フィリイプ
banner
fillip.pro
フィリイプ
@fillip.pro
Strategic technologist and author focused on safety-by-design human experience, sustainability, and privacy-enhancing technology at @omnifi.foundation.
If you define a standard ABI in the host, then you can support interpreted languages. JS—for example—could use QuickJS, and Python could use CPython enabling the imports and exports to be generated against the defined ABI.
August 7, 2025 at 6:10 PM
I’ve written quite a few systems that support WebAssembly plugins and I’ve never used a plugin library.

It always felt easier to just write the plugin code manually. Over the last 12-months it’s just gotten easier.

If you want a fast loop, I’d recommend that direct approach these days.
August 7, 2025 at 2:25 PM
I spent an absurd amount of time trying to find an adaptive solution to this that wouldn’t require preprocessing.

Hell of a rabbit hole to go down.
August 1, 2025 at 7:23 PM
I call back at my convenience.

No bombardments with messages, and I only use the web versions of tools if possible, including Google Meet, Teams, and Slack, so no notifications there.

Essentially nothing breaks my focus. I choose when to be distracted.
July 5, 2025 at 12:08 PM
I’ve scheduled and contextual focus modes on all devices, and rarely have notifications enabled on anything.

Even phone calls don’t come through if it’s during work time, and otherwise if the number isn’t in a group in my address book, of which there is currently only one person.
July 5, 2025 at 12:08 PM
Reposted by フィリイプ
Do you think that Europe should bargain away its digital sovereignty to appease Trump and the broligarchy? Strong majorities in Germany, France, and Spain are against that (YouGov).

@coricrider.com and I have a better plan:
www.politico.eu/article/digi...
Digital sovereignty can’t be bargained away
The European Commission has tools, public support and a mandate to act on Big Tech. Trading that away for short-term calm would be a costly mistake.
www.politico.eu
July 3, 2025 at 7:00 AM
Reposted by フィリイプ
it feels important to keep saying: SPA as an app framework default (and not an opt-in feature) was a mistake
July 1, 2025 at 4:08 PM
The promise is easy. Over a year of prototyping gives me confidence that the destination is feasible.

Shared protocols is critical to making this work, as is a very modular, open framework for building the worlds.
July 1, 2025 at 7:04 PM
Enjoying building tools from scratch in an effort to reshape how the digital tech industry develops.

rai.onl is another pillar forming a set of foundational tools that I hope provides more responsible alternatives in the future.
Rai - Responsible intelligence without compromise
Local-first, enterprise ready, sovereign AI. Responsible intelligence without compromising safety or privacy.
rai.onl
July 1, 2025 at 6:42 PM
Projects are in Git repositories with CI for building and CD for deploying with separate workflows for each.

When deployed to registries (e.g. crates, NPM, JSR, Harbor) they are tagged and can be picked up when versions are bumped.
July 1, 2025 at 8:32 AM
Every project has a composer. The composer program follows these steps:

- ensures resources are available
- copies itself over SSH
- disables SSH access
- provisions workloads (e.g containers)

100% automation to avoid repeating mistakes.
July 1, 2025 at 8:32 AM
I’ve been slow burning a modular and composable AI agent toolkit for WebAssembly as an alternative to some well used Python projects.

Anything in that space is interesting.
June 26, 2025 at 1:04 PM
Researchers and engineers are not meant to exist in the future techbros and C-level is envisaging.

Just cheap agents automatically churning out homogenous bloat.
June 23, 2025 at 8:10 AM
Would help if folk stopped asking product teams, “where’s the AI?” rather than focusing on the features and capabilities.
June 19, 2025 at 4:41 PM
When I work on product strategy or tech projects I also avoid the term. It’s largely irrelevant.

I’ve taught AI development to software engineers for 15-years, as I think will inevitably be as fundamental and invisible as applying databases or forms when needed.
June 19, 2025 at 4:40 PM
Another: stochastic neural network, with convolutional and recurrent neural network layers. Fundamentally a family of deep neural network models for prediction.
June 19, 2025 at 4:27 PM
A bot that replies to every “AI is bad” post with a legitimate use case might save me from having to read these ill informed takes.

For now, a single legitimate use: tropical cyclone prediction.

deepmind.google/discover/blo...
How we're supporting better tropical cyclone prediction with AI
We’re launching Weather Lab, featuring our experimental cyclone predictions, and we’re partnering with the U.S. National Hurricane Center to support their forecasts and warnings this cyclone season.
deepmind.google
June 19, 2025 at 4:18 PM
How are we defining AI? Neural networks, deep learning, etc. We’ve been applying it to vehicle safety, disease prevention, and the climate emergency for over a decade, invisibly.

Your novelty consumer app is not the most useful or valuable use of AI, certainly.
June 19, 2025 at 4:00 PM
Google threatens to pull apps from Play if they don’t receive updates within 12 months.

Building a stable service and having a feature-complete app is fundamentally shaped to be a bad thing.

Yes, I just had an account with several apps closed. They won’t reopen it once they deactivate.
June 18, 2025 at 1:46 PM
In September I’ll be giving a special talk at the @mydataorg.bsky.social conference on zero-compromise privacy-preserving AI, using private data spaces, able to run anywhere.

No cloud, no surveillance, no theft.
June 18, 2025 at 11:59 AM
Reposted by フィリイプ
Bluesky CEO @jay.bsky.team is right, we need to stop thinking of social media as a zero-sum game that platforms are “winning” or “losing.”

Why not measure friendships, or genuine connection? We need to rethink what we value in our online tools for socializing.
Bluesky is most definitely alive and kicking
CEO Jay Graber says the days of winner-take-all social networking are over. Thank heavens she’s right.
www.fastcompany.com
June 17, 2025 at 8:09 PM
As a teenager I’d have penpals all over the world. I keep thinking I should do this again.

The slow pace and wait is as enjoyable as receiving the letter.

The consideration taken in responding is meditative.

Maybe I should just add a significant delay to my email server. 😁
June 18, 2025 at 9:28 AM
For offline analysis it could be a fun time to try out SurrealDB instead. I replaced a few neo4j use cases with it.
June 17, 2025 at 5:40 AM
According to information from ~5 years ago some providers were hosting 7 million non-federated accounts, so it definitely should be capable at this point, considering the significant improvements made to the homeservers since then.
June 17, 2025 at 5:35 AM
It largely depends on the provider and the implementation. Some struggle with large numbers, especially when federated.

AFAIK some Matrix implementation providers have hit 2 million accounts.
June 17, 2025 at 5:26 AM