would imply if we had a normal distribution (simple image example: https://biology.stackexchange.com/a/87851).

It is almost never reasonable to assume normality and make calculations like this. This is particularly the case when you are dealing with lifespan, which isn't normally-distributed even in the slightest. The actual ranges are likely smaller than you are stating here, and variance is just not a very practical or interpretable metric to use when dealing with such a skewed distribution.

We should be stating something like a probability density interval (i.e. what is the actual range / interval that 95% of age-related deaths occur within), and then re-framing how much genetic variation can explain within that range, or something like it. As it is presented in the headline / takeaway, the heritability estimate is almost impossible to translate into anything properly interpretable.

https://biology.stackexchange.com/questions/87850/why-isnt-l...

Why does this whole post read like very standard / default personality ChatGPT output though?

Thank you for saving me the time writing this. Nothing screams midwit like "Em-dash = AI". If AI detection was this easy, we wouldn't have the issues we have today.

"Pattern matching" is not sufficiently specified here for us to say if LLMs do pattern matching or not. E.g. we can say that an LLM predicts the next token because that token (or rather, its embedding) is the best "match" to the previous tokens, which form a path ("pattern") in embedding space. In this sense LLMs are most definitely pattern matching. Under other formulations of the term, they may not be (e.g. when pattern matching refers to abstraction or abstracting to actual logical patterns, rather than strictly semantic patterns).

I'd need a citation for that statistic, and I'd also need to see which themes are actually used.

> IMO, the best themes do typically have minimal/functional highlights, which results in more text that is the base color

And IMO, those are the worst themes.

These things are just preferences, but it is an objective fact that a good highlighting scheme makes certain information immediately visible, without requiring the reader to parse the actual characters. Whether or not this information is something you find helpful or annoying depends on your processing styles and preferences.

I have about 1500 lines in my VSCode settings.json dedicated to custom syntax highlighting and text decorations (this could be trimmed, some is from before the days of semantic highlighting), but regardless, the amount of differentiation I can achieve with this is simply not possible on a light background. I've tried! (Solarized light is a nice theme though)

Code can be read without any syntax highlighting or text decoration, obviously. But adding those things is an additional information stream that makes processing faster and more reliable (redundancy in general has this effect).

As you said, it is especially useful for making certain code smells instantly visible at a glance.

I also find that different kinds of code will get different "color rhythms" (e.g. low-level algorithmic code vs. high-level code that calls a lot of functions vs. code that does a lot of operations / mutation of object or class properties) when syntax highlighting is properly semantic. This makes scanning for certain types of things (where objects are being mutated, where variables are introduced, etc) extremely fast, since you don't even need to read the characters.

I also find that rich syntax highlighting makes the codebase easier to remember, since the color (along with things like the line-lengths) gives each function a sort of unique visual texture.

Of course, all of this is personal preference. I am a very visual thinker so this kind of stuff helps a lot for me. Some people are far more verbal in their mental imagery or may remember code chunks solely based on semantics. Then, obviously, a bunch of color and/or text decorations might not matter much, or even just be a distraction.

Seconding this, light themes cripple syntax highlighting, which in turn makes it far more annoying to quickly scan through code and glean structure. You can make up for this to some degree with text decorations, but, well, with dark schemes you have that too.

This is a pretty awful post, the problem is his example uses a horrible syntax highlighting scheme that makes use of far too few colours and no other text decorations.

In a competent highlighting scheme, you have enough differentiation that every distinct type of thing indeed has a different way it pops.

SeriousEats is great most of the time, and if you can "acquire" copies of any of the Modernist series (Modernist Cuisine, Modernist Cuisine at Home, Modernist Bread, Modernist Pizza), those are all done by mass with baker's percentages.

Heh, true, good point.

But yeah, it is the messiness and art of it that keeps it fun for me (especially after a day of math and coding)!

>> Kitchen work is all about proportions

> Only in Imperial/United States customary units.

Cooking is only about proportions in some very narrow fields (e.g. baking), and, even then, adjustment to ingredients, environment, and other contextual factors is paramount, and most adjustments need to be non-linear (whether by mass, volume, or surface-area). If the proportions are anything other than guidelines, you are doing mediocre cooking, at best.

Sifting, IMO from experience, does not solve the mass-to-volume ratio problem enough compared to just going by mass.

As a quick sanity test, if it did, serious baking resources would just always specify to use sifted flour (as this is easier and requires less equipment than a scale), but since they don't (e.g. Modernist Bread/Pizza, if you really demand a citation), you can infer that sifting is not effective in making reproducible results. Also, note e.g. chemistry is not done using sifted volumes (peruse quickly the amount of articles trying to assess the bulk vs "tapped density" of various powders: https://scholar.google.ca/scholar?hl=en&as_sdt=0%2C5&q=%22ta...). This should cause some skepticism about claims that sifting your flour is going to make baking results particularly consistent.

Sifting definitely helps remove variance (especially if you always buy the same flour and use the same sifting method into the same bowl, and then put un-needed sifted powder back into the jar), but IMO is far inferior to just weighing.

You're still right everyone overthinks home baking. Precision only matters if you are aiming for perfection, and even a horribly misspecified recipe made at home, but consumed fresh, is still generally going to be good, and definitely better than anything you buy at a supermarket. (And this is precisely why using a slide rule for precision is massively missing the point). As you said, there are many indicators that are more important to pay attention to.

> you're going to be estimating anyway once you've gotten familiar with a recipe

I would disagree slightly for this when it comes to making precise doughs or other things like brines, syrups, candy, and etc. Or at least I would change "estimating" to "adjusting" in your statement above. When it comes to trying something new (whether in baking from a proper source, like e.g. Modernist Bread or Modernist Pizza, or otherwise), a scale is invaluable.

But yeah, once you have some something a few times and have the feel, you can convert to volumes and go based on your senses. There's a baseline science / formula to some cooking, but the rest really is art.

This feels like a nit, because really I am just glad to see someone else pointing out the obvious realities here. While I would be hesitant to try Mr. Slide Rule's cooking, I'd try your cooking without fear!

The imprecision of volumetric measurements can absolutely ruin much baking, and many other recipes based on things like surface areas, or where the perception of flavours does not scale linearly with things like either volume or mass of the ingredient.

You're right volumes seem easier, at first blush, but the cost of this easiness is a dramatic / considerable reduction in consistency, compared to when measuring by mass.

Once you switch to regularly scaling by mass (just as a guideline, and still adjusting to taste, texture, and other factors), you'll realize the apparent easiness of volumes is pure illusion, and actually makes getting good results much harder.

Baking--along with fermentation, curing, and certain brines or other solutions--is the subset of cooking where accuracy of the masses of ingredients matters more than most others.

And yet still you are right you must often adjust significantly in baking for other factors (temperature + yeast activity, humidity, flour grind and composition, and general feel on kneading).

This. Belief in linear scaling of recipes is such a quick tell for someone who hasn't done even the most basic home cooking (or someone who has no sense of taste / texture at all).

> 2 cups of flour works regardless of the size of your cup

This couldn't be more wrong and no serious baking is done by volume for dry ingredients (flour, yeast, sugar, salt preferments, other additives).

EDIT: It is clear from your other comments you almost certainly know what you are doing, but this particular part is very wrong. You can't measure powders reliably by volume, regardless of sifting, tapping, or tamping.

This is impossible for most ingredients because many ingredients (flour, oil, or almost all such ingredients) vary considerably depending on packing, composition, and a whole host of other factors, and, also, not all recipes need to be scaled by mass.

If you see a recipe involving flour and it uses volume, it is trash, will not be reproducible. All serious baking is done by mass and mass only, except for glazes / coatings and/or if a very specific product / brand is specified. EDIT: as another commenter here noted, yeast also does not scale linearly (obviously) except in special cases.

Also, oils in general should be measured neither by volume nor by mass, but relative to what they need to coat / submerge (be that an ingredient, a cooking surface, or some combination of the two), or, for deep-frying, based on the amount needed to not drop temperature too significantly for whatever batch you are frying. That is, much cooking is about surface areas of your ingredients.

In general seasoning (or saucing) anything solid is more about exposed surface area than mass, and this depends on things like cut sizes, evaporation shrinking, and god knows what other factors. It doesn't scale with simple math, because there are all sorts of other factors involved that complicate this (surface texture just being one).

It is also all moot because ingredients (especially spices) have massive variance in potency, sweetness, bitterness, sourness, etc., so recipes are only ever a guideline. I.e. if you double a spice that is twice / half as potent as expected, you can get an unpalatable / bland dish, and IMO factors between 0.25 to 4.00 are extremely common for plenty of ingredients. So you always just need to taste and adjust accordingly. This is also ignoring that certain ingredients can vary in multiple dimensions (e.g. a lemon that is a lot sweeter than expected but less sour, and so simple scaling of the lemon alone can't get you want want: you need to reach for white sugar and/or citric acid to get your desired pH and sweetness).

It is also a fantasy that all flavour concentrations are perceived linearly anyway (and this is especially the case for acidity / sour / pH generally, but also spiciness in e.g. ginger, pepper, capsaicin).

As another commenter noted, few things in cooking actually scale linearly, and, in general, if you are following recipes mechanically like this, you produce sub-par results. You always have to adjust quantities for ingredient freshness, humidity, ingredient variance, and other variables, so recipes are only ever guidelines at best. And seasoning is always to taste (your own, and whomever you are cooking for) anyway.

But, sure, I guess this helps you scale up those guidelines in some rare cases where that math isn't trivial to do in your head...

Ugh, you're right. This was not intended. Conflating LLMs with GenAI is a serious error, but you're right, it is obviously a far more common error than I realized. I clearly should have said "move beyond solely LLMs" or "move beyond LLMs in isolation", perhaps this would have avoided the confusion.

This is a really hopeful result for GenAI (fitting deep models tuned by gradient descent on large amounts of data), and IMO this is possible because of specific domain knowledge and approaches that aren't there in the usual LLM approaches.

Sigh. If I start with a pre-trained LLM architecture, and then do extensive further training / fine-tuning with different data and loss functions and custom similarity metrics for specialized search and specialized training procedures, and use feedback from other automated systems, we are far, far more than an LLM. That's the point. Calling something like this an LLM is as deeply misleading as calling AlphaFold an LLM. These tools goes far beyond simple LLMs. The special losses and metrics are really so important here and are why these tools can be so game-changing.

The "sequential treatment" or "tailored treatment" approach is at least plausible and what is done in practice, yes, if the prescribing doctor is good, and if this is feasible for the patient.

However, since this takes time, and most depression is temporary, it is hard to know if you really are tailoring the medication to the person in many cases, or it has just been long enough you are seeing regression to the mean (or a placebo response, which is still strong even in treatment-resistant depression https://jamanetwork.com/journals/jamanetworkopen/fullarticle...).

There aren't really any double-blinded or even just properly placebo-controlled / no-treatment controlled studies to test this, but the closest thing to looking at the sequential approach also doesn't find very impressive results (https://bmjopen.bmj.com/content/13/7/e063095.abstract).

I do believe the drugs help some people, and almost certainly take some experimentation / tailoring. The average effects are just very weak.

Nope, you can't say how many people return to average from standardized effect sizes. I wish we had a standardized effect size that was more useful and actually meant something. Cohen actually proposed something called a U3 statistic that told us the percent overlap of two distributions, but that still doesn't tell us anything meaningful about practical significance.

You can't make decisions / determine clinical value from standardized effect sizes sadly, so when I see studies like this, my assumption is unfortunately that the researchers care only about publishing, and not about making their findings useful :(