Artgor

When I was interviewing at my last company, during the raising bar interview, the interviewer said that he looked at my personal website, liked it, and was really impressed by it. Then we spent one hour chatting about our experiences. So, I can say that it definitely helped me get hired. https://andlukyane.com/blog/

Last year, I started playing DnD with 5e rules and lots of homebrew content for the first time. It is an online game with my friends on Discord. We have character sheets, roll a dice using a bot, and so on. This is really fun, we had some crazy adventures and this campaign is just at early stages :)

> I wonder how hard it would be to modify this code to run on a 64GB M2 Mac.

It isn't that hard, I was able to run in on M1. The changes are:

remove or modify multiprocessing - it doesn't work on Mac the same way as in the code;

replace `device = "cuda"` with `device = "mps"`

In this line ` att_idxs = (torch.clamp(torch.arange(context_size)[None, :] - torch.arange(context_size)[:, None], -pos_emb_radius, pos_emb_radius-1) % pos_emb_size).to("cuda")` replace cuda with "mps"

in `optim.AdamW` remove `fused=True` - we can't do it without CUDA

Replace ```with autocast(device_type='cuda', dtype=torch.float16): _, loss = mlm_head(bert(batch_data_torch_xs[mb_start_idx:mb_end_idx]), batch_data_torch_ys[mb_start_idx:mb_end_idx]) ```

with simply `_, loss = mlm_head(bert(batch_data_torch_xs[mb_start_idx:mb_end_idx]), batch_data_torch_ys[mb_start_idx:mb_end_idx])`

replace `scaler.scale(corrected_loss).backward()` with `corrected_loss.backward()`

replace ``` scaler.unscale_(optimizer) scaler.step(optimizer) scaler.update() ``` with `optimizer.step()`

It should work.

The paper sounds interesting, but without the code being available, it is difficult to say whether the idea is really viable.

In general I agree. But if this approach is applied for the same system many times in a row, it can cause technical debt.

https://andlukyane.com/blog/ I write about Data Science and my interests (like language learning). Recently most of my posts are dedicated to reviewing deep learning papers.

I created a personal website a couple of years ago: https://andlukyane.com/ 1. It is a convenient place to keep and manage information about career - jobs, talks, other activities. Makes it easier to share info with recruiters. 2. Most of my blogposts are paper reviews (on ML), some are about my experience. 3. I got several interviews thanks to this blog. During some interviews people shared very positive feedback on it. The most notable example was the last interview in my current company: it was a bar raiser, the interviewer told me that he looked at my website and really liked it. It made the interview very positive and resulted in me getting this job. 4. I got a couple of small consulting gigs thanks to my website.

> the fact that the linear lessons now also have large yet optional testing components at the end of every section.

If you're referring to the legendary lessons, then yes, I complete them as well and find them to be very useful. However, many people don't mention them because they cost a lot of gems for non-premium users (I use Duolingo Super), so they can't consistently complete these lessons. Sentence translation is an important part of language learning, yet it's kept behind a paywall. While I understand that the company needs to earn money, for non-paying users these lessons essentially don't exist.

> For all the complaining that I've seen in the in-app comments for certain translations being incorrectly rejected, I've yet to come across one myself. Unfortunately, this has become a serious problem for me. I didn't encounter such situations at the beginning of the course either, but the further I progress, the more frequently I come across them. It's to be expected, as the content is mostly created by crowdsourcing, and later lessons have fewer people studying them, so not as many people check them. Despite reporting dozens of such cases on Duolingo in the past year, I have yet to receive a single response.

I hope it is allowed to share links here. https://andlukyane.com/blog/studying-foreign-languages - I wrote a detailed blogpost about my experience of learning languages, including the resources that I used (not promoting anything).

COVID. People started spending more time at home and needed new ways of spending time.

As a user, I feel that Duolingo is a great tool for developing language learning habits and allowing people to "play", but it may only take people up to B1 level, not higher.

I would love for Duolingo to focus more on content, and not just on gamification. Right now, the main courses (such as Spanish and German) are great, but many courses are short, lack grammar tips, and have mistakes.

Even the Spanish course has problems: I am currently on unit 175 out of 211, and I have noticed that the lessons gradually have more and more problems. Some phrases are stiff, some are completely unnatural (and people complain in the comments), sometimes there are only a handful of accepted translations and we are forced to guess, and sometimes translations are inconsistent - even though several questions have a similar structure, the accepted answers are different.

Another point is the recent redesign. Personally, I'm okay with it and I have no problem with changing the tree into the path. What I didn't like is that the course was artificially inflated, and now we have to repeat the same things many times. One could argue that repetition is key to learning a language, but it can get tedious when you have to answer the same question multiple times without making any mistakes. Furthermore, the exercises themselves are becoming easier - previously there were more questions where it was necessary to type in the full translation, but now in most exercises we only need to input a single word or rearrange the word cards.

Lastly, many people who seriously participate in the leaderboard ranking don't actually learn the language, they simply farm experience. Their usual approach is to rack up XP bonuses and redo a single exercise dozens and even hundreds of times in a row. It's sad that this behavior is encouraged.

As for myself, I use Duolingo simply as an additional practice tool. I use many other resources for actual language learning. I recently reached Spanish B2 level, and many years ago, I reached German B2 and Japanese N3.

I want to share my experience of studying different languages, describe how my approach evolved over time, and give advice on studying more efficiently. Russian is my native language, I speak English fluently, in 2013 I reached B2 in German, got to ~N3 in Japanese in 2014, and recently I reached B2 in Spanish.

I started my website and blog in 2021. I used Jekyll + github pages with custom domain. I didn't want to setup everything by myself, so I bought a Jekyll theme that I liked and customized it. I did it lazily, so it took me around 1 month to complete the whole process, including copying my old blogposts from Medium.

You could start with Word2Vec or GloVe, for example. Another option is to to with CV papers and start with Alexnet or first ResNet.

One of the new ones: I started using chatGPT to write feedback for an annual review. I would have never be able to write such bullshit without feeling sick, but the model amazingly well - I apply only minuscule changed to the generated text.

Several years ago, a wrote a blog post about a DS roadmap, which turned out to be useful for many people.

Since that time I got more experience, and many people reached out to me with different questions about career, skills, mindset, etc. Now I decided to write a blog post based on my answers to these questions and my experience. I hope it will be useful

When I had started reading this article, I had thought that it would be a sad story about another startup failure. The blogpost turned out to be a fascinating story of the success. I really liked it.

But after I had finished reading it, I have realized that it is a sad story, if we look from the eyes of data scientists in the team. People were hired to do cool machine learning projects, but it turned out there is no infrastructure for them. After the new boss had arrived, they had to work as analysts for months. What is more sad - the new boss dangled a carrot before them several times, but each time the carrot disappeared.