semisol

👨‍💻 software developer
📨 nostr.land relay

all opinions are my own.

Public Key

npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkj

NIP-05 Address

Profile Code

nprofile1qqs99d9qw67th0wr5xh05de4s9k0wjvnkxudkgptq8yg83vtulad30gpzpmhxue69uhkummnw3ezumrpdejqzrthwden5te0dehhxtnvdakq37jets

Publishing to

nostr.land

nos.lol

nostr.mom

Author Public Key

npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkj

Show more details

Last Notes

2026-07-28 03:09:02 UTC

On Nostr, you are free to do anything, except to criticise Bitcoin

2026-07-27 22:52:55 UTC

- reply

This will only accelerate the failure of Bitcoin if it is happening (likely yes) as it rapidly shifts from actually useable money to corporate hoarding asset.

2026-07-25 23:14:46 UTC

I cannot find a single person outside of Nostr who wants to actually use AI vibeslop vomit.
Maybe because they care that their software works, and is not blasting them with AI overexplanation syndrome?

2026-07-25 20:58:12 UTC

- reply

I am not sure. Something has certainly happened. Probably the AI generated slop takeover.

2026-07-25 18:04:55 UTC

- reply

s/USB-C/any connector/

2026-07-25 00:57:30 UTC

- reply

yeah because it costs like 5x more on API and is slow as a snail and can’t be used with codex

2026-07-24 19:21:49 UTC

- reply

Such a general security product is a bad idea. You increase the attack surface a ton, and get little benefit for many of the things.
For example, software OTP is secure enough and you can always use HW keys. An encrypted storage can be a dedicated disk that performs better.

2026-07-24 16:33:45 UTC

- reply

You can do something much better yourself and much more secure, at a lower cost 🤔

2026-07-24 16:31:34 UTC

- reply

Yes. And honestly, both that and the Passport Prime are a waste of money.

2026-07-24 16:26:48 UTC

If you are concerned about the BIP-110 fork and your LN node: don’t
As long as you do not care about the fork-coins (in either side), nothing changes for you. You need to take 0 action.

2026-07-24 12:29:12 UTC

- reply

Good because Pi doesn’t come with them by default

2026-07-24 11:59:50 UTC

- reply

NVK said “general purpose computers are not safe for storing keys” but apparently that is not the case when Coinkite does it.

2026-07-22 19:36:50 UTC

- reply

Why use omp when you can just pi

2026-07-22 19:36:28 UTC

- reply

1. More efficient => more usage
(Higher cache hit rate, less tokens)
2. Faster and more reliable at doing tasks
(no need for subagents!)
3. Extensible to hell
4. Works with basically any AI provider except Anthropic because their policies

2026-07-22 18:10:35 UTC

There are two types of AI users:
- Those who use Pi
- Those who don’t use Pi yet, but won’t switch back once they do

2026-07-21 17:30:17 UTC

- reply

Dump a whole bunch of it on the public, keep the rest, make profit on inflated API.
Or maybe they are looking for a way to get their other portfolio companies to funnel more revenue at the AI providers at the same usage.

2026-07-21 17:29:17 UTC

- reply

Probably both

2026-07-21 16:37:05 UTC

It seems that we are now in the part of the bubble where AI companies try to profit.
In a year, the price of models has gone up 3x for OpenAI models, for example. This has especially impacted nano and mini models.
Same with Gemini, 2.5 => 3.1 going up in price 5x for output tokens.

2026-07-21 16:26:40 UTC

- reply

Took a good 30m or so to write.
I have also included some additional details here:
#nevent1q…06wj

2026-07-21 16:09:50 UTC

- reply

Making it open would not have helped with interoperability, unfortunately. Jam was not designed for that, which both Nests and Corny Chat were based on.
It would have been good to fully open up, but it’s gone now. There was no attempts though to “lock it down”, and it remained compatible with the base Jam software.
The goal of the rewrite was to provide specifications anyone could implement to get their own fully compatible audio room. That got lost after the takeover, somehow.
I attribute the downfall more to Primal and fast-moving specs, more than “scrutinizing what is a true client”.

2026-07-21 16:04:17 UTC

- reply

I could not bring myself to work on the Nests rewrite while being exhausted, under public pressure, and other concerns. The unilateral decisions about a project I contributed to so much did not help.
I forgot to mention, we held a bounty pool to be paid out to any contributors to the ground-up rewrite, that was separated.
I transferred any custodied sats for this bounty I might have to Derek when I left.
If I remember correctly, we also compensated both TC and Karnage out of pocket from Nostr Plebs for their work. It was however unfortunate to see Karnage’s design work go to waste..
——
I do less Nostr development now. Mostly because the AI slop takeover, and the focus on quality going downhill. I have tried to prevent that to some extent, but I have failed.
(Especially on the NIPs, where my access was revoked)
But, it is great to be able to work with amazing people like @npub1r0r…q9ac and the @npub137c…k37w.
I try to do the right thing. Sometimes I make mistakes.

2026-07-21 15:39:15 UTC

- reply

It was more complex and weirder than that.
Derek started hosting Nostr Nests independently based on Jam software. They made some minor modifications early on, but got bigger.
With our existing work together on Nostr Plebs, I agreed to help.
With there UI leaving a lot to be desired, me, @npub1jdg…2mys and @npub1r0r…q9ac worked on a UI redesign. Most of the core remained the same, with mostly some UI changes.
I have been clear that Nests was based on Jam, and I have never tried to obscure that.
Anyway, at this point, it was decided that a complete rewrite, fully independent from Jam was a good idea.
There were too many issues with large rooms on Jam. Also, it was built with a different identity system than Nostr, which made it hard to integrate.
The main reason that we, together, decided to not make it open-contribution is due to the rewrite. It would be fully open source and open contribution, and more suited than that.
(Looking back, I’m not sure if that was a good choice or not. The code depended a lot on internal details to work.)
Anyway, I did not have the motivation to work on the rewrite due to burn-out and personal reasons. This was not helped by the fact that our informal agreement with Derek that it was a shared project fell through, of sorts.
Derek had made it clear that he treated it mostly as his own project including making unilateral decisions, and I felt like that was disrespectful to the work of me and others.
Around that point I left and took a break from Nostr.
I did not track what happened fully next, but I believe it was replaced by a rewrite by Kieran and then proceeded to become inactive.
——
If it was ever unclear that Nests was based on Jam, I apologize for that.
There was never a goal to gatekeep contributions, but more that it was simply not suitable for it due to it being designed with so many internal components.
The situation was more complex than this person was solely responsible about it”.
Even though I have mixed feelings about it, I will not share my personal interactions with Derek in depth out of respect for privacy.

2026-07-21 07:10:46 UTC

gm nostriches 🤔

2026-07-20 16:43:56 UTC

- reply

Example
https://mempool.space/tx/84c22d5a76d923e32819e45a66ba20ff7c467ab1ec94592faec476e0a663d97a

2026-07-20 16:42:44 UTC

- reply

Because mempool.space identifies them for you

2026-07-20 16:41:40 UTC

- reply

Text file as a subscription

2026-07-20 12:20:08 UTC

- reply

Unfortunately, especially for 2, you need to pad your profile with fake activity so that people notice you more.

2026-07-19 23:41:04 UTC

- reply

I feel like it falls into these broad categories.
1. Sharing ideas and finding people.
2. Commerce.
3. Getting attention for self-worth / validation.
4. Manipulating people to achieve your own goals.
I feel like the influencers fall into 3 or 4, and vibe coders + “we are going to make it” people into 3. I personally feel like 1 and 2.

2026-07-19 19:25:57 UTC

- reply

It’s a website isn’t it

2026-07-19 16:35:15 UTC

- reply

It’s a spambot

2026-07-19 02:01:07 UTC

- reply

Though, this would fall under incitement

2026-07-19 01:59:13 UTC

AI has worsened the problem of armchair experts.
The only people that benefit from you publishing your unverified and crappy slop of any form, whether it be text, code, images or video, are the AI companies squeezing you for $$$
Please stop polluting the internet with garbage that is not even worth the fraction of a cent worth of disk space it occupies.
#nevent1q…7eev

2026-07-19 01:50:09 UTC

It appears now on Twitter, a random idiot with an AI tool has gained traction by writing an article longer than STM32 reference manuals about a BIP-110 vulnerability.
However, the vulnerability was completely hallucinated
(this post is not in favor of/against BIP-110)

2026-07-17 19:12:41 UTC

- reply

And I think LN is going to converge to marginal cost at some point.

2026-07-17 18:36:12 UTC

- reply

This already happens by fee prioritization. New nodes are not ranked solely by reputation/success rate but by cost.
And many nodes only keep forwarding histories for a few hours. They go stale very fast.

2026-07-17 10:15:05 UTC

- reply

Less tokens used, better results, and then you can instantly swap token providers ;)

2026-07-17 05:54:33 UTC

- reply

Switch from Codex to Pi to improve your life

2026-07-16 01:08:30 UTC

- reply

The best way to solve this is to keep the relay policy, to indicate that this is not welcome.
This does not eliminate the policy, however it makes it too uncertain for VC investors.
If anything, they should use witness data, because witness data can be discarded without losing 99% of the benefits of a full history.

2026-07-16 00:27:10 UTC

- reply

What if instead of misinterpretable hand-wavey specs, we had concrete implementations in code of the core functionality? We might even package this up into an easy to use thing, which we could call a library.
But the UI is still very problematic. So we also make a base UI. Then, why bother with people reimplementing their own thing all the time? We add 99% of the functions everyone needs into the base UI.
You have built a client

2026-07-15 14:02:18 UTC

Instead of more clients/features/infra/whatever, there needs to be more things developed long-term and focused on stability.
The state of Nostr usability has barely improved pre and post AI slop, if anything it got worse.

2026-07-15 13:59:14 UTC

Good morning

2026-07-15 12:02:39 UTC

- reply

Well, I had designed a swap protocol that looked exactly like a normal payment using Taproot except in the case of the hash-lock path.

2026-07-15 12:01:28 UTC

- reply

This however requires multiple UTXO spends. If anything, Taproot prevents the UTXO bloat that this would cause.
There’s also the option of paying to bare multisigs.

2026-07-15 11:59:54 UTC

- reply

What people forget is that Taproot did not cause this. Data can already be stored in standard P2WSH using OP_POP.

2026-07-14 18:30:14 UTC

I will not be running BIP-110 or a similar softfork. Such a fork solves nothing.
However, the behavior of Bitcoin Core is also harmful to Bitcoin.
The OP_RETURN change had no technical basis and had other problems:
- People will not pay 4x more to use OP_RETURN. The “cheaper at small sizes” argument was fixed by Taproot annex already.
- SegWit data can be fully pruned (or even not downloaded) while being able to verify supply. OP_RETURN can’t.
- Both SegWit and OP_RETURN do not inherently pollute the UTXO set. It is due to the specific implementation of Ordinals.

2026-07-14 16:31:05 UTC

Why do people hand over full access to their computer to an LLM?
The vendors are logging everything and these models are unpredictable. Your information will be leaked or used one way or the other.

2026-07-14 15:48:20 UTC

- reply

gm!

2026-07-12 22:58:10 UTC

- reply

This is an intentional design decision in SegWit.
As it is not part of the “main block” data and is segregated, it allowed increasing the effective block size without breaking old clients.

2026-07-12 16:43:36 UTC

- reply

Witness data is more prunable than OP_RETURN. People have been sold a lie.
The witness data is not included in the main txid, so it can be thrown away COMPLETELY while preserving verifiability of the supply. It is what would be pruned by a theoretical “non-archival node”
You cannot delete an OP_RETURN without breaking the TXID hash and by extension breaking verifiability. So you must always store the data even if you prune it.
Not that there is any incentive to use it, anyway, since it costs 4x more per byte.

2026-07-12 15:09:30 UTC

2026-07-12 13:58:42 UTC

- reply

It was me that said that, but @npub1m3x…a5sf probably did too as well.
I think Luke has gone completely off the rails. The CSAM argument is not a good one as it will always be possible
However, Core merging a change no one wanted, against common opinion, is a bad idea.
It was always possible to have large OP_RETURNs or embed data. This does not mean we should encourage it or say it is welcome (such as via relay policy), though.

2026-07-11 05:59:46 UTC

It seems that part of GPT 5.6’s token efficiency came from reducing the required “thinking budget” for high/xhigh.
So, xhigh is no longer truly xhigh.

2026-07-10 01:33:18 UTC

- reply

It seems Sol is being rolled out to Pro first, in Codex?

2026-07-10 01:15:37 UTC

- reply

Initial test is a bit interesting. It is very fixated on instructions, which is good? But also amplifies biases
I gave it a definition “UP TO 5 players MAY be cheating” and it tried to satisfy 5.
I had an example cheating method listed (not the one in this case) and it also got fixated on that and misidentified players.
It is very hard to identify that method from data.
Re-running with a debiased prompt.

2026-07-10 00:36:18 UTC

- reply

I did not use 5.3 too much so I cannot comment. My main test now, is to see ML training performance.
With 5.4 I made it modify an LM to a different architecture and purpose. It needed tons of hand-holding and sucked.
With 5.5 I gave it a bit simpler task, where it could use more off-the-shelf solutions, but model stability was still hard.
One thing I noticed is that 5.4 struggled figuring out an environment quirk, while 5.5 easily solved it.
I’ll test 5.6 on the same task I gave to 5.5 soon, and maybe after that the 5.4 task.

2026-07-10 00:32:17 UTC

- reply

Did you use the 5.3/5.4 also? I have noticed 5.6 goes deeper on even basic inquiries with web search.

2026-07-10 00:08:12 UTC

- reply

@npub1yvu…v74p @npub1s02…fxtw @npub1aqa…237n @npub1r0r…q9ac
Curious about your experiences as well, especially with the older models

2026-07-10 00:05:55 UTC

I’d recommend trying OpenAI models. The flexibility with your own harness is worth it.
5.3 Codex was the first time they had a well-polished coding model, imo. But it always needed high/xhigh.
5.4 was much better on a lot of fronts:
- Better usability in terms of personality.
- Much better technical capability than 5.3 on hard tasks
- Can do a lot on high/medium, xhigh not needed
and 5.4 mini can replace 5.3 Codex easily.
5.5 fixed a lot of the “early stopping”, and had better depth and technical capability.
The main advancement was token efficiency and medium/low(!) working for 90% of tasks.
5.6 is much more in depth with tasks, works well on low reasoning levels, and seems to have significantly better capabilities.
#nevent1q…ddjm

2026-07-09 23:53:07 UTC

5.6 feels much better than 5.5.
It goes into further depth on a task, and has very little “early stopping”.

2026-07-09 21:10:33 UTC

Did damus relay die?

2026-07-04 19:43:41 UTC

- reply

It blasts to everywhere.
The relay list is updated pretty regularly for large relays, and sometimes small ones.

2026-07-04 19:22:33 UTC

- reply

nitpick: BIP 113 was not relevant for LN, as it uses block-timestamps instead of real-world “time” :)

2026-07-04 00:25:21 UTC

@npub1xts…kk5s Could I get a dump of the Damus relay, along with the last dump from the last nuke?

2026-07-03 17:03:52 UTC

- reply

It is not related to your GM, but glad that helped

2026-07-03 16:57:37 UTC

- reply

Pondering the missing 2nd operator

2026-07-02 13:19:50 UTC

gm nostriches

2026-07-02 12:40:23 UTC

- reply

The great thing is that, the import/export will complete faster than you can get up from your chair.

2026-07-02 12:11:22 UTC

- reply

No, it uses other relays as well

2026-07-01 22:30:27 UTC

Unfortunate to see, but oh well.
If you want to save your notes, sign up to Nostr.land to import your notes with a click and 0 hassle.
It is the longest operating paid relay, having done so for over 3 years.
https://nostr.land/subscribe
#nevent1q…73ke

2026-07-01 15:58:03 UTC

gm 👀 👀

2026-06-30 13:15:26 UTC

Many ecash projects are solutions in search of nonexistent problems

2026-06-29 11:47:55 UTC

gm nostriches 👀

2026-06-28 21:46:00 UTC

- reply

This is a great reason to use NIP-44 v3

2026-06-28 21:20:13 UTC

- reply

On the extension side, yes. Same can be done on the app side to autodetect the version.
For DMs, there is a designated tag to the DM relay list so that it is only used for when the recipient supports it.

2026-06-28 15:54:52 UTC

- reply

Can’t. Too busy right now, even if I let an LLM barf up a PR.
However, if you’d like to provide a more secure encryption UX for users, you can point your LLM to the repo. It is more than enough.

2026-06-03 16:30:46 UTC

The Pi-pilling will continue until everyone learns the one true way

2026-03-03 22:23:39 UTC

- reply

fine tuning for actually useful tasks than a slop machine

2026-02-14 15:53:07 UTC

I am increasingly getting fed up full black UIs. They feel lifeless and dead

2026-02-12 14:02:22 UTC

- reply

nostr.land will soon have optimized support for XL events

2026-01-28 21:21:09 UTC

this applies to here as well…
#note12dx…xhns

2026-01-17 18:00:20 UTC

- reply

Once you add it, the app should also broadcast the current list.

2026-01-16 17:03:47 UTC

- reply

Nostr.land aggregates notes from many relays and pushes your posts to many relays. It’s a paid relay but it should speed up your experience a lot and increase your visibility.

2026-01-09 17:25:10 UTC

when did nostr start using the world’s easiest to manipulate algorithm to mass label everyone based on if influencers follow them or mass-report them due to a grudge

2025-10-21 15:20:30 UTC

- reply

fact: many people that say don’t trust, verify do in fact trust marketing claims

2025-10-21 03:31:32 UTC

- reply

Coldcard is not a solid HWW at all. I work on secure element design and several other people that also do agree.
Don’t use an exchange. SeedSigner so far is the best approach though it needs more code auditing.

2025-08-11 05:06:04 UTC

first person to figure it out gets 750 sats
#note12e0…tl78

2025-08-11 00:24:42 UTC

https://i.nostr.build/Qkkrn63zR5LL0ybT.jpg

2025-06-03 12:33:15 UTC

- reply

almost a majority of the OpenSats board has corporate interests, and they have barely any transparency, and until that changes there's a lot of reasons to believe that it is how OpenSats works
I have also looked at public nostr grant information, and many of those are excessive or pointless...

2024-06-24 00:46:07 UTC

so long and thanks for all the fish
I’m going to be taking a step back from here
nostr.land will be maintained for the forseeable future, for issues about it please DM @npub1gt9…lxjt

2024-06-13 21:09:23 UTC

yeah deleting damus again

2024-02-27 16:13:41 UTC

I'm disappointed at the standard being set for bounties.
It has been half a month since the release of hist.nostr.land, which was made to fulfill @nprofile…9h9m 's bounty for a relay that kept old contact lists. Since then, I have not been paid, with my submission being initially dismissed for not meeting criteria not mentioned in the bounty post, with the only time estimate for payment I have gotten a few days after that being "this year", and no further follow ups after that.
This sets a dangerous precedent and creates a bigger power imbalance between people offering bounties and developers: developers can have their submission dismissed on terms that were never mentioned, and the people offering bounties can postpone payment for however long they please, while developers have to open source what they made before they get paid to be eligible.
A bounty is already a risk for developers: someone can claim it before them, and their efforts could go to waste. Setting precedent that allows people to postpone payment to developers for whatever reason and add arbitrary terms will only reduce the amount of people willing to work on bounties, and goes against common practice:
You pay first (or sign a contract and/or escrow funds), and once the work is complete, you have to pay. No subjective terms, no arbitrary rejections, if it meets the terms then you get paid. Bounties as they are completely flip this on their head by requiring people open source their work without a guarantee to get their promised amount, and terms can be added after the fact.
We can't do much about this due to the nature of bounties, but we have a few options to improve the situation:
- For @nprofile…9h9m and others to pay bounties when it meets the terms without adding arbitrary terms after or withholding payment. If you don't have the money to pay right now then don't make a bounty.
- Switch to alternative solutions. Grants after project release do not work as they rely on some entity like OpenSats to accept the project after developers have done all the work, with no guarantees since they can be rejected due to whatever reason, such as it not matching with the vision of the grantor or due to personal issues. The ideal solution is to have a contracting system: have some group of people escrow funds, have clear defined terms, and only pick one developer (and add more later if nothing gets done).