Guus Bosman

software engineering director

You are here


Generating Dutch language using Nescio's works and a GPT

I'm mostly done with the Natural Language Processing specialization and tonight I was playing around with a fun proof of concept written by Andrej Karpathy.

The proof of concept is a tiny transformer architecture, "GPT mini". Unlike its big GPT-3 and GPT-4 brothers, this tiny model is character based, not word based. The text I used are three short stories by author Nescio -- which, in fact, constitute his complete works, around 200k characters. I trained it for 20,000 iterations.


Learning about Attention

I've completed the first three courses of the Natural Language Processing specialization at Coursera, and started the fourth one today. I really enjoyed the the courses so far but this fourth course is the best part: learning about Attention.

The basic concept of this was introduced in 2014, three years later the famous "Attention is all you need" paper came out.


Continuous Bag of Words

With the rise of Large Language Models, I decided to dive into machine learning again and catch up on how these LLMs truly work. I'm following a course by on Natural Language Processing.

I'm at the end of the second course and we're going through Continuous Bag of Words, which was part of the original word2vec algorithm that was so influential. It's really fun. The previous week was about n-grams, which was also nice to program with.


A 2-mile race with Nora

Today Nora and I did the ASFS race -- a 2 mile out-and-back course on the Custis trail. It was a lot of fun today and great to run together.

Her time was around 18 minutes, mine 20 minutes. That's pretty slow for a 2 mile race but it has 2 big hills in the course. We had a little bit of rain which kept the temperature down.

It was nice to see Nora hanging out with her classmates at the end.

2 miles
20 minutes

Lunch in Potos


We spent two hours at Antonio's Beach this morning. Leo and I tried snorkeling but there wasnt much to sea and the water cloudy, so we swam instead. The kids made friends with a small Romanian boy while make their sandcastle.

Around 2 pm we left for Potos for lunch. Were turned to Taverna Laskarina because the food was so delicious there. The chicken souvlaki in particular is outstanding. Sasha had European anchovy which was delicious, kind of a bigger version of цаца. Leo had grilled large shrimps, Nora her souvlaki and Adrian the pork ribs.


Bicycling to Enkhuizen

For our trip to Sprookjeswonderland I bicycled from Middenmeer to Enkhuizen and back, a gorgeous trip that took me through the heart of West-Friesland.


"Onze familie"


Adrian drew a picture at school of our family. Half of it in the Netherlands, half in Bulgaria. Can you guess which side is which?

Hint, on the left they are wearing rain jackets.

Adrian explained: "ik was vergeten mezelf te tekenen".


Medieval literature and scientific research

One of my favorite books is Het Bureau, about the politics in a scientific institute about culture and language.

Today arstechnica wrote about a study published by the actual institute from the book, and it was very interesting to read. Ever since I've been reading the ancient classics, I've been fascinated by "books that disappeared". The study tries to estimate how many works from medieval literature still survive -- their headline number is that only 9% survived.

One of the characters in the book predicted my interest in studies like this, quite rightly:

'Er komen daar natuurlijk allemaal natuurkundigen, chemici, biologen, geologen. Wat moeten die met de Nederlandse volkstaal en volkcultuur?' 'Dat interesseert ze juist!' zei Balk apodictisch. 'Eindelijk eens echte wetenschap!' - hij lachte met een grimmig sarcasme. Voor wat wij doen, heeft iedere intellectueel belangstelling. Dat geeft die bètamensen juist status! En anders prikkelt het het snobisme!' - hij glimlachte gemeen.


Microsoft Certified Azure Fundamentals

I successfully completed a certification for Azure, the cloud offering from Microsoft.

I took the exam in our basement; it was strange to be on video all the time. But the process was nice and smooth.


Leo & Adrian completed Veilig Leren Lezen


Leo & Adrian completed Veilig Leren Lezen, a Dutch reading curriculum for first grade. Today they received their official diploma; they really completed the work several months ago.

When the boys were little, I was thinking that it might be hard to get the Dutch reading skills to the same level as Nora. I spent quite a lot of time with Nora and with 3 kids instead of 1, it would be hard to give that same attention to the boys, I figured. The pandemic changed that equation dramatically of course and since March 2020 we have done extensive Veilig Leren Lezen exercises, most days of the week, and the result is there.

Both boys read Dutch very well and read books well above their grade level. Adrian loves reading about history ("De weg naar Titicaca" was one of his recent reads), Leo enjoys the Dolfje Weerwolfje series.

I'm really proud of them, a great achievement.


Recent comments

Recently read

Books I've recently read: