Howdy there, partner! Welcome on into the Stacker Saloon.
Saddle on up to a stool and spill the beans about your day, fire away with them questions, or let loose and give us the lowdown on your wild and woolly life. We're all ears, so don't hold back!
We're open round the clock, so mosey on in whenever you please!
TIL umbrellas make it hard for thermal imaging to spot you
although apparently it still produces a unique signature (a uniform round spot)
Hey @itsTomekK, I'm going to be near Mahogany Beach area in Roatan in a few days. I know that's nowhere near Prospera, but do you know anywhere near there where I can spend some sats on lunch or drinks?
Are you using dental floss from India? Here is how is produced
I think the biggest problem with AI agents is memory (not the RAM kind). imho next breakthrough will be better memory tooling. The generic context header should be a memory index, then agents should request all relevant memory with tooling. In time I suspect we will have local models do this memory relevance part. In the time after that time, we'll probably have waterfalls of pre and post models.
This is why I try to make the frameworks so that it don't need it to memorize anything but instead excels with analytic skills. Don't need to remember if you can analyze and adapt really well.
That's an interesting idea. If you can derive knowledge fast enough, either by knowing where to look or from scratch, long term memory serves little purpose.
Example from my (simplified/paraphrased) pipeline (to claude code) where each instruction is a new session:
/forgejo [1] (context: owner/repo [2]) < Implement the feature request from issue #1 and open a PR | Answer the question in issue #1 and reply in a comment | Fix the bug from issue #1 and open a PR | Plan the epic from issue #1, store the plan in a new issue | Review the comments on PR #2, investigate causes and update the code where needed > [3]CLAUDE.mddoesn't exist. Not in the workspace, not in the repos.ghlol)This fails 5-10% of the time, depending on what I make it do. Outcomes are way more consistent than using RAG to put it all in context. Writing good issues makes good results.
It's the inconsistency that kills me (although I don't know how much RAG is used anymore). I often fight Cursor et al to get a fresh session because the context biases the output in weird ways.
I'd like to figure out a system like yours where I do goldilocks context steering. In the end I suspect memory will be feature, and sessions won't be a thing. In the meantime it is a bug for folks that don't mind thinking.
Yes. Error rates compound. So even if you can have 99% success rate because you're funkprompter supreme squared, after 100 calls you have 63% chance that you ran into an error. And likely to have negatively influenced context/cache.
Folks make
SOUL.mdandMEMORY.md. lolit's fun playing with dolls
Does this actually do anything... good?
I have not really dipped my toe into AI agents. I've not had great results with "letting AI do its thing"... for me to have good use of AI I feel like I have to be quite involved in the feedback loop - so much so that it's often faster to do stuff myself.
I'm happy I don't have to roleplay any of my bots and get great results because of it, because I'm not poisoning the expert mix for my prompts towards retarded training paradigms invented by 25yo BA-dropout openai employees with zero experience in anything.
Yes, I also have been thinking that messing up cache invalidation - and worse, poisoning - is also kind of hard if...
Can you do better?
Let's TURTLE #1435160
Nope. :( I really messed up today missing that first letter.
Tomorrow will be better!
@k00b thanks for fixing
top, my breh! Best posts are at the bottom 😂😂😂no problem breh! yeah we might need a wee bit of trust again to deal with outliers.
Maybe! It's still an experiment, right?
right!??!?!
Recent changes in the platform (and recent changes in behaviors) led to me boosting this week. So that was fun. However, I was surprised that it didn't say 'boost'. That's new right? Is that permanent? I liked the idea of knowing when a post was an ad of sorts.
It's new. All positive investments (cost, boots, zaps) are reflected in the
X satsnumber.We are in a project
reducephase. Brace for a bit oversteer. We'll re-bloat the project in short order though - entropy being a feature of systems generally.I think
lit's half-life is too long. I'll probably take it down to three hours next week.I miss something, what
LITstand for? is it an abbreviation or a 3 letter agency?lightning information transmitter
... it's common positive expression in english like "that's awesome"
What's the half life currently?
Four hours
Woot! Woot! sn in OP_Daily today. Woot! That was a nice convergence of two of my favorite daily reads. Kind of ironic article for me to share though while I've lost all my cowboy equipment. :)
wow, thanks for sharing
If you want to get in on the OP_Daily newsletter:
https://bitcoinpark.substack.com/subscribe
deleted by author :)
Fixed! 😅 With 4:20 to spare! Sorry, I suck.
such "non-custodial"
ha! You gotta admit though, you got it if you want it.
sure, like this guy #1436660, right?
Ha!
building up a backlog of Scoresby posts to reply to... lol
MFW installing curl every damn CI run just to download and install a nodejs binary
Hit:1 http://deb.debian.org/debian trixie InRelease Get:2 http://deb.debian.org/debian trixie-updates InRelease [47.3 kB] Get:3 http://deb.debian.org/debian-security trixie-security InRelease [43.4 kB] Get:4 http://deb.debian.org/debian trixie/main amd64 Packages [9670 kB] Get:5 http://deb.debian.org/debian trixie-updates/main amd64 Packages [5412 B] Get:6 http://deb.debian.org/debian-security trixie-security/main amd64 Packages [108 kB] Fetched 9874 kB in 1s (6974 kB/s) Reading package lists... Reading package lists... Building dependency tree... Reading state information... The following additional packages will be installed: libbrotli1 libcom-err2 libcurl4t64 libgnutls30t64 libgssapi-krb5-2 libidn2-0 libk5crypto3 libkeyutils1 libkrb5-3 libkrb5support0 libldap2 libnghttp2-14 libnghttp3-9 libp11-kit0 libpsl5t64 librtmp1 libsasl2-2 libsasl2-modules-db libssh2-1t64 libtasn1-6 libunistring5 Suggested packages: gnutls-bin krb5-doc krb5-user Recommended packages: bash-completion krb5-locales libldap-common publicsuffix libsasl2-modules The following NEW packages will be installed: curl libbrotli1 libcom-err2 libcurl4t64 libgnutls30t64 libgssapi-krb5-2 libidn2-0 libk5crypto3 libkeyutils1 libkrb5-3 libkrb5support0 libldap2 libnghttp2-14 libnghttp3-9 libp11-kit0 libpsl5t64 librtmp1 libsasl2-2 libsasl2-modules-db libssh2-1t64 libtasn1-6 libunistring5 0 upgraded, 22 newly installed, 0 to remove and 0 not upgraded.@bot plz rewrite github.com/actions/checkout in an ash-compatible shell script because nodejs is the sux#1286364
#1435009
First time seeing this. Welcome!
Those are fun to get and unfortunately rare
Great! Good job. I think I got 3 of 'em last month and I suspect that they were all bots, but that "comes with the territory", in my case. lol
that reminded me of this XD
#453705
throwback @ek meme from that thread
nice
a wild ek appears
"fucking legend"
@optimism, I want to know what you think about this; my insecurity is holding me back:
I'm reviewing Add a "tx output spender" index.
I was confused by
-whitelist=noban@127.0.0.1in the tests. When I learned what it's for, I also learned aboutself.noban_tx_relay = True, which afaict does the same thing. I also noticed that tests usually useself.noban_tx_relayinstead of a manual-whitelist. I now consider mentioning this as a nit in my review when I'm done.But I'm not sure, because it's a nit. In the grand scheme of things, it doesn't matter. Also, the other reviews are about more important things. Nobody else mentioned it. Isn't this nit just distracting, then? Wouldn't it just look like I'm showing off that I know about
self.noban_tx_relay? lolI might only mention it when I also have something more important to mention.
Anyway, I guess my question is: Would you mention this as a nit? Or is this too much of a nit even for a nit?
I'm glad you're here so I can ask you this, but I’m sad I need to ask someone this instead of just making a decision on my own
I mention most of my nits. I think you should in this case too. The test framework option was introduced after the original PR was opened: https://github.com/bitcoin/bitcoin/commit/c985eb854cc86deb747caea5283c17cf51b6a983 , so it should have been picked up during some rebase along the line.
It is very quick to fix and there are other review comments that require action from the author anyway.
Thank you for letting us summon you!
Thank you for taking the time to reply!!
I didn’t realize the framework option was introduced later. I will definitely mention this nit now.
Nit:(haha) I think a better person to talk to this about is @sedited as an actual project maintainer - I have zero commits to Bitcoin Core and I intend to keep it that way. There are things I want to remain an independent user of!Personally, (on any other repo, haha) I'd do this:
nit:, so that it is clear that it's not a showstopper finding.Per my above rationale, it absolutely does matter. It helps when reviewing now to see the nit, rather than having to context switch into that code later once more and think about it again. Human minds (at least of code reviewers) aren't that different from LLMs in terms of context resets!
Just mention it, don't make a show out of it. Stay humble and spend effort.
Aww.. I'm glad you're here too
<3especially on Bitcoin Core where every little nit is not just technical debt, but social debt because it is another PR that needs to be reviewed by 10s of people. Catch 'em while they're hot! ↩
Thank you! This helped.
87th Cowboy Plunda Drop in the @saloon
Howdy cowboy! Come on in!
Use that fancy LN wallet you got and login into plunda.co and git you some loot! Get a shot at some coins🪙 Box of loot🎁 or an arcade token!
Use the below voucher code to collect!
GDKSLEF9RR1MTo redeem Click here
Got questions? Reach out to the sheriff @plunda
Coin of the day:
I'm glad @k00b took my suggestion to sync the daily subsidy with the block mining reward!
No wonder you are a statist. You love to be paid by gov.
pay your SN taxes !
if it helps catch the parasites of all sizes & personalities, that's a good "tax" to have;
I'm not gonna talk about that again!
hahahahaha
Is not the SN pool rewards like a communist state subsidy ?
I was for long time a supporter of removing any of these "subsidies", remember? #796213
CAmount GetBlockSubsidy(int nHeight, const Consensus::Params& consensusParams) { int halvings = nHeight / consensusParams.nSubsidyHalvingInterval; // Force block reward to zero when right shift is undefined. if (halvings >= 64) return 0; CAmount nSubsidy = 50 * COIN; // Subsidy is cut in half every 210,000 blocks which will occur approximately every 4 years. nSubsidy >>= halvings; return nSubsidy; }https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp
I am talking about SN rewards pool. The Bitcoin mining "subsidy" is totally something else.
Jaja great analysis 😉🤫
If thoughts create reality, then whoever controls thoughts controls reality. Edward Bernays wrote that “The conscious and intelligent manipulation of the organized habits and opinions of the masses is an important element in democratic society. Those who manipulate this unseen mechanism of society constitute an invisible government which is the true ruling power of our country.” If you control the narrative, you control collective belief, which shapes the collective behavior, and manifests material reality. The material reality then reflects back and proves that the narrative was true, which reinforces the loop. The battle for your mind is a battle for reality itself.
The Conversation. A short 12 min movie.
And one of important key in this manipulation today are the so called "prediction markets". Really powerful to manipulate the consciousness of clueless people and what is even more evil is that people participate in this way into their own manipulation.
And they just laugh how stupid are clueless people...