1. Entering the Fried Liver Attack
2. The Polerio Defense
3. The Ponziani-Steinitz Gambit
4. The Anti-Fried Liver Defense
5. Parting Thoughts

Entering the Fried Liver Attack

Just as a quick refresher, we enter the Knight Attack variation of the Italian Game after we jump our knight forward to g5 to create a powerful double attack on the vulnerable f7 square.

The Polerio Defense

A common response to this double attack is to block our bishop’s vision of f7 by advancing the pawn to d5, and attacking our bishop at the same time.

We naturally capture this pawn with our pawn on e4, which removes the attack on our bishop and counterattacks our opponent’s knight.

The most common response to this counterattack is for black to capture our pawn with their other knight. However, there is another approach they could take. They could simply move their attacked knight to safety on a5, where it also attacks our undefended bishop. This is called the Polerio Defense.

The followup for white is quite simple, we will get out of the attack by using our bishop to give a check.

After this, your opponent may block the check with their bishop, their knight, or a pawn. At this point, we have safely left the opening up a pawn. From here on, by playing solid you can gain an advantage, but you will not have an overwhelming tactical advantage as you might in many of the other lines.

The Ponziani-Steinitz Gambit

This is an unconvential way that black might respond to our dual threat on the f7 square. They could instead take our pawn on e4 with the knight, entering the gambit.

This seems like a mistake at first, because we could simply capture the knight back. The problem is that we run straight into a fork when black pushes their d4 pawn.

Similarly, if we continue with our plan of attacking with the knight on f7, our opponent is able to simply move their queen to h4, and begin to threaten checkmate.

To avoid all of these complications, we should simply offer a check to the king on f7, by capturing with our bishop.

The Anti-Fried Liver Defense

The final variation actually begins before we can even move our knight to threaten the f7 pawn. This is called the Anti-Fried Liver Defense, so while not part of the Italian Game: Knight Attack, it is certainly related. Essentially, after entering the Italian Game our opponent pushes their pawn to h6 instead of moving their second knight out to f6.

Because our opponent has chosen to push a pawn rather than develop a piece, we can choose to attack the center by pushing our pawn to d4. After this point, much of the ideas of a standard Italian Game apply, so I would encourage you to explore further there.

Parting Thoughts

As you can see, even with a niche opening like the Fried Liver Attack, there are so many variations that can arise. It’s important to study all of the different variations, especially by practicing them in actual games. When a new variation arises, it’s important to analyze the best rebuttal and keep it mind for future games.

1. Get a dedicated machine with an NVIDIA GPU installed (or install one yourself) to allow your ML library of choice to use GPU acceleration. This involves setting up and updating drivers and libraries to make sure your machine is compatible with whatever model/code you are trying.
2. Use a cloud-based solution such as Colab, or Kaggle notebooks to do your experimentaion in the cloud.

Both have their advantages and disadvantages. While having your own dedicated GPU might seem cheaper in the long-run, you may find that your machine quickly becomes outdated, and now you’re just stuck with an expensive computer that does not do its primary purpose well. It’s a problem I personally experienced when I purchased my System76 Gazelle a few years ago. Now it just feels sluggish compared to newer GPUs especially when working with language models. Not to mention the noise that the fan makes when the GPU kicks in, and the annoying process of install the proper libraries with conda, and keeping the NVIDIA drivers up-to-date.

Cloud notebooks like Colab on the other hand remove the need to maintain drivers, hardware, and library installations. In addition, the free tier is usually generous enough for beginners to learn. However, you will almost certainly run into problems if you want to try training/fine-tuning language models. Given the fact that language models need to be able to predict the next word in a sequence of words, the matrices that they train are quite large, leading to long training times. You will soon discover with platforms like Colab that the notebook will disconnect long before your training is finished, but it’s not at all obvious from Googling what you can actually do to fix this scenario.

This is where Vertex AI notebooks from Google Cloud come in. Vertex AI lets you start up a compute instance with a GPU, and then run a Jupyter notebook connected to that instance. Because you have a dedicated instance, the notebook will just keep running. The notebook may still disconnect when your computer sleeps, because the websocket connection gets broken, but there are many solutions to simply keep your computer awake, such as the Caffeine/Caffeinated apps.

I would just start with the g2-standard-4 machine, which comes with one GPU. If that is not enough, then you can consider upgrading to the g2-standard-24 machine, which has two GPUs. For learning purposes, it is unlikely you will need more than that, but the g2-standard-48 and the a2-highgpu-* and a2-megagpu-* instances also exist. However, do keep in mind that the cost increases with the power and number of GPUs you utilize.

So, if you find yourself wondering why time is flying by, ask yourself whether you’ve tried challenging yourself lately.

For example, when you run the install command:

!pip install d2l==1.0.0-beta0

You will most likely encounter an error similar to this:

  error: subprocess-exited-with-error
  
  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> See above for output.
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
  Preparing metadata (setup.py) ... error
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

I found the solution for this in a Github issue for the setuptools library. To fix the issue, you need to revert your version of setuptools, as well as the version of wheel to match:

!pip install wheel==0.38.4 setuptools==57.1.0

Then, restart your Colab/Jupyter runtime in order to pick up the change. Afterwards, if you run the following command again, it should succeed:

!pip install d2l==1.0.0-beta0

1. Setup
2. Process Authorization header in Python
3. Pass the token in from the webapp

Setup

First, we install the firebase libraries using:

pip install firebase-admin
npm install firebase

Next, we need to setup the Firebase service account key that will allow our backend API to make calls to Firebase. To do that, first we set up a service account, and download the service account key JSON file by following the instructions in the Firebase documentation.

Then, you take the JSON file and place it somewhere on your computer and set the GOOGLE_APPLICATION_CREDENTIALS environment variable. Finally, you initialize Firebase in your Python app, by adding the following line somewhere in you application:

from firebase_admin import initialize_app

_FIREBASE_APP = initialize_app()

Process Authorization header in Python

When Firebase authenticates a user for you, it generates a user token that can be used to make calls to Firebase on the user’s behalf. So, to check that a user is authenticated on the API, we will pass the token from the webapp in the Authorization header of every request. To process the header in Python, we will define a get_current_user function:

from typing import Optional
from flask import request
from firebase_admin.auth import verify_id_token

def get_current_user() -> Optional[str]:
  # Return None if no Authorization header.
  if "Authorization" not in request.headers:
    return None
  authorization = request.headers["Authorization"]

  # Authorization header format is "Bearer <token>".
  # This matches OAuth 2.0 spec: 
  # https://www.rfc-editor.org/rfc/rfc6750.txt.
  if not authorization.startswith("Bearer "):
    return None

  token = authorization.split("Bearer ")[1]
  try:
    # Verify that the token is valid.
    result = verify_id_token(token)
    # Return the user ID of the authenticated user.
    return result["uid"]
  except:
    return None

Then, in any of the Flask endpoints that require authentication, we can simply check that get_current_user returns a valid user ID:

from flask import Flask, make_response

app = Flask(__name__)

@app.route("/api/resource")
def run_endpoint():
  user = get_current_user()
  if not user:
    return make_response("Unauthorized", 401)

  # do something

Pass the token in from the webapp

Now, we’ll discuss how to propagate the token from the React webapp to the API. When Firebase authenticates a user on your app, it will also issue an ID token. We will listen for this token, and maintain it in a Context so that we can always have an up-to-date token when we make the request to our Python backend.

import { initializeApp } from "firebase/app";
import { getAuth } from "firebase/auth";
import {
  ReactNode,
  createContext,
  useContext,
  useEffect,
  useState,
} from "react";

const firebaseApp = initializeApp({
  // Follow the instructions at
  // https://support.google.com/firebase/answer/7015592
  // to get your Firebase config object.
});

const TokenContext = createContext<string>("");
const TokenProvider(props: { children: ReactNode }) {
  const [idToken, setIdToken] = useState("");

  // Register a listener for the ID token.
  useEffect(() => {
    const auth = getAuth(firebaseApp);
    auth.onIdTokenChanged(async (user) => {
      setIdToken((await user?.getIdToken()) ?? "");
    });
  }, []);

  return (
    <TokenContext.Provider value={idToken}>
      {props.children}
    </TokenContext.Provider>
  )
}

function useToken() {
  return useContext(TokenContext);
}

The TokenProvider will be used to make the token available anywhere in your app. To do so, you just need to wrap your app component with the TokenProvider component, like so:

<TokenProvider>
  <App/>
</TokenProvider>

Then, in any component that will make HTTP requests, you can get the latest context like so:

function SomeComponent() {
  const token = useToken();

  return <div>component</div>;
}

To make HTTP requests with our token, we just pass it in via the Authorization header as we discussed earlier. If we use something like Axios, this is straightforward:

import axios from "axios";

function SomeComponent() {
  const token = useToken();

  const handleClick = async () => {
    const headers = { 
      Authorization: `Bearer ${token}` 
    };
    await axios.get("/api/resource", { headers });
  };

  return (
    <button onClick={handleClick}>
      Click me!
    </button>
  );
}

Warriors of the Western Ghats

As Thucydides tells us, the climate and terrain of a land informs us greatly about the nature of the people. Just as the Spartans’ military discipline, and self-sufficient nature derives from the Laconian hills and mountains, so too can we assert that the hardy and militaristic nature of the Marathas derived from the rough terrain of the Western Ghat mountains.

In 1647, when Shivaji first began his campaign to establish his own state, India was a divided land. Much of the north was controlled by the Mughal Empire ruled by Shah Jahan. In the middle of the country there were two main powers, both with Muslim rulers:

1. The Bijapur Sultanate in the west ruled by Muhammad Adil Shah.
2. The Golconda Sultanate in the east ruled by Abdullah Qutb Shah.

To the south, we see the remnants of the Vijayanagara Empire which began fracturing into much smaller kingdoms.

Shivaji would grow up in the Bijapur Sultanate in the west. The region was part of the Deccan plateau, a rocky terrain in between the Western Ghat and the Eastern Ghat mountains which line the respective coasts of the Indian peninsula. Shivaji’s father and his forefathers were all soldiers and officers in the Bijapur army. The rough terrain of the Deccan and the Western Ghats meant that upward mobility was easier to achieve through the military rather than agriculture or commerce, which led to a simple society devoid of refinement, but rich in equality.

So important was their equality that though Shivaji was considered by most Brahmins (save the Pandit Gaga Bhatt) to be just a Shudra, he still rose to be king. Similarly, the Brahmins who were generally considered to be priests, produced some of the greatest generals and leaders in Maratha history, like Peshwa Baji Rao I. This sense of merit earned by battle allowed the Marathas to elevate men from every social class to the title of Maharaja, including the Holkars of Indore, the Scindias of Gwalior, the Gaikwads of Baroda, and many more.

This rough and rugged populace allowed the Bijapur Sultanate to form an army that could fend off the powerful Mughals to the north. But, it was Shivaji, the last princeps amongst the Hindus who would weld these mercenaries together into a nation.

Shivaji’s luck

In addition to the robust nature of his people, Shivaji himself was born into circumstances that greatly aided his rise. First and foremost was the advantageous terrain of his native land. The Western Ghat mountains with their narrow paths and rugged terrain were excellent natural fortifications against any enemy forces. It allowed small groups of Maratha warriors to withstand much larger invading forces.

Moreover, Shivaji’s father, Shahaji, was a jagirdar (feudal lord) and military officer in the Bijapur sultanate. This afforded him an excellent military education. In addition, he also enjoyed a great deal of independence given that Shahaji left his young son and wife, Jijabai, to marry a much younger woman. He left the young Shivaji in Pune under the care of one of his Brahmin officers, Dadaji Kond-dev. Shivaji spent much of his childhood in the company of soldiers. becoming intimate with the hills, forests, and valleys of the Western Ghats. This knowledge would prove crucial to his future military exploits.

The departure of Shahaji also brought Jijabai and Shivaji closer together. She would be the one who imbued the future king with a deep passion for religion that drove his reverance for holy men of any faith, and thrust his image in the eyes of his subjects to mythic status. During the infancy of his realm, Shivaji was faced with the surging Bijapuri army led by the famous general, Afzal Khan. Local legend has it that when pondering whether to surrender, the goddess Bhavani came to him in a vision, and urged him to fight, guaranteeing his victory should he do so.

By the age of twenty, with Dadaji Kond-dev suffering an early death, Shivaji became independent with the land of Pune completely under his control. Without the old Brahmin to hold him in check, and Sultan Muhammad Adil Shah falling into a years-long illness, Shivaji had free reign to enact his plans for swaraj (freedom).

Shivaji the fox

According to Machiavelli, a ruler should be in equal parts “a fox to recognize snares” and equal parts “a lion to scare the wolves.” The unforgiving terrain of the Deccan, and the protection of Bhavani (herself astride a lion) imbued the young king with his leonine heart. But, it was his unique vulpine mind that would win him most of his early victories in decisive fashion.

Afzal Khan

One of the most memorable incidents was Shivaji’s confrontation with Afzal Khan. From 1647 to 1659, Shivaji diligently annexed land throughout the Konkan region, the coastal area between the Arabian Sea and the Western Ghats. Most important of his conquests was Javli, which gave him access to a highly disciplined Mavle infantry. On the other side, Afzal Khan was a seasoned general with an experienced army at his back. Neither side was interested in a head-to-head clash. To cajole Shivaji into making a mistake, Afzal Khan attacked the temple at Tuljapur and pulverized the idol of Bhavani, the family deity of the Bhonsle clan. However, Shivaji did not take the bait, instead choosing to lure his enemy closer.

Using a network of spies, Shivaji learned that Afzal Khan’s mission was to meet with him on friendly terms in order to capture or kill him. So, Shivaji lures the Bijapuri army through narrow mountain paths and dense jungle to have Afzal Khan meet him near the Pratapgad fort. To the meeting held outside in a luxurious tent, Shivaji came prepared wearing a steel cap under his turban, chain mail under his clothes, steel claws (bagh nakh) hidden in his left hand, and a dagger hidden in his right sleeve. As the two leaders embraced, Shivaji waited for Afzal Khan to strike first, his dagger glancing off the Maratha’s hidden armor. Then, still restrained in the tight embrace of the Khan, Shivaji reaches around and disembowels the much larger Bijapuri general using the bagh nakh hidden in his left hand.

With the Bijapuri general nearly dead, the Maratha army, which lay in waiting all along the mountain paths, laid waste to the now scattered Bijapur army. In just one day, Shivaji crippled the Bijapur Sultanate, and formally earned his swaraj. This marked the decisive point in history when the Marathas became the dominant force on the Western Deccan. In a show of his generosity, Shivaji freed captured officers and soldiers, allowing them to return home with food and other gifts. It would be this magnanimity that would win him allies throughout his life.

Shaista Khan

After the death of the Mughal Emperor Shah Jahan, Aurangzeb, his son through Mumtaz Mahal, was crowned emperor following a struggle for the throne with his older brother Dara Shikoh. Recognizing the threat of the young Maratha upstart, Aurganzeb appointed the capable general, Shaista Khan, as his viceroy of the Deccan. From 1661 - 1663, Shaista Khan harassed the Marathas, even capturing the future Maratha capital of Pune. But Shivaji, ever the master planner, was just waiting for the right moment to strike.

Shaista Khan and his army retired to Pune for Ramadan in April of 1663. Nearly the entire Mughal encampment was asleep after a giant feast following the day’s fast. Shivaji and a detachment of 400 soldiers snuck into the camp by pretending to be Mughal Deccani soldiers arriving to take up their posts. Having spent his entire childhood in Pune, Shivaji was able to lead half his soldiers discreetly to the Khan’s quarters. By midnight the Marathas had begun making a hole in the wall of the Khan’s harem, containing all his wives and female slaves. The other half of band went to create a distraction by having the Khan’s musicians begin to play as loudly as they could.

Soon, the women and the Khan roused from their slumber. In the ensuing skirmish, many of Shaista Khan’s guards were killed, including one of his sons, and a captain. The Mughal forces were thrown into disarray and forced to scatter as Shivaji’s troops stationed outside of the encampment fell on the drowsy and shocked Mughals. Shaista Khan would harass the Marathas no more as Aurangzeb transferred him to the Bengal as punishment for his humiliating defeat. The stunning and swift victory would elevate Shivaji to a hero in the eyes of his people, and a demon in the eyes of the Mughals. The Mughals took him “to be an incarnation of Satan; no place was believed to be proof against his entrance and no feat impossible for him.”

Where so many kings might face large armies head-on in an open field, Shivaji gladly played the part of the fox. He routed the Mughal viceroy in just one night and lost only six men. Few kings are so remarkable for how little blood their countrymen have shed to water the tree of their empire. Should anyone say that Shivaji lacked the heart of a lion, let them know that he personally led the raid on Pune that night, and injured Shaista Khan with his own blade, rather than hang back in his fort while his soldiers carried out the deed. As General Patton once said (or at least George C. Scott’s portrayal of him), “no bastard ever won a war by dying for his country. He won it, by making the other poor dumb bastard die for his country.” Patton famously believed in reincarnation, and that he was a soldier in all his past lives. Perhaps his sentiment was inspired by a former life as a Maratha soldier from the Western Ghats.

Shivaji’s early demise and his legacy

Shivaji would not become coronated as Chhatrapati (meaning the umbrella lord, protector of his people) until 1674 under the auspices of Gaga Bhatt. It was Shivaji’s secretary, Balaji Avji, who would engage Gaga Bhatt of Varanasi to produce evidence of Shivaji’s kingly lineage and consecrate his throne. The learned scholar, considered the Vyas of his age, delivered on his task and declared Shivaji to be descended from the Maharanas of Udaipur in Rajasthan, part of the solar dynasty of Ramachandra. With his coronation Shivaji hoped to legitamize his realm, bring himself to equal status with the rulers of Dehli, Bijapur, and Golconda, and draw those in the upper echelons of Maharashtrian society to his side. He desired to transition from a marauding warlord to a monarch.

From 1647 to 1674, Shivaji annexed territories, formed a powerful army, built a large navy, and plundered the cities of his enemies. But, many a historian has wondered whether Shivaji could bring a realm to prosperity during peactime, or whether he had created a horde of pillagers and raiders like the Mongolic bands of Timur and Hulagu? The speculation raged, due to Shivaji’s untimely death in 1680 at just 50 years of age. He was widely suspected to be poisoned by his queen-consort, Soyra Bai, who had hoped to make her son Raja Ram the next Chhatrapati, as opposed to Shivaji’s older son by his first wife, Sambhaji. Nevertheless, the crown would pass to Sambhaji, who put Soyra Bai to death either for her part in his father’s death or at her attempt to usurp the throne. Ironically, Raja Ram would still become Chhatrapati when Sambhaji met his own early demise.

Unfortunately, we can never know how Shivaji, the monarch, would realize his vision for swaraj, because for the entirety of his short life he would be embroiled in war. However, it is the fire that he lit in the hearts of every Maratha soldier that would one day allow his grandson, Chhatrapati Shahu to grow the Maratha Empire from the mouth of the Kaveri river to the sands of Peshawar.

What his life means for us today

In the age of democracy, one wonders what, if anything, can be learned from the life of Shivaji. He was a man who grew up in a completely different time. He was illiterate, his livelihood was killing, and he had eight wives. However, there is in fact much that we must appreciate about his life even in the 21st century, but not in the way that corporate middle managers think they can learn something from Sun Tzu’s Art of War. Today, we must cope with the fact that world leaders do not understand their people. In fact, most even show open disdain for their own people, and gleefully plunder the public purse. Furthermore, we must cope with the fact that they are not brave. They send the sons of the poorest among their people to die for a war where there is no plunder, where there is no end, and where there is no dignity upon return. And lastly, we must cope with their imperfect morals. We must give a wide berth to their personal failings in hopes that they may at least be competent.

This was not so with Chhatrapati Shivaji. His care for the common folk and high-born alike is comparable to the Daanveer Karna. On his trip to Hyderabad in 1677 to form an alliance with the Golconda Sultan Abul Hassan Qutb Shah, Shivaji strictly forbade his soldiers from robbing or harassing any inhabitants of the realm, on penalty of death. For this march through the city of Hyderabad, he distributed amongst his officers, strings of pearls, gold jewelry, and bright armor. For the people of the city, Shivaji kept ample gold and silver on hand to distribute to the masses, reciprocating the blessings of every household that came to see the most feared man in the Deccan. And, his words were just as honeyed and sweet as his gifts. His recounting of the Maratha conquests across the country mesmerized the sedentary Sultan, who quickly accepted alliance.

Moreover, the evidence of the morality and godliness of Shivaji’s conduct is as long as the evidence of military prowess. To honor his spiritual guru, Sant Ramdas, Shivaji placed the deed to the Maratha realm at his feet. Pleased with the gift, Ramdas accepted, and “appointed Shivaji as his vicar, and bade him rule the realm thenceforth not as an autocratic owner, but as a servant responsible for all his acts to a higher authority.” From that moment forward, the Maratha standard was a plain saffron penant fashioned after the sannyasi’s cloth, his bhagwe jhanda. A symbol of renunciation, and servitude to God, it became an appropriate standard for the simple but reverant Maratha soldier.

So, I say again, what can we learn from the life of Chhatrapati Shivaji? We can learn that victory in battle need not drench the soil with the blood of your own people. We can learn that it is not too much to expect a leader to be as beyond reproach in his private life as he is in his public life. We can learn that a king who bows his head to the needs of his people, finds that his people bow their heads even deeper in reverance. And most importantly, we can learn that far from being a barbaric way of choosing a worthy ruler, battle is the meritorious way of choosing a worthy ruler. The Marathas became successful on their merit, a merit measured by their performance on the battlefield. Can we say that democracy has chosen leaders in a more meritorious way? I think not.

Sources

My main source is Jadunath Sarkar’s Shivaji and His Times.

1. The Upanishads
   1. Brihadaranyaka Upanishad
   2. Kaushitaki Upanishad
2. The Mahabharata
3. When does life begin?
4. Incidence of abortion in Hindu nations
5. Parting thoughts
6. Bibliography

The Upanishads

First, let’s start by looking at the Upanishads, which expound many of the core principles of the philosophy underlying Hinduism.

Brihadaranyaka Upanishad

In the chapter 4, section 3 of the Brihadaranyaka Upanishad, the oldest of the Upanishads, Janaka, the king of Videha, is in discussion with the sage, Yajnavalkya, about the atman, the self or the soul. When describing the state of shedding all desires and only knowing the atman, Yajnavalkya says:

Here a father is not a father, a mother is not a mother, worlds are not worlds, gods are not gods, and Vedas are not Vedas. Here a thief is not a thief, an abortionist is not an abortionist, an outcaste is not an outcaste, a pariah is not a pariah, a recluse is not a recluse, and an ascetic is not an ascetic. Neither the good nor the bad follow him, for he has now passed beyond all sorrows of the heart.

Yajnavalkya describes that in this state, you are free from both the good and bad of life. He begins by listing good things, such as fathers, mothers, gods, etc. But then, he lists things that are bad, one of which is an abortionist. Clearly, Yajnavalkya has a low opinion of the abortionist.

Kaushitaki Upanishad

Next, we turn our attention to the third chapter of the Kaushitaki Upanishad. Here, Indra asks Pratardana, a king, to request a boon. When Pratardana refuses a gift, Indra instead asks Pratardana to simply perceive Indra:

When a man perceives me, nothing that he does—whether it is stealing, or performing an abortion, or killing his own father or mother—will ever make him lose a single hair on his body. And when he has committed a sin, his face does not lose its colour.

Here Indra, similar to Yajnavalkya, lists abortion among things that are obviously bad, such as theft and patricide. He shows the extent of his power, by showing he is able to wipe away sins of any degree, and Indra certainly believes that abortion is among those sins.

The Mahabharata

One of the most important stories in the Mahabharata has to do with Ashwatthama attempting to kill Parikshit, a fetus in the womb of Uttara. Ashwatthama, along with Kripa and Kritavarma, attacks the Pandava camp and kills most of the warriors on the Pandava side in their sleep. With all the sons of the Pandavas murdered, the only viable heir to the Kuru lineage remaining was the Parikshit, still just a fetus in the womb of Abhimanyu’s widow, Uttara. To ensure that the Pandavas have no heirs remaining, Ashwatthama directs a bhramastra at her womb, declaring:

this weapon of mine shall fall on the foetus that is in the womb of Virata’s daughter, upon that foetus which thou, O Krishna, art desirous of protecting.

In response to Ashwatthama’s declaration, Krishna declares that he will not succeed:

The fall of this mighty weapon will not be fruitless. The foetus will die. But being dead, it will live again and have a long life! As regards thyself, all wise know thee for a coward and a sinful wretch! Always engaged in sinful acts, thou art the slayer of children. For this reason, thou must have to bear the fruit of these thy sins.

Ashwatthama is condemned in the harshest terms for his act. He is not condemned by an ordinary man. He is not even condemned by a sage, or king. He is condemned by Krishna, who revealed to Arjuna on the fields of Kurukshetra that he is the incarnation of Vishnu, the lord of the universe himself. As punishment for his act, Krishna afflicts him with the most ghastly curse:

For 3,000 years thou shalt wander over this earth, without a companion and without being able to talk with anyone. Alone and without anybody by thy side, thou shalt wander through diverse countries, O wretch, thou shalt have no place in the midst of men. The stench of pus and blood shall emanate from thee, and inaccessible forests and dreary moors shall by thy abode! Thou shalt wander over the Earth, O thou sinful soul, with the weight of all diseases on thee.

Much like Indra, it is safe to say that Krishna, through his punishment of Ashwatthama, condemns in no uncertain terms the act of abortion.

When does life begin?

The question of when life begins is a complicated one as well. Many who consider themselves pro-life say that the only consistent position is that life begins at conception. There is actually some Upanshadic literature that comments on this topic. In one of the minor Upanishads, the Garbha Upanishad, literally the Upanishad of the womb, we see a description of the development of the fetus. The Upanishad tells us that the fetus becomes united with the atma in the seventh month:

… it (the embryo formed in the womb) is like water in the first night; in seven nights, it is like a bubble; at the end of half a month, it becomes a ball. At the end of a month, it is hardened; in two months, the head is formed; in three months, the region about the feet; and in the fourth month, the region about the stomach and the loins and also ankle is formed; in the fifth month, the back (or spinal) bone; in the sixth, the face of the nose, eyes, and ears; in the seventh, it becomes united with Jiva (Atma); in the eighth month, it becomes full (of all organs); in the ninth, it becomes fatty

Does a fetus without an atma constitute a human life? Does the absence of the atma in the body mean that abortion before the seventh month is fine? To help answer this question, I attempted to determine at which month of the pregnancy Ashwatthama attempted to kill Parikshit. However, the best we could do is say that Parikshit was no older than seven months. There is no direct evidence of this, but we could infer it from the fact that Abhimanyu and Uttara were married for a mere seven months. After his death, Uttara lamenting the loss of Abhimanyu tells us:

Thy union with me in this world had, it seems, been ordained for only six months, for in the seventh, O hero, thou hast been bereft of life!

There was no definitive guidance that I was able to find on this matter during my search. However, I will say that even though the atma unites with the fetus in the seventh month, there is still no definitive time during the seventh month that this happens.

Incidence of abortion in Hindu nations

So, why is this issue of importance. Why should we care? Despite the injunctions from classical Hindu literature, when we look at the two largest Hindu-majority nations in the world, India and Nepal, we see that abortion is rampant. India has 1.38 billion people, out of whom about 80% are Hindus. Yet, according to an article in The Lancet, it is estimated that 15.6 million abortions were performed in the year 2015 based on data from six Indian states. This meant one in three pregnancies in India ended in abortion.

Nepal has 26 million people, out of whom 81% are Hindus. Yet, according to an article published with the NIH, in 2014, Nepalese women had 323,100 abortions with 56 abortions per 100 live births. In other words, just like in India, approximately one in three pregnancies that year ended in abortion.

These are staggering numbers. So, how can two nations with about 1.1 billion Hindus between them do this, despite the words of Yajnavalkya, Indra, and Krishna?

Parting thoughts

This evidence is presented to the reader to inform them about the true prevalence of abortion among Hindus, as well as relevant theological material on the matter. From there, we can make a conscious decision about how our actions affect our karma. Whether you think abortion should legal or illegal, it is without a doubt that we should at least find a way to decrease the need for all of these abortions.

I also understand that this analysis is not perfect, because I have read these works in translation. Translators are often not perfect, but learning Sanskrit and reading the original source just means that now I become the imperfect translator. I might try reading the original Sanskrit in the future, but for now I trust that the translators, Patrick Olivelle and Kisari Mohan Ganguli, have done an honest job.

Bibliography

• Olivelle, Patrick. Upanisads. Oxford University Press, 2008.
• Ganguli, Kisari Mohan, translator. The Mahabharata of Krishna-Dwaipayana Vyasa, Volume 3. Pratap Chandra Roy, Sacred Texts, sacred-texts.com, Accessed 2021.

When developing an Electron app, you will often want to run expensive or long-running operations in a different process. In the past, this might have been done using the child_process module. But, Electron has recently introduced a new utilityProcess that uses the Chromium Services API. This article outlines how to compile the code for your utilityProcess using your existing Webpack config.

1. The problem: how to compile the utilityProcess code with Webpack?
2. Electron Forge’s Webpack config
3. The solution: write our own Electron Forge plugin
4. Appendix: getting a first-party plugin

The problem: how to compile the utilityProcess code with Webpack?

To initialize the utilityProcess, you must provide a node module path, like so:

import { utilityProcess } from "electron";

this.child = utilityProcess
  .fork("./utilityCode")
  .on("spawn", () => console.log("spawned new utilityProcess"))
  .on("exit", (code) => console.log("existing utilityProcess"));

This code assumes that you have a utilityCode.js file that is a Node module. But, this doesn’t work so well if you’re building all of your other code with Webpack. You might be using Typescript, and integrating other compilation steps. Naturally, you would want all of your utilityProcess code to also run through Webpack first.

Electron Forge’s Webpack config

When you initialize an Electron project with Electron Forge, it initializes a Webpack config for you. This configuration consists of the following files:

• webpack.rules.config.ts: which Webpack loaders to use for this project.
• webpack.main.config.ts: the Webpack config for the main process.
• webpack.renderer.config.ts: the Webpack config for the renderer process.
• webpack.plugins.ts: contains plugins for the renderer process.
• forge.config.ts: passes the main and renderer process configs to Forge.

The WebpackPlugin is used by Electron Forge to run Webpack for the code of each process, but there’s no API for adding a utilityProcess.

The solution: write our own Electron Forge plugin

First, we create a new plugin called UtilityProcessPlugin in a separate file called utility.plugin.ts:

import { webpack, Configuration as RawWebpackConfiguration } from "webpack";
import { PluginBase } from "@electron-forge/plugin-base";
import { ForgeHookMap } from "@electron-forge/shared-types";

export default class UtilityProcessPlugin extends PluginBase<RawWebpackConfiguration> {
  name = "utility-process";

  getHooks(): ForgeHookMap {
    return {
      generateAssets: this.generateAssets.bind(this),
    };
  }

  private async generateAssets(): Promise<void> {
    return new Promise((resolve, reject) => {
      webpack(this.config, (err, stats) => {
        if (err) {
          reject(err);
          return;
        }
        if (stats?.hasErrors()) {
          const json = stats.toJson();
          for (const error of json.errors ?? []) {
            reject(new Error(`${error.message}\n${error.stack}`));
            return;
          }
        }
        resolve();
      });
    });
  }
}

This plugin takes a Webpack config and runs the Webpack compiler on it. Then, we create a Webpack configuration for the utilityProcess:

import { Configuration } from "webpack";
import path from "path";

import { rules } from "./webpack.rules";

const modulePath = path.resolve(__dirname, "dist_utility/utility_process");

export const utilityConfig: Configuration = {
  entry: "./src/utility.ts", // Change to your own entry point
  target: "node",
  module: {
    rules,
  },
  output: {
    path: modulePath,
    filename: "index.js",
  },
  resolve: {
    extensions: [".js", ".ts", ".jsx", ".tsx", ".css", ".json"],
  },
  // TODO: find a way to infer this based on whether we run electron-forge start
  // or package.
  mode: "development",
};

This configuration will look for a src/utility.ts file as the entrypoint. Feel free to change this to whatever file is your entrypoint. Then, it will compile the code and output it to a dist_utility/utility_process directory. Then, we need to tell Electron Forge about our plugin, by modifying forge.config.ts:

// ... other imports
import { utilityConfig } from "./webpack.utility.config";
import UtilityProcessPlugin from "./utility.plugin";

const config: ForgeConfig = {
  // ... other configuration
  plugins: [
    new UtilityProcessPlugin(utilityConfig),
    // ... other plugins
  ],
};

export default config;

Next, to make sure our main process knows whether this code has been compiled to, we use the Webpack DefinePlugin to define a global variable in the main process that contains the location of the compiled utilityProcess code. In the webpack.main.config.ts file, we add:

import { Configuration, DefinePlugin } from "webpack";
// ... other imports

const modulePath = path.resolve(__dirname, "dist_utility/utility_process");

export const mainConfig: Configuration = {
  // ... other configuration
  plugins: [
    new DefinePlugin({
      UTILITY_PROCESS_MODULE_PATH: JSON.stringify(modulePath),
    }),
  ],
};

Finally, we can modify our code in the main process to load our compiled utilityProcess:

import { utilityProcess } from "electron";

declare const UTILITY_PROCESS_MODULE_PATH: string;

this.child = utilityProcess
  .fork(UTILITY_PROCESS_MODULE_PATH)
  .on("spawn", () => console.log("spawned new utilityProcess"))
  .on("exit", (code) => console.log("existing utilityProcess"));

Appendix: getting a first-party plugin

I have opened a feature request with the Electron Forge team to help push this proposal into the WebpackPlugin. Please see https://github.com/electron/forge/issues/3169 for progress.

import { shell } from "electron";

mainWindow.webContents.setWindowOpenHandler((details) => {
  shell.openExternal(details.url); // Open URL in user's browser.
  return { action: "deny" }; // Prevent the app from opening the URL.
})

This handler prevents all window open requests from changing the URL inside your app, and instead causes the URL to open in the user’s browser. However, this only works if your links open new windows/tabs. You can have anchor tags with target="_blank", or call window.open with a _blank target:

<!-- Allow the user to click and open a link. -->
<a href="http://example.com/abc" target="_blank">Some link</a>

// Open a URL programmatically.
window.open("http://example.com/abc", "_blank");

Google as a forum search

Once we dig a little deeper, it makes sense why people are doing this. There are essentially three categories of search queries.

• Facts: examples are “temperature outside,” “NFL scores,” “height of Michael Jordan,” etc. In other words, queries that have a clear 1:1 query and answer. Google is amazing at these.

• Objective information: examples are “George Washington,” “Nelson Mandela,” “The Seven Years War,” etc. These are queries that have objective information attached to them. Usually for these types of queries the Wikipedia link is a good starting point, and the user might have to jump to the external links inside to get some in depth information. In many cases, Google will also present other information about these queries in the other search results, but these are “low trust” results, or worse versions of Wikipedia such as Britannica. They have a lot of ads, they are possibly unmaintained, and they have questionable veracity.

• Subjective information: examples are “What’s the best programming language,” “how to get into Harvard,” etc. These are queries which are handled even worse by Google. And, these are the queries that people often append “reddit” for. They do this, because that will lead to a place where the topic is discussed, multiple inputs are provided, and there is feedback on those inputs so that you can judge all of the info before making a decision.

Ultimately, if there is no authoritative source on a matter, people want a democratic answer, hence the desire to search on a forum. If there’s no factual result, they want to see a discussion involving real people so they can weigh the many sides of an argument themselves. Google is heavily optimized for the first category of queries, and partly optimized for the second category of queries. The third set usually surfaces untrustworthy results, such as random blog posts, or SEO spam sites.

If Reddit, StackOverflow, and other forums have the information that people are really looking for, why do people still use Google? Unlike Google, all of these disparate forum sites have lackluster search. Google unsurprisingly has best-in-class search.

Democratized information

Implicitly, people are looking for expert opinions on topics. They want to solicit the thoughts of someone who has already looked into it, and let other people give their thoughts. Of course they could pose their own question on a forum, but that would take time for the question to be answered (if it is even popular). Forum search also accounts for the long tail of information. Information that is sufficiently niche that Google cannot serve anything useful or trustworthy. For example, if you search for “Guliga daiva” after watching Kantara (extremely niche subject), you get a bunch of surface level news articles and blog posts. There is nothing authoritative or democratic.

Recently, there have been great strides in large language models. ChatGPT has gone viral. If what people really want is an expert opinion on topics, why can’t we make a ChatGPT for all of the “experts” we trust? For example, if you want to know what Mahatma Gandhi would think about the “antifa” riots, we could fine tune a ChatGPT style LLM on Gandhi’s writings and sayings, and generate something plausible.

However, the main issue that we currently see with ChatGPT and others is that they struggle with surfacing actual facts. For example, if you have never read William Faulkner’s Absalom, Absalom! then you would not be able to tell that while ChatGPT’s synopsis of the novel sounds plausible, it’s actually inaccurate. Thomas Sutpen does have a son named Henry, but Quentin is not his grandson. He is the narrator of the story and belongs to the Compson family.

Explainable AIs

Solving the accuracy problem is the main issue that prevents us from using LLMs for information retrieval. Some models such as GopherCite, WebGPT, and LaMDA attempt to do this by essentially composing the answers by augmenting the response with search engine results. However, if people do not trust many search results for most opinionated topics, then using search results to provide post facto justification for the responses would not give users any confidence.

Instead what we really want is to have an explainable AI that can cite how it arrived an answer using the resources that were used to train it. This would hit several birds with one stone. It would give users confidence that the training was used with data that did not violate anyone’s privacy, that the data is from trustworthy sources, and that the response was formed sensibly. Obviously this is an extremely hard problem to solve, but it’s absolutely essential for us to be able to use LLMs widely.

As A Survey of the State of Explainable AI for Natural Language Processing (arXiv:2010.00711) explains, we are looking for a global self-explaining model. Many explainable language models are local, meaning they explain what parts of your input were most important. What we care about is global explainability, which means we want an explanation of the output regardless of the input. We want an explanation based on what the model actually learned.

One interesting approach is REALM, which trains a knowledge retrieval model that sits in front of a masked language model. The knowledge retrieval model provides documents that can be used to fill in the masked tokens in the language model’s response. Because the knowledge retrieval model is essentially a function that given a query and a document generates a relevance score, we can use the model to interpret which documents were used to inform the response you received. Breaking down the documents into paragraphs or sections could also help us rate the relevance of each part of a document.

Parting thoughts

No matter how well an LLM can formulate answers, or how well it’s able to explain its answers, many humans will always prefer to see the thoughts and opinions of real people. But, LLMs can help you summarize long threads, and cite the most important comments in a thread.

There is also a long tail of information buried in sources that are extremely hard to surface through regular Google search results. Often, if you have an obscure topic, you can try looking up the information in Google Books. For example, we could try looking up Guliga daiva:

These searches will often surface some high quality sources, but you might have to do quite a bit of reading to get the information you’re looking for. The same is true for scientific topics where relevant answers are only in dense research papers. An LLM that explains itself can be used to present the user with the paragraphs in books, papers, and long forum threads that are most relevant, along with an easy to understand summarization. In short, LLMs will not replace search, but will help to create a more useful interface for search.