Notices by Daniel (djh@chaos.social)

@thomaskujawa @jackpearse Cycle.Travel ist super, nutzt die OpenStreetMap Daten und wird von @richardf entwickelt, der schon seit Ewigkeiten bei OpenStreetMap mit dabei ist. Die routing engine ist auch open source.

@thomaskujawa @jackpearse Was unabhängig dazu super geht: per @streetcomplete app die Aufgaben für's Radfahren erledigen.

Also z.b. Oberflächenbeschaffenheiten ("surface tags") zu OpenStreetMap Wege hinzufügen.

Das hilft für das Fahrrad Routing egal welche app dann die Daten nutzen wird.

@obrhoff Yeah I don't know if sci-fi is good at predicting the future or we're just discarding all the space lizard lore - or a mixture of both.

@obrhoff def. a sprinkle of survivorship bias in there for sure 🫠

@obrhoff Okay aber was kannst du zu den stabilen Spatzen sagen 👀

leider nein 🐦✨

@obrhoff This goes up all the way to institutional investors.

Just look at the tech sector: Even state pension funds / sovereign wealth funds invest overseas. One of the reasons they do invest in Europe is because of incentives or requirements they have to invest in local companies.

This is the European tech VC space in a nutshell.

@ianthetechie @obrhoff @everydoor @zverik I believe the more interesting question is not technical but how can we build something that incentivizes the less nerdy folks to contribute to free and open geo data and/or the fediverse.

With nerdy apps such as StreetComplete or EveryDoor, or Mastodon or Pixelfed, we're still not getting the incentives working as e.g. Meta does: Most companies want an Instagram or Facebook account, simply because it's good for their business.

@obrhoff Ha! Sure it's more about the geo x fediverse intersection and what could be! 🙂

@obrhoff and from a while ago: https://chaos.social/@djh/111937041846184707

If you want to go down that route I can recommend coming by the Berlin OpenStreetMap meetup or even giving a lightning talk at GeomobBER.

I can connect you to both and always happy to talk geo 🤓 I think the intersection of geo x fediverse is underserved for sure.

@obrhoff The ActivityStream spec has a (very limited) Place object for very basic geo integration into the fediverse

https://www.w3.org/TR/activitystreams-vocabulary/#places

Unfortunately on the Mastodon side there seems to be resistance to implement anything geo related; various tickets, here are just some

https://github.com/mastodon/mastodon/issues/281

https://github.com/mastodon/mastodon/issues/29002

If you don't mind about participating networks like Mastodon not making use of it, there's nothing stopping you from sending geo data.

@obrhoff

"Edit prediction won't be free forever, but right now we're just excited to share and learn."

https://zed.dev/blog/edit-prediction

🦧

Not sure these small AI models are there yet in terms of quality / latency to run them offline. Especially when most folks don't have powerful GPUs in their dev laptops. Mac folks are probably best positioned here.

#SamAltman of #OpenAI fame on a panel discussion live right now at TU #Berlin

Very first question if folks on the panel say "please" and "thank you" to #ChatGPT.

Anthropomorphising #LLMs — let's maybe not?

Overall I would have loved for a more critical discussion than giving Sam a platform for unchallenged narratives and takes.

@tuberlin

@obrhoff Ah that sucks. I can feel (!) it's getting worse, too, these days. But unless there are tents and needles on the streets and toilet paper getting thrown around, I would in no way compare it to SF / NY.

@obrhoff How is it comparable here in Berlin? Care to elaborate? Do you have numbers on this or is it just a feeling or perchance a hyperbolic mid-day rant? 🌝

@obrhoff 450M but not a single market; and not much access to capital. I believe those two specifically make up for most of our differences.

@obrhoff Yeah the smaller distill models you can locally are great, too, it's just important to understand that you won't get the proper deepseek-r1 everyone is hyped about.

What models to pick for running locally depends on: What you can fit in memory, if you run on cpu or gpu, if you want to optimize for latency e.g. for interactive use cases, and what kind of quality output you're looking for.

Just try a bunch and see what happens 😛

@obrhoff It's all open research

https://arxiv.org/search/cs?searchtype=author&query=DeepSeek-AI

For details on deepseek-r1 and the qwen / llama distilled models, see

https://arxiv.org/pdf/2501.12948

for the distilled model benchmark see table 5.

They're qwen / llama model architectures and different compared to their main contribution.

@obrhoff because it's not deepseek-r1.

Only the 671b / 404gb model is deepseek-r1 and most folks won't be able to run that locally.

All other variants are the qwen / llama distilled models. It's highly confusing that folks like ollama name them deepseek-r1. They're completely different models.

@obrhoff talk is cheap, send pull request 🤓