Vladyslav Siriniok

Making Instinctive Software Estimates

Sat, 31 Dec 2022 21:01:00 +0000

I want to give you some insight in how I do my estimates. But it’s a pretty substantial topic, and it would take a book to explain it properly. So please excuse me if my explanations won’t sound sufficient sometimes, I will be happy to discuss those moments later. :)

My estimates are represented by two variables—duration and a likelihood of success.

In my estimation process I’m guided by three mathematical heuristics:

Fermi estimates: I only do order-of-magnitude estimates to cancel errors in single estimates. So when I say “we need 8 hours” I actually mean “we need a day or two” and when I say “we need 1 hour” I actually mean “we need 10 minutes or a day”.
Paretto principle (80-20): I assume that distribution of complexity of our tasks follows the power law, therefore the least complex 20% of our tasks will take 80% of our time.
Instinctive Bayesian updating: We can apply the Bayes’ theorem to calibrate a probability of completing a project in a given timespan. But that’s complicated and relies on our intuition anyway. So let’s just use an alternative approach proposed by Douglas Hubbard:
1. Start with your calibrated estimate.
2. Gather additional information (polling, reading other studies, etc.)
3. Update your calibrated estimate subjectively, without doing any additional math.

It all may sound scary, but it’s really just a formalization of the intuitive heuristics our mind uses for planning daily without even our conscious awareness. Let’s see how these principles apply to our case.

First, we need to plan our project, asses the size of the scope we are dealing with, and estimate the magnitude of how long will it take us to get it done. We can get those numbers from our experience with similar projects in the past and add a 20% buffer to appease the uncertainty.

That’s a nice number, but we still don’t know how likely it is that we guessed it right. Estimation is essentially betting, so we need to know the odds. We also know that software engineers tend to underestimate how long they will work on a project.

What can we do about that? Let’s start with some outrageously large number to overcorrect our bias in favor of shipping the project fast. How long will it take us to ship this project with 99% certainty? What a pessimistic manager would say? 1 year? Ughh, that’s a bit too much. 1 month? Sounds a bit too low. 3 months? Huh, easy. So be it, that’s our prior guess.

3 months have 160 * 3 = 480 working hours, so our estimate of 226 hours gives us only 228 / 480 = 0.48 probability of success then. We could have just flipped a coin!

So, the probability of finishing the full project in less than 228 hours is P(A) = 0.48.

Our second data point we will obtain from the real world, more specifically from building our PoC. Building a mini-version of our project gives us some understanding of the problem space and helps us to set more correct expectations for the rest of the project. As it “took” me 10 hours to build rather than expected 12h, we can calculate a probability of that happening: P(B) = 10 / 12 = 0.83.

Let’s calibrate our first prediction using this new data: P(A) = 228 / (480 * 0.83) = 0.57. Wow, it’s not that bad now.

But I actually want our project to have at least 80% chance of success. That means that we need to calibrate it to (480 * 0.83) * 0.8 = 319 hours. It’s almost 100 hours more than our initial estimate, but also more than 100 hours less than the worst case scenario! And now we can be much more confident that we will ship the project in time.

Is 80% chance good enough? Well, we can pick a higher number, but it’s almost twice as likely than our initial estimate already, so it’s fine by me. The goal of these calculations is to counter the planning fallacy rather than to predict the completion date.

We can repeat this process after every iteration to calibrate our estimates even better.

What Really Guides Me? Part I

Fri, 31 Dec 2021 20:12:00 +0000

Facing lots of changes coming into my life, I had to once again ask myself “What really guides me?“. Psychology and common wisdom say that our values guide us, so that’s where I started to look for the answers.

To get more or less objective picture, this time I decided to begin with some science-backed self-assessments. There are a lot of personal values tests you can find on the Internet, but as far as I know, there is no a de-facto standard, like Big Five personality traits assessment. Many of them represent the bleeding edge ongoing research, but most of them are likely just a junk. So I’ve tried a bunch of tests to evaluate their results on my own.

If you are wondering if there is a rigorous scientific definition for the concept of “value“—yes, there are a few. Here is the one I like, it’s used in acceptance and commitment therapy:

Values are “freely chosen, verbally constructed consequences of ongoing, dynamic, evolving patterns of activity, which establish predominant reinforcers for that activity that are intrinsic in engagement in the valued behavioral pattern itself” (Wilson & Dufrene, 2009).

So basically values are arbitrary self-reinforcing effects of our behavior.

I had only three criteria for evaluating the results of the tests, sorted in the order of importance:

Validity: Is there any evidence for the scientific validity of this test?
Resemblance: Are the results of this test are similar to the rest? If no, why?
My feelings: How do I feel about the results?

So here is following my review for every each of them.

You can skip to the Conclusions section for a TL;DR version.

1) Personal Values Test

[Link]

Validity: Unknown

Resemblance: High

My feelings: “Yeah, that’s me“

The first one is one of the top Google search results, so it’s probably quite popular on the Internet, and that’s why I’d decided to try it. I have no idea about its validity or even methodology, the site doesn’t disclose anything. However, it gave me a highly-relatable list of values, similar to the rest of the tests, so I’ve included this test on the list.

My most important core values according to this test are wisdom, intelligence, meaning in life, freedom and health.

2) The Core Values Index

[Link]

Validity: Doubtful

Resemblance: High

My feelings: “Yeah, that’s me, but WTF“

This one is rather a personality test, as the authors claim “the most reliable assessment ever created“ because of “a 94% repeat score reliability rate“. However, the link to the paper is broken, and I couldn’t find a copy.

According to this test, my primary core value is “Innovator“ and the secondary one is “Merchant“. And their “core value energies” are wisdom and love respectively. Ughh, yeah, I know how New-Agey it sounds. Let’s move to the next one.

3) The Portrait Values Questionnaire (PVQ)

[Link]

4) Find Your Values Test

[Link]

5) Life Values Inventory

[Link]

Validity: All seem to be based on the PVQ test which is backed by some studies.

Resemblance: High

My feelings: “Yeah, sounds like me“

These three seem to be variations of the same PVQ test. They all gave me pretty similar results and insight. The 4th and 5th tests also provide an extended analysis and exercises.

According to these tests, my core values are achievement, health, benevolence, intelligence, prosperity, freedom, and security.

6) Personal Values Assessment from Barrett Values Centre

[Link]

Validity: It has some validity according to this article, but I was unable to verify this claim.

Resemblance: High

My feelings: “Yeah, that’s me“

The Barrett Model is about mapping your values on something pretty much like Abraham Maslow’s Hierarchy of Needs. Quite an interesting exercise, but I’m not sure if it’s backed by any science, given that the Hierarchy itself is often questioned nowadays. It also gives an 8-page long PDF with some insights and practical exercises.

According to this test, my core values are wisdom, ethics, making a difference, personal fulfillment, and well-being.

Conclusions

Despite all the tests have been very different, I’ve got pretty similar results on all of them. This gives me some confidence both about the validity of these tests and my personal integrity and consistency. I’d say these tests gave me a pretty abstract picture of the areas I should explore to identify more concrete values. I do believe it’s hard to act on vague images and to make my brain focus on my values I need to turn them into something more specific and actionable.

Also, I’m honestly wondering, aren’t these results too good? None of the tests asked me whether I like killing little kittens. I mean, I definitely do not, but none of the tests included any “dark“ values and I don’t understand why do psychologists have such high faith in humanity. Do these tests give an “honest” picture or just a “feel-good” one? That’s worth exploring as well.

Another objections against all these tests is that all their dimensions feel rather arbitrary and many of them do not look like fundamental values worth to be a core of one’s identity.

And finally, here is my free interpretation and summary for what my top-5 core values are based on all the tests above:

intelligence & wisdom
freedom & independence
health & well-being
prosperity
security

In the next stage, I want to make my values more specific and derive my core principles from them. So please stay tuned :).

Can Technology Make Discussions More Fruitful?

Fri, 29 Nov 2019 08:11:00 +0000

Throughout the history of our civilization, the speed of long-distance communications tends to raise. It started with messenger horses and pigeon post, and likely ended with video calls and chat messengers¹. I’d say conservatively, no one needs faster communications than we have today, and even if you can imagine something more rapid, like quantum telepathic networks organized through a brain-computer interface, you will probably agree that our communication technologies are fast enough for our daily needs.

So nowadays we have everything in terms of speed and access. And it looks fine: what could possibly be wrong about the ability to communicate fast and freely, right? Well, yes, but only until it comes to group discussions. Quick pace and hurry aren’t the best friends of a good argument.

You had probably noticed some frustration related to the speed of discussion yourself: discussions on Twitter are grueling, discussions in group chats are grueling even more. I bet, if you are old enough, you experience some nostalgia for sweet discussions on forums or mailing lists from time to time. And have you noticed, it’s easier to discuss a challenging topic with friends at the bar than to discuss that topic with the same friends in your group chat?

For me, the most frustrating aspect of chat conversations is their intensity. It never was appropriate for a good discussion to have more than one person speaking at the same time. It’s still not appropriate in real life or on Skype calls. But when it comes to texting… well, no one can be stopped. Taking a few deep breaths probably could help, but alas, you can’t force all participants to do that.

Even when this issue arises in a voice discussion, you can manage it by raising your voice and asking everyone to shut the fuck up, but I don’t remember a case when STFU ever worked on the Internet.

Do certain technologies make discussions more or less productive? Maybe, you have wondered about that too after having a hot, but unproductive and exhausting, discussion on Twitter or some messenger. My guess, it mainly depends on the average message length and average message frequency on the specific media.

And it seems like I’m not the only one who came to similar conclusions, I see people frustrated by this issue quite often.

Instant messaging lowers the bar to compose a message. I think this is not a good idea for collaboration. Composing a message should require more time and thought.
— Þotacon (@obsgog) March 22, 2019

Unfortunately, I haven’t found much research on this topic, but from my subjective experience, I’d sort the popular text communication means by their discussion quality in this order:

blogs (as article responses, not comments)
emails (mailing lists)
forums and Q&A sites
Facebook
Twitter
chats

So it looks like the quality of the discussion correlates with the average message length and average message frequency on the respective media. Maybe, a technology that artificially limits these two variables would help to have more thoughtful and fruitful discussions?

One potential solution that came to my attention is the so-called “slow media” platforms like Letter Wiki as a modern alternative to mailing lists, or Threads, with a similar concept for B2B. They are trying to create a space for meaningful discussions and better decision making by encouraging the email-like style of communication while providing a better UX.

Existing platforms are also trying to mitigate this problem. For example, Telegram introduced “slow mode” for group chats: you can throttle posting frequency down to a specific number of messages per hour.

There is a broad space of possibilities to solve this problem, and a technology that will solve it has the potential to become a new level of social networking. As Naval Ravikant once put it: “The people designing the algorithms to shape online debate are now the most powerful people in the world”.

So how do you see the future of online debate? Can technology make online discussions more fruitful and less exhausting? Please, leave your thoughts in the comments below.

UPD (2020-11-14): In the last year, we could observe a raise of slow media. Letter Wiki is gaining popularity, Threads got a competitor from Clariti, and Substack gave a new life to email as a medium. Even Google entered the market with their own solution for meaningful discussions—Currents. So I’m really glad that I have spotted the trend early.

Amazing how it all started up with messengers and ended up with messengers too. ↩

My MIT Challenge

Sat, 01 Sep 2018 18:00:00 +0000

Hi, everyone!

I’m Vlad Siriniok, a software engineer and entrepreneur from Kharkiv, Ukraine.

Like many other guys in tech I dropped out of my university after a second year of study. And I have no regrets about that—tech universities in Ukraine is not the best place to actually learn something. But honestly, I’ve always dreamed about receiving a degree from an American university, like MIT or Berkeley.

Fortunately, we’re all living in America. I mean, today, in the era of the Internet, anyone has an access to a world-class education. Massive online courses, online degrees, whatever.

So since I’ve learned about MIT OCW I lived with a plan to complete MIT Undergraduate program in Computer Science using only MIT OCW materials. Later, I found an article written by Scott H. Young about his experience in studying through the entire 4-year MIT computer science curriculum, without taking any classes. He finally motivated and convinced me in attainability of such crazy goal.

And today I begin ~~a new life~~ (finally)! Over the next few years, I’m going to study through the entire 4-year MIT curriculum for computer science with help of materials available on MIT OCW.

My Goals

Obtain fluency in math.
Improve my knowledge about software engineering and design.
Prepare myself to the deep dive into studying AI.
Improve my ability to read CS papers.
Develop an ability to write good essays and articles on computer science and tech topics in English.

6-3. Computer Science and Engineering Curriculum

I’ve used the latest MIT curriculum. Also, I used this catalog to get pre- and co-requirements for the courses.

General Institute Requirements (GIRs)

6 subjects for Science Requirement

3.091 Introduction to Solid-State Chemistry

Prereq: None, 12 Units

8.01L Physics I: Classical Mechanics

Prereq: None, 12 Units

Also, I want to read The Feynman Lectures on Physics.

8.02 Physics II: Electricity and Magnetism

Prereq: Physics I (GIR), Calculus I (GIR), 12 Units

18.01SC Single Variable Calculus

Prereq: None, 12 Units

18.02SC Multivariable Calculus

Prereq: Calculus I (GIR), 12 Units

7.013 Introductory Biology

Prereq: None, 12 Units

21W.755 Writing and Reading Short Stories

Prereq: None, 12 Units. HASS-A

9.00 Introduction to Psychological Science

Prereq: None, 12 Units. HASS-S

9.85 Infant and Early Childhood Cognition

Prereq: 9.00, 12 Units. HASS-S

9.46 Neuroscience of Morality

Prereq: 9.00, 9.01, and 9.85, 12 Units. HASS-S

24.09 Minds and Machines

Prereq: None, 12 Units. HASS-H; CI-H

24.08[J] Philosophical Issues in Brain Science

Prereq: None, 12 Units. HASS-H; CI-H

24.211 Theory of Knowledge

Prereq: One philosophy subject, 12 Units. HASS-H

24.231 Ethics

Prereq: One philosophy subject, 12 Units. HASS-H

2 subject for Restricted Electives in Science and Technology (REST)

9.01 Introduction to Neuroscience

Prereq: None, 12 Units. REST

18.03SC Differential Equations

Prereq: None. Coreq: Calculus II (GIR), 12 Units. REST

1 subject for Laboratory Requirement (12 units) [satisfied by 6.01SC the Departmental Program]

Total GIR Subjects Required for SB Degree 17

Departmental Requirements

6.001 Structure and Interpretation of Computer Programs

Prereq: None, 18 Units

Actually, new curriculum requires modern 6.0001 program. But I don’t want to learn the basics of Python, so I replaced this one with the legendary SICP course. There are two online versions of the textbook: original and modern.

6.042J Mathematics for Computer Science

Prereq: Calculus I (GIR), 12 Units. REST

6.01SC Introduction to EE and CS I

Prereq: 6.001; Coreq: Physics II (GIR), 12 Units. Institute LAB

6.UAT Oral Communication (Management Communication for Undergraduates)

Prereq: None, 12 Units. CI-M

Computer Science Requirements

6.004 Computation Structures

Prereq: Physics II (GIR) and 6.0001, 12 Units. REST

6.005 Software Construction (or 6.031 if available)

Prereq: 6.01, 15 Units

6.006 Introduction to Algorithms

Prereq: 6.042[J]; 6.01, 12 Units

6.033 Computer Systems Engineering

Prereq: 6.004 and 6.005 (CI-M), 12 Units

6.034 Artificial Intelligence

Prereq: 6.001, 12 Units

6.045[J] Automata, Computability, and Complexity

Prereq: 6.042[J], 12 Units

Elective Subjects

6.035 Computer Language Engineering

Prereq: 6.004; 6.005 or 6.031, 12 Units

6.041SC Probabilistic Systems Analysis and Applied Probability

Prereq: Calculus II (GIR), 12 Units

6.172 Performance Engineering of Software Systems

Prereq: 6.004, 6.006, and 6.005, 18 Units

Unrestricted Electives

18.06SC Linear Algebra

Prereq: Calculus II (GIR), 12 Units. REST

6.050[J] Information, Entropy, and Computation

Prereq: Physics I (GIR), 9 Units

6.003 or 6.007 Signals and Systems

Prereq: Calculus I (GIR) and (6.0001 or 6.145), 12 Units. REST

6.803 The Human Intelligence Enterprise

Prereq: 6.034, 12 Units

6.814 Database Systems

Prereq: 6.033; 6.006, 12 Units

9.66[J] Computational Cognitive Science

Prereq: 6.034, 6.041, 12 Units

9. Minor in Brain and Cognitive Sciences

I’ve used the latest MIT curriculum. Also, I used this catalog to get pre- and co-requirements for the courses.

Courses marked with * were included into the 6-3 curriculum.

Core Subjects

9.00 Introduction to Psychological Science *

Prereq: None, 12 Units. HASS-S

9.01 Introduction to Neuroscience *

Prereq: None, 12 Units. REST

9.641J Introduction to Neural Networks (or 9.40)

Prereq: 6.041, 18.02, 18.06 12 Units

Specialized Subjects

9.66[J] Computational Cognitive Science *

Prereq: 6.034, 6.041, 12 Units

9.85 Infant and Early Childhood Cognition *

Prereq: 9.00, 12 Units. HASS-S

9.46 Neuroscience of Morality *

Prereq: 9.00, 9.01, and 9.85, 12 Units. HASS-S

Rules on Exams

I’m going to imitate the MIT’s examination rules to get my self-assessments right. Here are some of them:

I will write the exams under the same time and material constraints (no looking at the questions or solutions beforehand, no notes or forbidden things). If the time constraints not specified I will try to pass exam in default 3 hours.
I have to get at least an 80% grade on final exam.
I will grade the papers myself, comparing them to the actual MIT solutions or answers on the Internet. I’m going to place my exam papers on Github, so anyone will be able to check my answers and grading.

F.A.Q.

1. How much you’re gonna work during the challenge? It depends. I don’t have a lot of time, but I want to study at least 12-16 hours per week.

2. Do you have any plans to get an actual degree (from MIT or other college)? No, and it is not the point of this challenge, I rather have dreams than plans. But if you’re living in the US and you’re interested in getting a DIY degree, as fast as possible, then you can follow this path.

Vladyslav Siriniok

Making Instinctive Software Estimates

What Really Guides Me? Part I

1) Personal Values Test

2) The Core Values Index

3) The Portrait Values Questionnaire (PVQ)

4) Find Your Values Test

5) Life Values Inventory

6) Personal Values Assessment from Barrett Values Centre

Conclusions

Can Technology Make Discussions More Fruitful?

My MIT Challenge

My Goals

6-3. Computer Science and Engineering Curriculum

General Institute Requirements (GIRs)

6 subjects for Science Requirement

8 subjects for Humanities, Arts, and Social Sciences (HASS) Requirement; at least two of these subjects must be CI-H

2 subject for Restricted Electives in Science and Technology (REST)

Departmental Requirements

Computer Science Requirements

Elective Subjects

Unrestricted Electives

9. Minor in Brain and Cognitive Sciences

Core Subjects

9.00 Introduction to Psychological Science *

9.01 Introduction to Neuroscience *

Specialized Subjects

9.66[J] Computational Cognitive Science *

9.85 Infant and Early Childhood Cognition *

9.46 Neuroscience of Morality *

Rules on Exams

F.A.Q.