yakkomajuri

Post author (not OP) here.

Would you want a Rust SDK for Skald?

In this case do you even need a RAG? Most models will have been trained on Wikipedia anyway.

Give Jan (https://www.jan.ai/) a try for instance. You'll need to do a bit of research as to what model will give you the best perf on your system but one of the quantized Llama or Qwen models will probably suit you well.

> Do many models underperform or not support non-English languages?

Yes they do. However:

1. German is one of the more common languages to train on so more models will support it than say, Bahasa

2. There should still be a reasonable amount of multi-lingual models available. Particularly if you're OK with using proprietary models via API. AFAIK all the frontier embedding and reranking models (non open-source) are multi-lingual

Didn't know he was such a fan of rap, interesting point about brioche buns.

Also: "Karaoke should only be performed with people who have already seen your genitals." :D

hey, author (not op) here. we do do semantic chunking! I think maybe I gave the impression that we don't because of the mention of aggregating context but I tested this with questions that would require aggregating context from 15+ documents (meaning 2x that in chunks), hence the comment in the post!

yup neopets was also my first contact with programming because I wanted to have a cool website.

put it aside for years and eventually became a programmer later in life

I applaud this, and I'm a big advocate of every person finding out what's best for themselves.

I also admire the author because I derive a lot of joy from learning languages (should get back to it) but wouldn't ever undertake this.

The one comment I have here is that this feels a bit like trying to have a lot of control ("no client can cancel it") and what I've personally found for myself is that the answer to having been shattered by lack of control is learning to let go of control and flow rather than seeking more control.

But again, this is my own realization for my own mental health process! Irrespective of anything this is really cool and thanks for sharing.

I mean it would be really hard to put guardrails in place in a way that wouldn't affect real users. Besides the fact that it's ofc really hard to build guardrails period.

I've been using Claude to scan my codebase and submit issues and PRs when it finds a potential vulnerability and honestly it's pretty good.

So preventing it from doing any sort of work that can surface vulnerabilities would affect me as a user.

But yeah I'm not sure what the answer is here? Is part of it for the defender to actively use these systems to test itself before going to prod?

I guess this is similar to what Astral is doing on the Python side?

As far as Astral goes, so far they've distributed all the tooling separately but it seems they might be going towards consolidation as well.

This is not just about management but life overall. You will mess up with the people you love -- and repair is the way to go there as well.

This has led to all sorts of opinions on the thread, which are all very interesting!

I do agree that just accepting this is not the way to go, and also that slowly making changes is a valid approach.

I do want to qualify though, for those who aren't in a WhatsApp-heavy country, how things work.

I looked at my latest messages and beyond all my friends and all my family, I have my accountant, my landlord, my barber, HOA, groups for birthday party invites (where you're asked to confirm attendance), a painter, etc. In many restaurants, if you want a reservation, WhatsApp is the only way. For people who work in Brazil (I work remotely for a company abroad), a lot of work communication happens on WhatsApp.

Again, this is not to say that not dong anything is the way to go! But I think abroad some people don't understand the extent to which WhatsApp is used here. Someone mentioned iMessage for instance and I don't think I know a single person who uses it. Most Brazilians have Android phones too.

I guess this was expected, but it makes me feel really powerless in the sense that I can't really move away from WhatsApp.

I have a couple of friends that I message via Signal and even convinced my dad to use it a while back, but here in Brazil WhatsApp is _everything_, and I doubt most people care about this at all. In my case, I'd love to just go over to Signal fully but then I couldn't talk to family, friends, and probably couldn't even book a haircut or pay my taxes (my accountant messages me on WhatsApp).

It's one of those where unless just about everyone were to go over to Signal, most people won't, because keeping track of messages in two apps is quite hard.

That leaves me stuck in this ecosystem, which is quite sad.

I think your point still stands, and I'm a big Postgres advocate/user myself btw.

But yeah we did migrate our _analytics_ data to ClickHouse (while still keeping Postgres for more transactional stuff) back when I was at PostHog.

Writeup: https://posthog.com/blog/how-we-turned-clickhouse-into-our-e...

Well NuBank might have a similar trajectory I don't know but I'm talking about Banco Inter who explicitly talks about the term "SuperApp"

Did a quick stalk based on your Brazilian name... I am talking about your competitor ofc!

I have an account with you guys too but haven't kept up with the developments at all. I do wonder what direction you're going in - particularly given the tech company valuation the US market has given ya.

(I don't expect you to reply to this)

Yeah definitely. I'm not oblivious to the potential gain to the business. I'm just frustrated with the user experience of the core banking products. And it seems like this is the direction other banks might like to follow.

I feel like this with my (current) bank of choice here in Brazil. They were one of the first to focus on being digital-first and allowed opening an account without going to a branch etc. They grew fast and became one of the largest banks in the country and generally considered pretty solid. I've been banking there for like a decade.

Now they've decided to be what they call a "SuperApp". This goddamn super app has a Twitter-like thing inside of it, shopping, and literally dozens of other products. Some core banking features are now hard to find but more importantly I had quite a few issues with investments as well. People who work there also tell me about messy problems on the financial services bits. It's very clear to me that in trying to become everything, they've deprioritized the fundamental products they offer, which are those related to banking. I want to store money, send and receive it, invest it, and have access to credit. But the experience of using those features has become significantly worse as new verticals sprouted up.

Got right to playing Donkey Kong Country! Thanks for this, really cool.

Sorry. I pulled my figure from the article in the post which claims "Google had $23.47 billion in cash and cash equivalents as of Dec. 31, implying it might have to seek financing for the deal."

Extracting this from my comment on a subthread to add color to the discussion here.

They announced in a blog post that they went from $1m ARR to $100m ARR in 18 months (Feb 2021 -> July 2022). [1]

Reuters in the article posted here reports they were at $500m ARR when they last raised in mid-2024, meaning they went from $100m to $500m in around 2 years.

One would thus speculate they are likely a few hundred million above the half-a-billion figure today.

The multiple still appears a little high to me (particularly given it's all-cash, which Google doesn't even have) but what do I know.

[1] https://www.wiz.io/blog/100m-arr-in-18-months-wiz-becomes-th...

Well 500m/year when they last raised in mid-2024. There are hints as to their growth rate from their post about their 100m ARR milestone [1] and thus one knows they went from 100m to 500m in two years (mid 2022 to mid 2024).

They're thus probably higher than 500m now although the multiple still seems really high to me. But what do I know.

[1] https://www.wiz.io/blog/100m-arr-in-18-months-wiz-becomes-th...

Usually we keep the mainstream news out of here but felt I had to post this one because I've never seen anything like it and was curious to hear other people's perspective here.

To me it feels like a setup to look good in front of their audience without any proper desire to engage in true diplomacy.

The Kafka engine gets a little bit of hate (well, I haven't worked with ClickHouse for like 2 years now) but honestly it will go a long way.

At PostHog we were inserting something around 50M rows a day into CH with it and it was quite nice to be able to pause ingestion to a table by just detaching the table via SQL :D

I think they're still using the Kafka engine today but not sure. In our case (at least back then) we had to live with suboptimal batch sizes because we were providing near realtime analytics so Kafka was a solid fit.

The Chinese investment into strategic spots around the South China Sea & Gulf of Thailand is really interesting to witness.

I was at Sihanoukville (Cambodia) around 5 years ago and was surprised to find what was seemingly a Chinese city. There was a lot of dust, construction, and the city seemed like it was being raised out of the ground really quickly. Massive Chinese casinos and hotels were around and locals reported some casinos didn't even let them in. There were some whispers about shady stuff too.

I suppose if I were to go there today I'd find there are no more dirt roads and that the city is "fully built" given the rate at which these things happen. Would be cool to see.

I tried them out and found the same thing as the author. The fact that someone dared to take a different approach gathered them a large amount of attention at CES but the 2min demo had me a bit dizzy for a good 10 minutes.

You inevitably have to look up to see the display and it's also a constant struggle to aim as best as you can to be able to see what's on the display effectively.

It's important that people in the industry try and test different approaches though!

"Whereupon, three days after my father formally retired in 1988, he died in his sleep."

There's a deep point here about approaching life as it pertains to all the things we say we'll do after we retire.

I quit a hell of a job (in the good sense!) kinda out of nowhere at a point in my career where I was in a great upward trajectory to go on an "indefinite sabbatical" (turned out to be 7 months).

I did not quit to start a company, but the sense of multiple plunges does resonate.

From my perspective:

- It's very hard to detach yourself from stable (and growing) personal finances

- The comparison aspect is always there, you just need to learn to manage it

- The most rewarding bit was re-adjusting my metric of what's sufficient/enough in life

- I was not familiar with that video, but, indeed, I did learn to ask myself "would you rather be doing something else?" and the answer was no

- I learned to dream again

- As soon as I pulled the trigger on the decision I was flooded by a sense of insecurity even if I was really certain of it

I wrote stuff about it at the time but never published it.

You might find Teeny (https://github.com/yakkomajuri/teeny) interesting for this.

Something I built to have my own site in static HTML and CSS.