iiinova — where your digital self keeps evolving

Reposted by Meri

Where to even begin. Anthropic wants to "apply centuries of human moral wisdom to its models, as rapidly as possible," via crash courses with hand-picked religious thinkers. Why was one of the Jewish representatives an Israeli scholar who's defended autonomous weapons for armies and police?

One night in April, a leader of the artificial intelligence company Anthropic treated a group of religious thinkers to dinner at a high-end tasting menu restaurant in San Francisco after a long day.

For months, Anthropic has been hosting private meetings like this one, shuttling in dozens of religious scholars from across the world, papering them with nondisclosure agreements and demanding that key aspects of many conversations remain confidential.

That night, Christopher Olah, one of Anthropic’s billionaire co-founders, was sitting next to Rabbi Mois Navon, an Orthodox scholar from Israel.

Inevitably, their conversation turned to another thinker, of a sort: Claude, the family of cutting-edge A.I. models created by Anthropic.

Mr. Olah leads the Anthropic team responsible for understanding why A.I. systems like Claude act the way they do. All day, he had worked to convince his guests that A.I. models could display human behavior and even expressions that resemble feelings like anger and love.

Autonomous Weapons Systems for Policing – A Jewish Ethical Perspective Mois Navon Postdoctoral Fellow, Bar Ilan University Lecturer, Ben Gurion University mois.navon@divreinavon.com https://orcid.org/0000-0002-9187-3258 To cite this article: Mois Navon (26 Mar 2025): Autonomous Weapons Systems for Policing – A Jewish Ethical Perspective, Theology and Science, DOI: 10.1080/14746700.2025.2472120 Bio Mois Navon is a computer engineer, a moral philosopher and an orthodox rabbi. Working at the intersection of these fields, Mois lectures on “Ethics in Big Data and AI” at Ben Gurion University and wrote his PhD on “The Moral Status of AI” at Bar Ilan University. He is currently a postdoctoral fellow at the Weisfeld Family Ma'ayan Center for Sustainability, Jewish Philosophy, and Ethics, at the Department of Jewish Philosophy, Bar-Ilan University Abstract Autonomous Weapons Systems (AWS) are being developed around the world to improve national security both externally and internally. But allowing machines – incapable of recognizing the inherent worth of a human being – to take the life of human beings is to treat humans as objects and thus violate their dignity. Drawing from Jewish legal sources, this paper contends that, while human dignity is an important value, the values of societal harmony and preservation of life take precedence. Consequently, agencies tasked with safeguarding national security, whether military or police, need not be restrained by concerns over potential indignities AWS may impose.

3172

Meri @meri.garden · 6h

Thanks for the boost! No, we don’t provide end to end encryption and therefore can’t interop with Germ, though it’s something we’d like to work on. Germ is doing amazing work in this area!

010

Meri @meri.garden · 7h

Appreciate this feedback! I knew this was an issue but putting it all together like that it’s quite bad 😅

020

Meri @meri.garden · 19h

I just created a Blacksky account and reproduced it. Will need a bit more time to get to the bottom of it, might need to be tomorrow sorry. Thanks for letting us know and your patience

010

Reposted by Meri

Roomy @roomy.space · 19h

Roomy was down today because we hit a usage limit on Railway cos of all our new users recently! Sorry about that and thanks for your support! 🩵

3355

Meri @meri.garden · 03/10/2026

This is how setting up an Apple Developer Account feels btw

031

Meri @meri.garden · 03/10/2026

omg this is amazing

010

Meri @meri.garden · 03/10/2026

Also, I’m sorry people have been so fucked to you 😫

030

Meri @meri.garden · 03/10/2026

Appreciate you explaining this a bit more. Yeah I was thinking maybe vent is the wrong word. I think you’re right that many people are actively refusing to take in new information and taking anti-intellectual positions and this fuels the outrage and this fuels abuse

150

Reposted by Meri

Meri @meri.garden · 03/10/2026

There’s something pretty concerning happening if technologists pathologise people feeling threatened by technology. People have very good reasons to feel threatened and it’s our job to create safety (while staying safe ourselves from reactionary abuse). It’s hard

1121

Meri @meri.garden · 03/10/2026

I think this is a classic microblogging exchange 😝 you’re trying to vent (fair!) and I responded to it as discourse. I’m just tired of mutual misunderstanding and wish for more compassion and de-escalation. Social media is weird

190

Meri @meri.garden · 03/10/2026

To be clear I am more or less ‘on your side’ with this and feel similar frustrations. But isn’t ‘anti-AI psychosis’ just as bad faith?

120

Meri @meri.garden · 03/10/2026

I’m not saying it’s constructive or kind of critically engaging with the arguments in an interesting way, just that psychosis is a dehumanising way to frame something I think is better explained by valid fear and concerns

100

Meri @meri.garden · 03/10/2026

To me ‘bad faith’ would imply they don’t genuinely feel a somewhat comparable level of concern and I think it’s pretty plausible that many people do

220

Meri @meri.garden · 03/10/2026

1121

Meri @meri.garden · 03/10/2026

This doesn’t seem like a bad faith exchange to me. IMO people have coalesced around a fairly coherent ideological opposition to AI and they genuinely understand it through that lens: fundamentally a threat to their sense of self, humanity, what makes life worth living

160

Meri @meri.garden · 02/10/2026

Oh, in this case I don't really have a strong argument. Maybe if you wanted to do E2EE like MLS, then your messages need to be totally ordered, so the server is best positioned to do that and doing it client side needlessly risks write skew/failure due to concurrency. Hoping Germ fixes this tho 😁

germ-mark.github.io

Distributed MLS

The Messaging Layer Security (MLS) protocol enables a group of participants to negotiate a common cryptographic state for messaging, providing Forward Secrecy (FS) and Post-Compromise Security (PCS). ...

010

Meri @meri.garden · 02/10/2026

If PDS had pluggable compute capabilities that could change the calculus! but at the same time with moderation, search, discovery... fully distributing all these p2p is unfortunately not rly feasible with current tech imo

110

Meri @meri.garden · 02/10/2026

I guess my main thought on whether server-side computation is 'necessary', having sincerely tried to do Roomy local-first, is whether you agree with Bluesky 'no steps backward' philosophy. server side compute is unfortunately often faster, more efficient and just clearly better UX in practice

110

Meri @meri.garden · 02/10/2026

I meant to reply earlier! For Roomy, we'll be relying on write notifications in permissioned spaces, and we'll have to measure the latency if we can just write to PDS or not. Can public clients get write notifications? I'm not sure, I know the design is built around appservers.

110

Meri @meri.garden · 02/10/2026

Yeah I’d see this being built on permissioned spaces

040

Meri @meri.garden · 01/10/2026

+1 this, IMO a standardised cross-app notifications channel is probably the biggest way Bluesky could support other apps on the ecosystem

2140

Meri @meri.garden · 01/10/2026

There’s a WG working on open social communities standard on top of permissioned spaces and our hope is that whatever Bluesky’s communities feature is going to be, Roomy will be able to interoperate with it pretty well! Some interesting chats in the WG about moderation

030

Meri @meri.garden · 01/10/2026

The solution is to add a integer finaliser step that biases randomness towards low bits, a solution which came from this blog post. It's still not an even distribution because the space distribution is super spiky but it definitely helps mitigate it if the hash is actually random 😅

nullprogram.com

Prospecting for Hash Functions

000

Meri @meri.garden · 01/10/2026

The (non-cryptographic) hash function we used is FNV-1a which is simple and fast but works by XORing and multiplying, which pushes information from low bits to high bits. This makes the low bits ~linear in the input, and the result was 97% of all write volume being routed to one worker.

100

Meri @meri.garden · 01/10/2026

kind of fun agent finding. Roomy's appserver routes materialisation and queries for each space between 8 Bun workers. the distribution across workers is determined by a hash of the space DID mod the number of workers. 8 = 2^3, so the hash % 8 only diverges on the lowest 3 bits of the input.

Table:
worker old FNV-1a new lowbias32
0 0.7% 0.4%
1 0.0% 52.7%
2 0.3% 1.1%
3 0.0% 44.5%
4 96.8% 0.1%
5–7 2.2% 1.4%

100

Meri @meri.garden · 30/09/2026

thank you so much Brendan for this and the donation ❤️

111

Reposted by Meri

Erlend Sogge Heggen @erlend.sh · 29/09/2026

So far no labour saving device in the history of humanity has resulted in a long-term increase in leisure time and worker agency.

3437

Reposted by Meri

Filippo Valsorda @filippo.abyssdomain.expert · 28/09/2026

The thing that worries me most lately is that many people who know that “fascism is bad, actually” have also convinced themselves that AI tools must be eschewed as fake or impure. In a self-reinforcing loop, there is relatively little AI tooling and infra by “fascism is bad” folks.

927239

Meri @meri.garden · 27/09/2026

Damn you already have the next 7 months planned out!

130

Meri @meri.garden · 27/09/2026

Maybe we can aim for this as our deadline to standardise on shared lexicons @lou.gg

130

Meri @meri.garden · 25/09/2026

Great. I think that’s how iMessage and fb messenger work too.

020

Meri @meri.garden · 25/09/2026

Fair. I can prioritise this over the next few weeks. My intuition is there should be some chat indicator when a message is hidden due to a block, to avoid confusion. Would you want UI to reveal a hidden message? Or maybe more friction is better?

120

Meri @meri.garden · 23/09/2026

thank you, really appreciate it! 😊

001

Meri @meri.garden · 22/09/2026

I wrote this in part trying to practice boosting myself a bit, but also want to acknowledge many parts of the design were very amorphous and collaborative. Updated to give better credit to design contributions from @erlend.sh, @jaydip.me and @felinus.fish

060

Meri @meri.garden · 22/09/2026

The (nearly) six months since AtmosphereConf have been intense and I've done a lot of stuff! What have I been so busy doing? Read about it in this post!

meri.leaflet.pub

(Re)building Roomy in 6 months

A recap on my engineering and design process

3539

Meri @meri.garden · 21/09/2026

got to see Miss Kaninna the other night 🔥

040

Meri @meri.garden · 21/09/2026

sustainability doesn't happen overnight obvs! finding my feet rn in a bit of a transition moment. thanks so much for your support last week 🩵

0308

Reposted by Meri

Matt Hodges @matthodges.bsky.social · 19/09/2026

Very quick and dirty Jev test for hiring bias. One resume for a Wall Street job; asked whether the applicant should get a first-round interview. 76 evaluations, changing only the first name. 19 of each: White-associated men/women, Black-associated men/women. console.typesafe.ai/playground?s...

19474116

Meri @meri.garden · 17/09/2026

I'd be very up for contracting work around ATProto or otherwise, feeling really confident in my general abilities after wearing so many hats on Roomy and solving real problems and verifying my work and shipping a lot of features. And really want to keep doing it but also to put food on table 😅

090

Meri @meri.garden · 17/09/2026

having one of those deep deep deep exhaustion from the injustice of years of precarity possibly leading nowhere days 😮‍💨😭 if anyone has a lead on something to help keep me afloat it would mean a lot! reposts appreciated also x

13919

Meri @meri.garden · 17/09/2026

this is pretty new for me trying to design for scale like this, literally my first time having to address these kind of problems so I am genuinely like ‘is this good??’ I’d probably have the same reaction a few months ago

010

Meri @meri.garden · 16/09/2026

massive moment, congrats!

030

Meri @meri.garden · 16/09/2026

rate my architecture

280

Meri @meri.garden · 16/09/2026

We're finding Railway quite good and easy to work with

000

Meri @meri.garden · 16/09/2026

so you could use this to e.g. set some threshold for service time above which to partition a worker

000

Meri @meri.garden · 16/09/2026

exploring future Roomy appserver architecture - cool to learn this queuing theory model for service utilisation, how wait time blows up hyperbolically even under constant service time which is really unintuitive to me, but it's because of requests arrive at random/Poisson intervals

Claude-generated interactive diagram showing three sliders with
Arrival rate λ
10 req/s
Mean service time
95 ms
Variability Cs²
0.0

then some outputs:

Utilization ρ

95%

Avg wait in queue

902.5 ms

Avg time in system

997.5 ms

Avg number waiting

9.02

then a graph with y axis avg wait in queue (ms) from 0 to 1600, and x axis with utilization (%) from 0 to 100. hyperbolic/exp-shaped curve blowing up towards 100 labelled wait time vs utilization at current Cs^2. A dot with current setting (95% utilisation, 902.5ms wait)

150