druskacik

How's the performance compared to ordinary 9B LLM with structured outputs? Both accuracy and speed?

Yeah, it's weird, considering ModernBERT, which the Laya models are based on, supports 8192 context window.

I'm really interested in technical details behind Jev (not this), how it can work so fast and so cheap. It's probably large (must be since the performance is so good) but somehow still fast, so it must include some really non-trivial stuff. The price suggests it may be runnable locally, but who knows.

If it was possible to re-create it as an open-weight, it would be exciting!

Classical music concerts database. The interesting part maybe isn't what it is, but how it's done: I'm aiming for a fully automated database, where "crawlers" for the classical music websites are created and validated by AI.

Rough idea: get a list of urls with classical music concerts -> send each one to a coding agent to build a crawler -> validate and run on an infrastructure for "free". Classical music is an ideal use case, because the number of concerts is small enough on an average website that it doensn't require many workarounds, and the websites are usually pretty crawler-friendly.

https://classicalbot.com/

I actually have a good use case for this. For my project, I run a lot of Codex sessions in parallel (via Codex SDK[0]), because they are solving self-contained tasks (building crawlers for websites) and in theory I could scale it to hundreds or thousands of parallel sessions. But my VPS can handle maybe 10 parallel sessions max.

Btw, the crawlers are for classical music websites, the project is https://classicalbot.com/ .

[0] https://learn.chatgpt.com/docs/codex-sdk

Interesting. I actually prefer when agents don't commit on my behalf unless I explicitly say so, I even had to add a custom instruction for Claude to stop doing it (Codex never does it). Even if I don't read all the code line-by-line, I at least want to see the changes at glance and commit myself. Git Fork[0] is a great tool for that, by the way.

In general, I don't like when I have to prompt models to NOT do something. It's probably difficult for the AI companies to get this right, they should understand ambiguity but still not over-do simple instructions.

[0] https://git-fork.com/

> This kind of happens today

What do you report to the police if the act isn't illegal?

I live in Prague and seeing the temperatures expected during weekend, I bought at least the portable AC. I'd say 2-4 weeks is quite a long period to justify this purchase.

Yes, historically, the AC didn't make sense in Prague, but there's no going back, the world is just going to get hotter.

Time off of work to get a haircut is insane. Aren't there barbershops open in the evening?

Yeah, the focus is mostly on night trains, although some day combinations leak to the data, so it's possible you'll find some combinations for daily trains.

Of course, I'd like to have the daily trains in the data, but that would drastically increase the number of combinations and require some more scaling of the scraping, which I don't want to do at this point. Focusing on night trains only is a nice niche.

I plan to add at least some daily trains between core cities, e.g. the EuroStar London <-> Paris route, which would make multi-leg search to farther cities easier (i.e. combination of night train + daily train).

It's definitely similar, but the philosophy seems different. NightRide supports exactly the use case on searching a route on a specific date. I also noticed that if they don't have the price pre-scraped, a scrape is called dynamically on user's request.

My idea is to pre-scrape the prices and answer questions such as: wanna go to Amsterdam for a weekend? -> here are the best dates and prices.

Added, but just the Milan - Brusells route from European Sleeper so far.

https://trainbot.eu/?from=milan&to=brussels&type=return

The 20€ Milan - Zurich tickets also don't look bad, but the early-morning departure / evening arrival are probably not that optimal.

https://trainbot.eu/?from=milan&to=zurich&type=return&seat_t...

EDIT: note, this route will be available from September.

Currently not possible, but:

- once you fill the "From" field with your start city, the "To" field will include only cities reachable from the "From" city

- at https://trainbot.eu/coverage/ you can search for all routes that go through your city

I plan to add a subpage for every relevant city with recommended routes once I'll have enough data. Something like https://www.seat61.com/ but with actual prices and dates.

Hm, thanks for the tip, I didn't know that. I scrape NightJet website, which lists only the 00:36 train. I'll look into it.

sorry about that :)

I have a working version of Trenitalia scraper (Intercity Notte trains across Italy), I just haven't added the routes to UI before verifying it works reliably. Hopefully I'll add it in the evening.

Also the Milan - Brusells route from European Sleeper, it's scraped but not yet in UI. I'll reply here when it's done.

That's the Prague - Zurich NightJet that's hellishly expensive, possibly because the target audience are Swiss people for whom it's not so much :D Also, NightJet does not price very dynamically as opposed to other providers, and their baseline price is always quite high.

But yes, in most cases this journey will be more expensive that an alternative flight, which is a shame. However, there are routes where it's comparable, e.g. the Prague - Amsterdam route.

Sure :) Thanks.

European night trains search engine. The plan is not to provide the functionality where user searches for a specific route on a specific date and the engine returns prices. The plan is to provide "tips for trips", e.g. I provide a starting city and it will recommend me interesting trip ideas any time in the future. There are many flight apps that provide this functionality, but no train app.

Currently I'm building scrapers for all relevant provider, then I want to connect the data for multi-city trips recommendation. Plus some connection to the day trains so that the trips are built more easily.

Some existing trip recommendations:

Prague <-> Amsterdam https://trainbot.eu/?from=prague&to=amsterdam&type=return

Berlin <-> Paris https://trainbot.eu/?from=berlin&to=paris&type=return

Zurich <-> Budapest https://trainbot.eu/?from=zurich&to=budapest&type=return

For me personally, I have the basic Claude Code subscription that I use to rewind on some evenings or on weekend, to code a bit for 1-2 hours. I have like 3-5 session with it every week.

The 5h windows are frustrating because I can go through them quickly if I have a more complex task. I haven't yet met the weekly limit. I'd say there are many cases similar to mine.

> If the man holding the flag had been wearing a thawb instead of a suit, or if the statue had been of a woman, I think the establishment's response would be quite different.

That's argumentum ad speculum[0]. You can speculate what the response would be if the statue was different in a way you imagine, but the thing is, it's not.

[0]: https://www.logicallyfallacious.com/logicalfallacies/Hypothe...

What truly deep art would you recommend for us laymans who enjoy Banksy's works?

Although it seems the night train offering in Spain is not great.

It's a work in progress and southern Europe is not covered well, Spain is not covered at all. I focuses on Eastern and Central Europe because that's where I'm from, I'll try to add Spain next :)

Altman is an advocate of Universal Basic Income, as far as I'm aware. That doesn't sound like he's not worried about massive job losses.

https://www.cbsnews.com/news/sam-altman-universal-basic-inco...

https://finance.yahoo.com/news/sam-altman-wants-universal-ex...

The 8B model response to my "Harry Potter knowledge-bench" question is too funny not to share.

> *Fathers of Harry and James Potter*: - Sirius Black is the *father* of *James Potter* (the older brother of Harry).

> - James Potter is *Harry's uncle* and the *older brother* of *Luna Lovegood*.

> - This means *Sirius and James are Harry's uncles*, though they are *father and brother*.

https://pastebin.com/WAAmFKfX

I agree, their autocomplete (tab) model is the best, but recently I realised I am using it less and less - the new models are so good that I mostly just do agentic coding, and I do very little changes in the codebase by myself. This is probably a general trend and if the usage of autocomplete models is dying out, it's understandable the companies are not investing resources into it.

That has been true for some time now, definitely since Claude 3 release two years ago.

According to the announcement blog Le Chat is powered by the new model as well: https://chat.mistral.ai/chat

I wonder when we will see an LLM deployed directly in a space rover on a mission like this. The time savings could be huge.