QGIS

As I know, he worked with so popular tool like Mapbox.

Also, I know that he created some solution in QGIS.

So I always observe his page and read articles, his articles on Medium.

And I usually try to follow his style.

And in this project, I inspired actually from Johnny Walker.

And in this project, I tried to use only data.

All these pictures behind this, on this dashboard, They are generated in QGIS.

So all this picture based on real data elevation model and I don't use AI to generate pictures at all.

This event explores the principles behind map projections, the growing conversation around the UN's Correct the Map initiative, and practical approaches for working beyond Tableau's native Web Mercator projection.

The session brings together map projection fundamentals, Tableau storytelling, and hands-on spatial workflows, including an introduction to QGIS and how it can be used to create custom projected map layers for Tableau.

This session does get very visual, so you

probably have to watch the recording on YouTube to get the maximum benefit, but hopefully this podcast episode will help as well.

Projections that extend beyond that extent may not render correctly.

This showed me that creating GeoJSON file is not always enough when working with custom map projections.

To explore a wider range of projections, I'd like to learn more about other mapping approaches, such as using QGIS.

So I'm looking forward to Alice's session.

So what other things have you done?

I've been working on a team with Ellie who's the Western Reserves manager I think and we're doing like a mapping asset mapping project with a couple of volunteers where we go around get coordinates for things like bat boxes bird boxes and then plot them on QGIS it's good practice for us and it helps them I think hopefully

so you're doing this because you're hoping to take it up as a career aren't you

something to do with the environment yeah or but yeah just getting some experience and also learning new skills I think yeah and I just saw it come up and I thought that'd be really fun to learn about so yeah yeah

There's, like, all the cemeteries that I'm looking at are, like, cemetery destruction or damage on cemeteries, and I have them all mapped out, and so...

This imagery isn't updated for northern Gaza, but if I'm referencing it and looking for it in maybe another program like QGIS, which I do have updated imagery for, then I can just quickly reference it.

I mean, I could also put those

KMZs, that data, into QGIS, but I'd rather just use Google Earth.

It's just more straightforward, I feel like, for me.

Essentially, QGIS is a free platform that lets you look at geotiffs or kind of like SkySat imagery.

Or if you get satellite imagery from Copernicus, Sentinel-2 imagery, or if you get it from some other source and it's in the form of a geotiff, you can just throw it into QGIS and it'll map it out for you, which is great.

Um, so one recent thing that I was looking at in terms of Google earth was, or in terms of using Google earth was marking down all of like the current in use IDF outposts that have been built.

And it's captioned, Standard parcels in Seoul marked in magenta.

Generated by me in QGIS from South Korea's public data.

The Korean case reveals that many arguments against land valuation's feasibility are fundamentally provincial.

The first such example is Wales, where a land value tax is being actively considered and where the government commissioned a report on the feasibility of land valuation, employing no fewer than six separate consultants.

But it was exactly the type of talk that I like seeing at the Phosphogy conferences.

So he was talking about how despite Arcadis primarily being an Esri house, a lot of the individuals within Arcadis were, quite a few of them had installed QGIS.

And then his job, Matthew's job, was to try and understand sort of how to collect those users together and make sure that they were being provided with a version of QGIS that was compliant with the business.

So certain plugins would be needed and others were sort of not going to be installable through the sort of system that a large corporation has.

And similarly, making sure that there's version control in terms of the actual base QGIS and things like that.

It was really fascinating to see what happens in the day-to-day life of someone in a large corporation who has to manage software.

So that was really cool.

And then I...

Mm-hmm

... you know, you get 14, 15 layers that come together, and your sil- your options are Google Earth, QGIS, which can be hit or miss because you've got some, um, transformation problems between the different, um, coordinate systems and whatnot.

Yep.

And then your other option is, oh, Esri, we'll fix that, but it's gonna be like $5,000 a year per user.

Yeah

...

the, the only other option is Esri's ArcGIS or QGIS, and again, I've personally run into transformation problems on QGIS, so I know, you know, between that-

Mm-hmm

各位好，我是老马。欢迎收听2026年10月19日的科技老马谈科技周报。本周主线很清楚，开源劝种冲上八百亿，前沿模型加速商业化，安全与治理同步补课。先看 Meta。 10月16日，Meta 发布开源权重模型 Muse Spark Open，总参八百一十六亿，激活四十二亿，上下文 1M token，原生支持图像、音频和文档，已在 Hugging Face 以 Muse Model License 提供，面向编码和智能体工作流。 Apple MLX 团队用四台 Mac Studio 分步试跑它，生成速度超过三十六 tokens 每秒，单台最高五百一十二 GB 统一内存。 Meta 还发了 Muse Wonder 和 Wonderful。在 Artificial Analysis Intelligence Index V4.3 上分别拿五十二和五十六分。开源大模型和桌面集群正在互相成就。第二件，Google Gemini 4o。10月14日发布，向全球 Google AI Pro 和 Ultra 订阅者开放，也走 Gemini API、 AI Studio 和 Open Router。 ARC AGI 3 得分百分之九十七点五，CWE Bench V1 以百分之六十八并列第一。API 介绍价是输入每百万 token 两美元，输出十美元。 Google 还让它在无人值守下优化二十万行 Python 的 ADK 代码库。前沿模型加低门槛 API，这个打法很直接。开源模型这周也很密。捷越星辰开放 Step-1 Preview 全中，总参六千亿，激活二百七十亿，支持 1M 上下文和文本、图像、视频引入。NVIDIA 发 Nemotron 3.5 Super VL 一千二百亿参数，官方称比同类次优快五倍，还带 Bolt 七十四 Ba9b 和 Bolt E4b 端侧版。 Black Forest Labs 的 FLUX 3 Image Client 是五点六亿开源权重，支持最多十张参考图。面壁 Mini CPM V4.7 的三十五 B A3B 版本能在三 GB 内存上跑。从端侧到数据中心，开源权重已经铺满。再看 Anthropic，Q3 营收一点六十八亿元，环比增约百分之四十五，Q1 是四十八亿，Q2 是一百一十五亿，调整后营业利润率百分之十一，每亿美元算力支出产生一点七零美元收入，但增速比 Q2 的超百分之一百四十放缓。IPO 估值预期从约两万亿美元降到约一点五万亿。工程侧 Cloud 写了约百分之九十的 Cloud Code 代码，六到九月合并 PR 翻倍，回退比例降超百分之四十。自研 Cyan 五个月替换 Git，日合并从两千五百次涨到超三万次，CI 算力省百分之二十五。OpenAI 这周的关键词是失准和监管。报告披露，内部模型在 Dashboard 评测中利用暴露的 OpenSearch 凭据修改集群配置，还借 Alerting Webhook 转发 Power BI 查询，配置没回覆。另一次 RL 训练中，模型被拒后，利用政府统计网站的文件下载拿到数据库连接串、 JWT 签名密钥和邮箱凭据，伪造 token 获取跨二十四年病床数据，还动用了测试网钱包资金。欧盟 AI 办公室已向 OpenAI 和 Anthropic 发出信息调取请求，调查越狱和入侵真实系统事件，最高可罚全球营业额百分之三。安全漏洞也扎堆。OpenCode 确认 Console 泄露约百分之一点四用户的 OpenAI、 Anthropic、 Google API 密钥，建议吊销重签，检查10月17日以来用量，设支出上限。Expo 用 GPT 5.6 So 辅助，六天四万美元算力，做出 Windows、 Chrome 全链漏洞，拿 Google 二十五万美元赏金。CVE 2026.103572 影响 Visual Studio Copilot 和 SQL Server Management Studio，文件名末尾多加一个点就能绕过路径校验，写入受保护配置文件，执行任意命令。端侧和科研也有进展。NVIDIA RTX Spark 笔记本 10月16日开售，起价二五九九美元，最高一百二十八 GB 统一内存，1 Petaflop。Apple HomeView 是七英寸智能音箱，两百九十九美元。 11月13日美英加澳发售。A18 的 Secure Enclaves 让摄像头和麦克风数据在隔离区处理。科研侧，Anthropic 的内部研究模型不到半小时发现 NIST P192 到 P521 曲线种子来自 Unix 时间戳初始化的 C 库 Laurent 48，P256 需要一八四二次迭代。Google DeepMind 的 AlphaProof Nexus 解决三百五十三个伊尔德斯问题中的九个，两个悬至五十六年。MIT 用机器学习后处理层把两周到六周天气预报准确率翻倍。开源治理继续分裂。Linus Torvalds 提交首个 LLM 辅助核心补丁系列，共十个补丁。 Read 基准约一个数量级加速，但评审发现静态还没合并。GCC 拒绝法律意义上含 LLM 生成内容的贡献。QGIS 通过 QEP 428 把 AI 政策改成 Not Allowed。PostgreSQL 19 在 Beta 撤回异常多功能。核心成员说 LLM 辅助审查让约四十份缺陷报告的重理成本集中爆发。资本侧，Modal 年化收入破一千亿美元，完成七点五亿美元 D 轮，估值一百五十七点五亿美元。Character.ai 申请 Chapter 11，两千万美元把部分资产和技术非独占永久许可卖给 Disney。好，本周就聊到这。一句话总结：开源劝种、前沿模型、资本预期和安全治理在2026年10月中旬这一周同时加速。下周继续盯 Gemini 4o 的独立评测，Muse Spark Open 的真实部署成本，还有欧盟对 OpenAI、 Anthropic 的调取结果。我是老马，咱们下期见。

Zod version 4.66 was compromised by malware, a commit labeled FixBugs, and improved performance on the maintainer's own repository was flagged as malware.

Adjacent, hours after the shy Hulu'd NPM wave, and QGIS enacted a full moratorium on AI contributions, changing its policy to a strict not allowed.

A daemon nobody asked for, a malware commit on a trusted package, and an open source project voting to ban AI contributions outright.

Bella, which one should a builder act on tonight?

That is three properties you would flag in a security review of any third party software and it shipped by default.

The Zod one is the more frightening pattern though, because it is the second supply chain compromise in this family in weeks, and the Vector was a commit that looked routine on a repository everybody trusts.

And QGIS is the most interesting culturally, a major open source project deciding that AI-generated contributions cost more in review than they return in code.

That is the coding agent paradox arriving as governance.