semisol

gm
there is a high correlation between using claude and using the word “shape”
and an inverse correlation of using the proper terms “pattern”, “schema”, “type” or “behavior”
AI is making language stupid

2026-10-26 19:52:18 UTC

- reply

Nope this is directly warranty

2026-10-26 17:39:01 UTC

- reply

That is still a cloud coordinator. You can use the underlying software Nebula yourself

2026-10-26 17:03:33 UTC

- reply

🐈

2026-10-26 17:03:13 UTC

- reply

I will have to check what protocol it uses. But if not supported, will add

2026-10-26 17:02:52 UTC

- reply

😑

2026-10-26 17:02:33 UTC

- reply

2026-10-26 17:02:29 UTC

- reply

2026-10-26 17:02:24 UTC

- reply

2026-10-26 17:02:20 UTC

- reply

good morning as well

2026-10-26 17:02:12 UTC

- reply

Maybe

2026-10-26 17:02:07 UTC

- reply

good morning

2026-10-26 17:01:55 UTC

- reply

Yes but it doesn’t look like morning
🖕☁️

2026-10-26 17:01:12 UTC

- reply

Wireguard

2026-10-26 14:20:56 UTC

- reply

Everyone says this until it happens. Not to mention, they could be hacked even if they have good intentions

2026-10-26 14:18:28 UTC

Tailscale always felt weird to me because their control panel could always decide to add a malicious node to my network and do whatever.
Same with the SSH access. Like, why?!

2026-10-26 14:17:38 UTC

- reply

This is referring to the 21K NFTs that the Nostrich app wants to offer

2026-10-26 14:17:15 UTC

- reply

You should run AI in a sandbox and assume everything will leak even if you use local AI (the agent could randomly decide to upload your data somewhere or prompt injection)
Private usage of AI is hard

2026-10-26 13:15:25 UTC

🤔 good morning I guess

2026-10-26 04:08:13 UTC

- reply

“Spoofing” DKIM is not possible. It means their email provider got hacked, either by their own fault or a 3rd party.
It is e-signatures for email basically

2026-10-26 03:42:25 UTC

It appears many people are trying to reinvent Nostr.land, but more expensive.
We have been operating relays since 2022 :)

2026-10-25 16:45:33 UTC

- reply

The problem is that the clanker is lying a shitton in ways that sound plausible:
- Lottie is a standardized format. And of course, a new format won’t work for clients for a while!
=> It has the benefit that it is like SVG but animated, is much smaller than GIFs, and can be scaled up infinitely
- Of course, migrated packs are signed by someone else. Duh.
- It does not run through Telegram except for imports. Anyone can download a Lottie file or make their own and turn it into a sticker

2026-10-25 16:43:11 UTC

Instead of overpaying for AI generated ostrich JPEGs, sign up for Nostr.land
✅ Trusted @nostr.land NIP-05
✅ Archival for your content
✅ Export your notes at any time
✅ High-speed spam filtered Nostr aggregator
and a bunch more coming soon 👀

2026-10-25 16:38:26 UTC

- reply

https://gruhn.me/blog/2026-08-03/

2026-10-25 16:28:07 UTC

- reply

Their sticker format is pretty good!
It’s basically lottie

2026-10-25 16:20:03 UTC

Fact check: RHXO does NOT stand for this
It is a single-person-led company that has existed since at least 1995, but has no demonstrable work anywhere
#nevent1q…uqdp

2026-10-25 15:06:30 UTC

Good morning Nostr

2026-10-25 14:24:27 UTC

- reply

I say this not because America is bad, but because they are too cheap to not use when your limit runs out

2026-10-25 12:43:28 UTC

- reply

Use cheap chinese tokens

2026-10-25 01:17:01 UTC

- reply

Hi. I apologize for the issue.
This has been fixed

2026-10-23 23:55:33 UTC

- reply

There is no solution. The problem grows the more you try to fix it, which causes more people to try to fix it, which means that it will eventually consume the entire planet like grey goo. The best way is for it to all be deleted and start over.

2026-10-23 23:53:55 UTC

- reply

In trying to solve the problem, you have realized you have become the problem

2026-10-22 19:29:02 UTC

- reply

ASNs or subnets

2026-10-22 18:55:30 UTC

- reply

Could you send me some samples

2026-10-22 18:45:58 UTC

- reply

And I guess under this categorization, Nostr.land is a big evil data harvester analyzing your notes to filter out spam and aggregate them so more people can see it

2026-10-22 18:37:33 UTC

- reply

I don’t get how this automatically implies that Nostrich is a data harvesting machine
They have patents on that topic, in general, that’s all

2026-10-22 18:36:45 UTC

- reply

what kind of patterns were you seeing

2026-10-22 18:11:52 UTC

- reply

it’s not like everyone and their dog has not mined the relays thousands of times over collecting every event in existence including the so called secure gift wraps

2026-10-22 13:06:09 UTC

- reply

Rust does not magically eliminate all vulnerabilities, and a good chunk of vulnerabilities are not caused by memory safety issues
(Also, the Go CVE is actually just a duplicate of the eMudhra issues)

2026-10-22 09:38:15 UTC

- reply

I would not, rather things like npm and pypi. One has well-defined release procedures

2026-10-21 22:09:31 UTC

- reply

I think that package managers may have been a mistake

2026-10-21 19:50:33 UTC

- reply

It’s not really possible to prompt out the slopstyle and it leaks to everything. :/

2026-10-21 16:30:43 UTC

- reply

Nostr devs in their infinite wisdom continue inventing solutions in search of a problem…

2026-10-21 15:29:16 UTC

- reply

the keys were:
4, 56, 66, 115, 155, 156, 230, 237

2026-10-21 15:28:02 UTC

- reply

CA generated private keys incorrectly

2026-10-21 14:44:15 UTC

- reply

I just wish someone made it stop sounding like a Claude distill with the awful jargon, pseudointellectual behavior and such

2026-10-21 14:16:18 UTC

- reply

AI will convert a single sentence into a whole email so that another AI can convert it back into 1 completely different sentence

2026-10-21 13:55:00 UTC

- reply

not self-signed

2026-10-21 13:51:46 UTC

Now you can be your own CA

2026-10-21 06:24:19 UTC

- reply

🫤

2026-10-19 17:35:21 UTC

- reply

Cashu is fully custodial

2026-10-19 16:56:07 UTC

- reply

Only Bitcoin, Bitcoin, Bitcoin, moar Bitcoin, Bitcoin, Bitcoin…..

2026-10-19 16:17:35 UTC

- reply

gm 💭

2026-10-19 16:17:26 UTC

- reply

☀️

2026-10-19 16:17:16 UTC

- reply

Good morning!

2026-10-19 16:16:50 UTC

- reply

2026-10-19 16:16:39 UTC

- reply

good morning! 👀

2026-10-19 16:16:25 UTC

- reply

Gm liminal 🤔

2026-10-19 16:16:15 UTC

- reply

I wonder

2026-10-19 15:00:17 UTC

Good morning Nostr.
👀

2026-10-18 16:54:26 UTC

Not all relay software is made equal

2026-10-18 02:27:28 UTC

- reply

We focus so much on irrelevant risks, and focus so little on relevant risks.
Losing your keys, phishing/scams, malware, locking yourself out. Much more likely than a $5 wrench attack, confiscation, and evil spies.
But note, I’m developing a hardware wallet.
For those that do self custody, I want to have a good solution. But I can’t see the need for anything beyond a smartcard or small USB stick with 2 buttons and a display.
I don’t think the other options have much benefit, except for show-off. Which the market unfortunately wants…
Bitcoiners spend too much time and money talking about self custody and doing a pissing contest of which setup is better, I wonder if it was worth it at all
Also, seedless is likely much better for normies. Advanced users can have it, but it’s not the way for normies that decide to go this route

2026-10-18 02:18:36 UTC

- reply

Worse than self custody however, is spending $300 on a Q or some other shitty overpriced overdesigned hww times 3 to protect $20k. Maybe do something nice with your $900 instead of burning it all on an overpriced overcomplicated setup
You’re protecting $20k against what exactly? “Oh but bitcoin will be $1M” do it when it actually is
If there was something somewhat better than an exchange, then that would be amazing:
1. Doesn’t have much business except to hold coins at a fixed % rate per year
2. More “security hardened” with 24h delay + phishing-resistant passkeys and no SMS 2FA
I don’t see reasons for self custody for the average normie. If you can get used to it, then it is a good choice for some cases, but it has its risks

2026-10-18 02:09:02 UTC

- reply

It’s not a solution for the majority of people.

2026-10-17 22:28:35 UTC

➡️

2026-10-17 21:34:07 UTC

👀

2026-10-17 21:26:20 UTC

- reply

I run my LN node for convenience. The convenience being I can do all sorts of weird things with it.
Many things in Bitcoin are not suited for the average person and have little benefit for them

2026-10-17 18:09:42 UTC

- reply

What are the problems

2026-10-17 18:08:34 UTC

- reply

Any sufficiently deep AI prompter’s setup contains an ad hoc, informally-specified, bug-ridden, and slow implementation of half of a modern programming language, labeled as “specs”

2026-10-17 17:23:50 UTC

🔶 M

2026-10-17 15:00:02 UTC

- reply

Please stop calling it the blake2b fork and call it BLV (Bitcoin Luke’s Vision) instead as it is an insult to the hash function

2026-10-16 15:20:50 UTC

- reply

The heat death of the universe

2026-10-16 13:17:43 UTC

- reply

Nostr DMs: so secure even you can’t read them

2026-10-16 13:02:54 UTC

Good morning Nostr

2026-10-15 22:10:41 UTC

- reply

It is a good way to check if a solution that is better exists. But it isn’t very good at inventing new ones.

2026-10-15 18:35:21 UTC

- reply

I was wondering if it was that one, though I just did a search for it in general
Also the note ref is corrupted

2026-10-15 18:34:51 UTC

- reply

I could easily find a description of this method before you replied:
https://github.com/DavidPal/discrete-distribution/blob/master/doc/algorithm.tex
Again, AI is mostly useful as a search engine compared to inventing new solutions.

2026-10-15 18:30:23 UTC

- reply

which one, exactly?

2026-10-15 17:46:16 UTC

- reply

It’s the exact opposite for me.
It’s mostly a regurgitation of already well-known designs, and rarely does it ever come up with something “new”. For the rare few times it “does”, it’s just my own input echoed back at me…
The world’s most expensive rubber duck, now with idea-stealing as a feature

2026-10-15 17:34:59 UTC

- reply

The problem with these models, in general, is that they are reward-maximizing.
They have no real intelligence, capability to engineer, or creativity to speak of.
This makes them great for verifiable tasks where you can get away with “guided brute force”, but for anything else it ends up being not very good.
I think the recent RL on “one-shotting apps” may have also made them worse at anything that isn’t one-shotting apps

2026-10-15 17:27:14 UTC

- reply

Why use automatic routing? There are a lot of factors like prompt caching and such.

2026-10-15 17:26:20 UTC

- reply

(Please look up what I’m referencing :)

2026-10-15 17:19:42 UTC

- reply

Nostr.land fixes this

2026-10-15 17:13:16 UTC

- reply

you’re holding it wrong

2026-10-15 17:12:39 UTC

- reply

Yes

2026-10-15 12:04:16 UTC

- reply

I can say that 99% of the things I do with a CLI on openSUSE (package installation) now can be done with a GUI as well.
Otherwise it’s mostly set and forget

2026-10-14 23:28:04 UTC

- reply

Linux kernel overreports vulns, though

2026-10-14 20:26:52 UTC

- reply

What a terrible day to have eyes

2026-10-14 20:25:03 UTC

Good day Nostriches
🤐

2026-10-14 02:28:32 UTC

- reply

I really hate the hype around these “system one” models. There is nothing innovative.
It’s just a promptable classifier. And the implementation is basically crappy logprobs.
Anyway, I was mostly referring to programs that generate programs rather than AI/ML

2026-10-14 00:37:50 UTC

- reply

what instead baffles me is that we do not have enough nail-hammering machines
these are more reliable than the LLM, have to be proven correct once, and can help *any* workflow
I feel like in software development, code generation, and abstractions are nowhere close to being decently exploited. and that is very disappointing!
though, software is much different than buildings at the same time. the blueprint *is* the building, and you’d mostly be outsourcing the design of individual components

2026-10-13 23:09:09 UTC

- reply

This is a heuristic

2026-10-13 23:03:14 UTC

- reply

AI is more “outsource it to 3rd world country” rather than “competent programmer”. It’s the exact same.
You can do all the tests and adversarial reviews and loops and subagents you want in the world, but any non-trivial property of software is uncomputable.

2026-10-13 22:56:08 UTC

- reply

Using AI in some cases is also outright deception of your users. Think an architect rubber-stamping designs.
Then, why do people trust you and use your product?

2026-10-13 22:52:08 UTC

- reply

Forgot 5 (implied by 4): Not innovative
——
The best code is the code you write, and then delete. As with science, negative results are often more important than the positive ones.
If your goal is to research the market by burning user trust, shutting down features constantly, and providing a subpar experience. AI assisted coding is for you!
——
Also: using AI as a heuristic is not bad. It can give you junk, and it should not replace your own information, but it can help you prioritize/reach the result faster depending on the use case.

2026-10-13 22:48:25 UTC

- reply

The only components where I keep AI code is where all of the following apply
1. the gains of writing it myself is not worth the cost
2. it is self-contained and can be replaced
3. it is not my product
4. it is hard to get wrong

2026-10-13 22:44:08 UTC

- reply

If implementation is not the product, what is?
[I had a huge wall of text here but deleted it] TL;DR Software is selling knowledge and is similar to science. Your code is a 1:1 representation of that knowledge.
And you learn from building software
With AI you slow yourself down by attempt to interpret incoherent, reward-maximizing slop. You don’t learn from your mistakes, or what works, and you are doomed to reinvent the wheel every time.
I use it for some things, but it is not a main part of my development flow. Prevention rather than remediation pays off big
If your goal is just to ship a shitty MVP to see if anyone will buy it, then sure, use AI. But this is the YC startup mindset, and is not what I consider serious software.

2026-10-13 17:24:22 UTC

- reply

AI does not eliminate the problem that the maintenance effort for a codebase scales superlinearly compared to its size
You can either have less code and higher quality, or more code and lower quality

2026-10-13 16:16:11 UTC

- reply

in exchange you get horendous spaghetti code