Ilasky

https://dailytokki.com

It's an email-only Korean (and English) language learning service started by me and my partner! We've both been interested in language and, since I've been learning Korean and she's been improving English, I've been looking for a modality that works for me -- apps don't click for me for some reason.

It started by me sending myself an email everyday with new Korean vocab/grammar, but we thought it would be nice if it responded with corrections and learned from my common mistakes to make better questions. So, we've built it out to work for us, and turns out other people also like it and are growing everyday!

Always excited to see others working in the language learning space! (Korean & English in my case)

Signed up and played around a bit, but was unable to get a video transcribed. It kept popping up with a "Error loading video. Please try again later." issue. I tried entering a URL and was given that the audio quality was too low.

Separately, I'm curious how you're approaching quality control in your translations for the words. While building my app, the GPT models regularly get translations wrong for Korean and miss nuance and require human-in-the-loop before sending to the user.

Great work shipping! It looks great!

This is eerily well-timed!

My partner and I do something similar for Korean & English (she’s Korean native and is fluent in English and I’m learning Korean). We actually built it out for ourselves and some friends and just released it yesterday[0].

Still working out some kinks, but it sends a question every weekday via email that you’d respond to. It then sends back feedback on vocab & grammar, all with spaced repetition baked in to keep track of words you learn/use as you continue.

It’s currently tailored towards those that can already read and have basics under their belt.

[0] https://dailytokki.com/?ref=hn

I started learning violin about 4 years ago. I used to play saxophone, but always loved classical.

To say that the beginning was rough is an understatement. Violin (and any string instrument, really) is just difficult, plain and simple.

But daily practice, private lessons, and I’m able to play pieces that I’ve only heard recordings of. Sure, it’s not professional, but it brings me immense joy to be able to have my hands and fingers know what to do on instinct now.

Private lessons for me were the key: accountability and expertise at every step. Self-learning would have been so much slower.

The more you learn, and I’m sure it’s like this with any art, the more you realize there’s so much nuance to it and that there’s always room to improve.

Actually just nailed down the date to Oct 6th!

Yes! This cohort we are actually splitting into 2 times so that everyone can join at a time that's convenient (had some people that needed to join very early before and want to avoid that this time)

There's been a couple articles about finishing projects posted in the last couple weeks[0][1] and in addition to this topic being incredibly interesting, it also seems pretty pervasive in the community.

A lot of people, myself included, tend to work well with external stimuli, e.g., other people. Especially when there aren't users of the product, there's no one you're really accountable for except yourself. When you start to have users, of course, that becomes different.

So, the question really becomes: how can I bake in accountability in the early stages of my project to see it through and finish?

My answer has been to build around others. A group of us meet weekly over the course of a 6-week period to chat about our project, progress, etc. with a "demo day" at the end where we're expected to show the final thing.[2] It's honestly really fun and we're about to start our third "cohort".

[0] https://news.ycombinator.com/item?id=41428705

[1] https://news.ycombinator.com/item?id=41293219

[2] https://lmt2.com

One use case I’d love to see is the ability to find the best price for multiple city tickets.

I’m fairly flexible on dates, but getting the best price for it is a super iterative process. Having some similar functionality like excel’s solve function would be awesome to find optimal dates within a range for each destination.

Ah — I see! Thanks for taking a screenshot.

Will work on making the spacing better on desktop!

Do you have a screenshot of what you got? Or can you describe it?

I just tried it and it gave the correct answer.

Agreed the ordering is a bit weird. Working on figuring out a way to adjust that. Thanks for the feedback!

Actively working on making the font more legible and adding a “speed slider”!

Unlike Fermat, the whiteboard has enough room in the margin :)

Using a variety of hand-written inputs, so font variation would be present. If you could give a screenshot, that’d be helpful!

Should be back up!

I made https://whiteboard.lasky.io, which is basically if you gave an LLM a marker and told it to solve math problems.

Also made https://resgen.app to help candidates tailor resumes

Great to hear!

Awesome -- still need to figure out exact time for next cohort. Since we have people from all over (SF to Seoul), finding a time is tricky, but so far has been okay! I'm currently EU time as well. So, I'd say, go ahead and sign up for now and I'll do my best to make sure I can find a time that works for everyone.

Thanks! That was a fun project to build. Would love for you to join!

I’ve always been in the camp that I enjoy making things and code is simply the medium in which I feel most comfortable to do so. Sometimes with money in mind, sometimes not.

In a similar vein, I really enjoy building alongside others. So, I’ve been running a group that all build things together over the course of 6-week cohorts.[0] (Just finished the second one!) it’s really fun to see everyone nerd-out about their project and challenges they face. And hits that pure “just for fun. No really” vibe for me.

[0] https://lmt2.com

I was inspired by https://calligrapher.ai to see if I can make something that looks similar to it, but for math.

I ended up hooking up an LLM on the backend to solve math problems as well.

This ended up being more of an exploration in trying to make alternative outputs for LLMs that are a little more dynamic than just text and/or LaTeX.

Those strategies for finishing projects are exactly why I’ve been running a six-week group[0] where we all work on our projects together.

Setting goals, time-boxing, accountability, celebrations — it’s all built in.

I’ve definitely found that during the 6 weeks that I find a laser focus on what I want to do and how I want to do it.

It adds a fence to the green field project, so to speak.

Highly recommend building alongside others, if you haven’t tried it yet.

[0] https://lmt2.com

Did something kind of similar (needs a little updating, but I find it still useful) for price vs performance on the LMSYS leaderboard.[0][1]

[0] https://chat.lmsys.org/?leaderboard

[1] https://llmcompare.net

AI is probably losing you a lot of good candidates. It can be easily gamed, and then you'll end up with someone who watched a TikTok video on "how to fool any ChatGPT in ten seconds". I also fail to understand how I can have multiple years of experience in a job as well as active certifications, and somehow still get filtered. My resume isn't that bad...what on earth is your AI doing? Do you know?

While the ol’ prompt injection tactic may soon stop working, tailoring your resume for each job to increase your interview rate won’t stop working anytime soon. Specifically, knowing which of your experiences align well with the requirements is super useful in tailoring. If you’re interested, I made a quick app[0] that does cosine similarity of your bullets to the job requirements to help visualize how well your resume aligns with the job description.

[0] https://resgen.app

I was in South Korea for about 3 months earlier this year and went in absolutely loving kimchi. But what surprised me was how much a meal is determined by the type of kimchi that’s available.

For example, fresh kimchi is best with boiled pork belly, while a more sour kimchi is better for stew. And then there’s this “peak” kimchi for eating by itself — it’s tremendous when it’s homemade.

Can’t wait to go back to SK and, if you haven’t had it, I highly recommend finding some kimchi & pork belly. It’s an unbeatable combo.

I love Typst. It’s especially good for dealing with programmatic input as the —input flag can take json. [0] Then, in the file it’s easy to get the json with the dedicated json.decode() function. Super easy to get solid results. Swapped to it from LaTeX in my own project [1]

[0] https://typst.app/docs/reference/foundations/sys/

[1] https://resgen.app

Something I could recommend would be working alongside others to get it done.

I’ve actually been running a little group[0] that gets together every week over the course of 6 weeks to push their products over the finish line. The weekly accountability really helps keep the motivation going. It’s actually a ton a fun! We’re just about to start our next 6 weeks and always welcome more people.

[0] https://lmt2.com

I’ve actually made the switch over to Typst[0] for my app [1]. I’ve previously used a quick jinja .tex template that then just pasted things in, but LaTeX can really throw some strange errors and overall handling the files was a hassle.

Typst was much easier to setup and the function-based operation meant that sending variables in was a breeze with better error handling there too. Also, I just grok the syntax a lot better.

Just another option for folks looking to redo their resumes/not use Latex.

[0] https://typst.app

[1] https://resgen.app

I dig it! Kind of related, but I made a comparison of LLM API costs vs their leaderboard performance to gauge which models can be more bang for the buck [0]

[0] https://llmcompare.net

I made https://gpupricecompare.com to compare GPU prices from eBay over the weekend and it was an interesting project because it seemed relatively simple upfront, but dealing with titles was a bit tricky. The only solution (I found at least) was to pull in some NLP packages to filter out and match GPUs, get the relevant models, etc. Was a great exercise for some non-LLM NLP.