Hacker Newsnew | past | comments | ask | show | jobs | submit | f311a's commentslogin

It’s better to compare this to computer science, especially, theoretical one. LLMs stop people from exploring new languages, architectures and so on.

Well... computer science really is math already. :)

You can use ClickHouse cloud to host it on AWS.

But given the majority of use-cases of CH, AWS can be quite expensive.


Using BYOC management reduces the costs significantly. The big cost in analytic SaaS offerings is generally compute, which vendors mark up significantly. (They keep margins low on storage.)

Disclosure: My company Altinity offers BYOC management of ClickHouse.


They got like 10 new sponsors, you can see it on their page. They are one of the top OSS projects in terms of revenue.

I can't recall any other OSS project that makes so much money and has nothing to do with infra/cloud.

They were making around $1M a year back when they complained about the revenue. A lot of Linux distros can only dream about such an amount of money.


$1M a year isn't a lot of money if you are paying for full-time employees.


They achieved $2 million revenue in the first 5 months of the first Tailwind UI release.[1]

They had a very good run for such a project.

[1] - https://adamwathan.me/tailwindcss-from-side-project-byproduc...


Right, and as a consequence lots of very important software (orders of magnitude more important than tailwind) does not have full-time employees. Good for them if they can manage it, but it's a gravy train.


Given how hard it is to build any form of financially sustainable model around an open source project I think we should actively celebrate anyone who manages to build a model that works, not dismiss it as a "gravy train".


I really just think they got lucky, I don't see anything to celebrate. It's like celebrating one of your friends winning the lottery - good for them, but it's not really an indication the system is working well at all.


9 years of work to "get lucky".


Again, lots of people put in 9 years of work and never get anywhere with it. The success tailwind has had is not a consequence of the work.


The success of tailwind would not have happened without the work.

Just because others have worked nine years without success doesn't mean you shouldn't get credit when you put that work in and are successful.


This was after they laid everyone off, so it was just Adam and Steve left. An they surpassed $2M a year in sponsorships a couple of months after that. They were not struggling for cash at all.


I'm not sure from "they could afford payroll because they laid everyone off" I would conclude that they weren't struggling for cash.


A lot of open source that we used to use as dependencies is trivial. Sometimes you only need 10% of the library. No one is talking about kernel.

Also, changing license does not prevent you from forking the source code before the change. There are no new libraries with kernel level effort that use uncommon license.


Well, people don't play like computers, so this kind of practice can be useless.


How much of professional chess is metagaming the opponent vs. finding the best move from current game state?


From what pros are saying after matches, especially in classical chess, it's almost always preparation and metagaming the opponent vs finding good moves.

It's a lot of mind games involved, going for specific openings, repeating same moves as last match, doing a poor by surprising, move etc.

Many GMs said that to win a match you have to play non-ideal moves many times, otherwise it will likely end in a draw.


I've been using clickhouse-local for quite some time, instead of DuckDB. There is also chDB.

After using pandas for 10 years, I favor SQL now, for some reason.


Maybe because it's like a swiss army knife for data work, regardless of whether you need it for OLTP or OLAP workloads. Having different SQL dialects is a bit annoying, but the base is the same more or less, so switching doesn't come at too big of a cost.


I haven't tried chDB yet, but I heard about it. Thank you for reminding me of that option.

I use SQL in data pipelines and processing that is going to require interoperability.

But for data exploration, I usually prefer Polars (imo it is easier to work with text, semi-structured data, etc.)


As a huge duckdb fan, I'd love to see chDB to get proper windows support - that would make it real competition (having WASM coverage is already a big step) which would be good for the space as a whole.


Interesting, why do you typically prefer clickhouse local to duckdb?


Because we use regular CH in production. I'm pretty familiar with it and I can always run large queries on a production server with 1TB of RAM.

Regular CH also support external data sources, so I can read 500GB of JSON from S3 and group by it on production server very fast and in memory.


Is the google infra stable enough right now? At the start of the year, the flash model was unusable for a whole month via gemini CLI. They could not fix it for a whole month and I was a paid customer.


I haven't had any issues lately.


I use it quite a lot and after a week of use I’m being hard rate limited


I feel the same, there are still lots of people who are interested in learning internals, but they are hard to reach. Even such people don't use Google anymore, so they won't find your content.


The whole point of his model is to optimize for a very specific benchmark.

BUT, he does not use labels when training, so the model does not know the answers.


> The whole point of his model is to optimize for a very specific benchmark.

But benchmaxxing is what we generally try to avoid for training, as there is no point really for it. We used to call it "overfitting", now you're saying this person does it intentionally? Why?


There are plenty of applications where a machine learning system needs to optimize for a very limited data set that is still intractable by linear logic systems of reasonable scale and complexity. It’s interesting, because he is using the legos of LLMs to build highly specialized machine learning systems, which is a very pragmatic approach. Obviously a lot of other ways to achieve similar goals, but it’s cool to see someone back porting the modern tools towards older style optimizations.

Also, the complexity of the task he is using occupies an interesting middle ground of ultra high dimensionality (for a “simple” problem) while being limited in width to a narrow set of solves- a space where one would be tempted to imagine you would need a much more capable system.


Overfitting, as well as the specific instances I've seen of the word benchmaxxing, involve knowing the answers and training to those answers. That did not happen here. The model is limited in scope, which means it's not being scored on generic intelligence, but neither is it defective and terrible at solving new problems inside its scope, like you get with overfitting.


Why not? There is $700k reward for the next iteration of this benchmark https://www.kaggle.com/competitions/arc-prize-2026-arc-agi-2...

I would not call this overfitting, it's finetuning for specific task where you have a benchmark.


Robotics cost way more to produce and support, unlike software


And physical damage is lot more expensive to repair in vast majority of use cases. Have fun when your humanoid window washing robot punches through your window. Or any other such cases.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: