I'm not saying beating children is good but.. if you think the 'state of the world' is bad, well, most countries have massively reduced how much beating children get over the last 40 years or so.

Where the proof that caning people for mistakes creates in the the desire for vengeance where they will hit their children.

Naughty children won't work, or do vocational training. Have you worked with children?

I think what we really need is better sandboxes languages. I’d be much happier if my compression algorithm only had an input stream and an output stream. Maybe my gui library shouldn’t have network access or filesystem access. It just draws what I give it, gives me back what users press. You could still make evil software in this world of course.

Sounds like functional programming could help with that?

Not a latex post with someone talking about typst. Come back when the html output works. Not having good accessible output was more acceptable back when Tex was invented, it definitely isn’t now, and they made a new system and somehow got this worse then modern latex.

Typst does have have accessibility features.[1]

I don't worry too much about HTML output still being WIP. Even if TeX had a massive head start, Typst has a good development speed, and a little bit of slope makes up for a lot of y-intercept.

[1]: https://typst.app/docs/guides/accessibility/

Deepseek is fairly uncensored. I tried pushing it and reached my limits before it did.

Is this satire? Ask it about June 4 1989, Taiwan independence, or Winnie the Pooh.

Not that you're wrong, but I think they were talking about it from a technical POV. I use deepseek to write exploits and red team("Malicious" code). It's alignment is under different values so it's nice to be able to at least swap between models for different uses.

So, this sounds exciting to me, but the postcode checker really feels like a spam as a user. All it tells me is 'Mixed results'. I could make a website that prints 'mixed results', I bet most results are 'mixed'!

I understand wanting to get money, but honestly, there is no way I would give money to this website in it's current state, you are giving me far too little info before asking me to hand over a credit card.

Then, if someone gives you £19, a crazy amount of money honestly, the last page of the report is an advert to give them 4 times more!

Really useful feedback, cheers. Yeah, "Mixed results" is kinda rubbish as you say. It should give you something concrete before asking for anything. I'll fix that today. Fair point on the £79 upsell at the end of a £19 report too. That's tone deaf and I'll move it. On the £19... I'll think about it, but you're right the site needs to do more to justify the spend before pulling out a card. Appreciate the honest take!

Just a quick follow up, if my reply seemed very harsh, view that as a sign of how enthusiastic I was to see the website at first. I understand wanting to make money, but I'd seriously consider giving a lot more away (maybe even the basic report stuff) away for free, I'd love to explore my local area, my parent's, be nosey what life is like in Oxford (a place I previously lived), but even if I was willing to pay (I'm not), having to stop, get PDF, download, really breaks the flow.

No, that's absolutely a fair follow-up and not harsh at all. It's very helpful. The "be nosey about places you used to live" use case is exactly what the postcode tool should serve (thinking about it), and right now it doesn't. You're right that PDF-downloads break flow badly. Tbh... that's a hangover from the "people want a thing they can save" assumption that I'm still stuck in, I guess. I'm still on the fence about giving the paid reports away wholesale, but the gap between "tells you nothing" and "£19 PDF" is way too big. I'm gonna need a middle layer of free but actually useful exploration on the site. Will have a solid think about this today. Appreciate the feedback!

I'm also enthusiastic, it's not often you see people find a genuinely underserved niche and you have.

I don't know if I would pay £19 for a general state-of-the-area report. I would almost certainly have paid £100-300 for a service that took my planning application, critically reviewed it and told me which aspects were and were not likely to pass, with references to specific examples within my local area.

Thanks, honestly that means a lot! Yeah, the pre-submission review idea is interesting and I've thought about it. I have the data to surface "applications similar to yours in your ward, here's what got approved and what didn't" but I haven't built it as a workflow because it requires the user to upload their plans... and that's a different kind of trust ask, but yeah, it is definitely worth revisiting. £100-500 is also a much more honest price for something that genuinely changes a decision. £19 is in the awkward "too much for curiosity, too little for stakes" zone you and the other commenter are both pointing at.

Just checking, are you using an LLM to reply? Your replies are riddled with things LLMs are good at, like making quoted analogies that make no sense. They're not even analogies

What benefit would people gain from the reports? Average rate of success/time is interesting, but I'm not sure what you'd do with this information other than a bit of local press discourse. I suppose it's nicely timed for the council elections?

Honest answer... I don't fully know, zero paying customers so it's still very much a hypothesis. The two use cases I think hold up: (1) people pre-buying a house with extension potential, who otherwise guess or pay £500+ for a planning consultant; (2) homeowners about to commission £2-5k of architect drawings who want a sanity check before proceeding. Someone else suggested £100-500 for a proper pre-submission review which is probably better for that second case than my £19 report. The "general state-of-area" framing is the weakest one and you're right it's mostly local press discourse — that's marketing not revenue.

They made a new format with basically no accessibility. We finally got latex usable by blind people with acceptable html output, I’m not moving to something worse.

In what ways is HTML produced by transpiling LaTeX more accessible that HTML produced by Typst?

The HTML generated by LaTeX is currently very good (you can read basically every paper on arXiV in HTML). The html generated by Typst just.. doesn't really work currently. I checked a few weeks ago, tables didn't output sensibly. Looking at the docs their plan doesn't seem to be to aim for total coverage in general.

By “The HTML generated by LaTeX” do you mean by latexml (the tool used, I think, by arxiv) or something else?

This is a really nice website!

In China it turns out there are lots of rule sets. The city I'm currently living in (Changsha) has it's own ruleset for example, with less tiles than these examples.

mahjong rulesets are wild. I play Japanese mahjong, and the difference between online and a mahjong parlor is quite different, making it interesting to see what people optimize for in those different settings

I think mahjong is probably "house rules the game" though. Pretty sure most mahjong hands probably just were a result of some guy being like "hey this hand looks like it should be scored man".

It's similar to dominos then - every region and cultural/ethnic group has their own variant, and every family has their own house rules. Or craps! I was so confused my first time playing in a casino after learning to play in the streets.

Yeah, the grad students near me always play Sichuan-style, because it's simpler (only numbered tiles, scoring is a bit easier).

In my city, Guangdong, Cantonese mahjong doesn't allow chows, only pongs, and you need to self-draw to win.

What's the current best framework to have a 'claude code' like experience with Deepseek (or in general, an open-source model), if I wanted to play?

You can use deepseek with Claude code

You can, but does it work well? I assume CC has all kinds of Claude specific prompts in it, wouldn't you be better with a harness designed to be model agnostic like pi.dev or OpenCode?

I've been using all Kimi K2.6, gpt-5.4 and now Deepseek v4 (thought not extensively yet) in Claude Code and I can say it works much better than you'd expect. It looks like the system prompt and tools are pulling a lot of weight. Maybe the current models are good enough that you don't need them to be trained for a specific harness.

You can use CC with other models, you aren’t forced to use Claude model.

claude-code-cli/opencode/codex

To me, the important thing isn't that I can run it, it's that I can pay someone else to run it. I'm finding Opus 4.7 seems to be weirdly broken compared to 4.6, it just doesn't understand my code, breaks it whenever I ask it to do anything.

Now, at the moment, i can still use 4.6 but eventually Anthropic are going to remove it, and when it's gone it will be gone forever. I'm planning on trying Deepseek v4, because even if it's not quite as good, I know that it will be available forever, I'll always be able to find someone to run it.

Yep, it's wild how little emphasis is there on control and replicability in these posts.

Already these models are useful for a myriad of use cases. It's really not that important if a model can 1-shot a particular problem or draw a cuter pelican on a bike. Past a degree of quality, process and reliability are so much more important for anything other than complete hands-off usage, which in business it's not something you're really going to do.

The fact that my tool may be gone tomorrow, and this actually has happened before, with no guarantees of a proper substitute... that's a lot more of a concern than a point extra in some benchmark.