Ddox

@cmuratori @grok what do you think about this?

Does any of it violate copyright? According to Anthropic itself, their devs do not write any code by hand. As far as I know, AI-generated code is not copyrightable under US law. So correct me if I'm wrong, but you should not be able to use the DMCA to take down this code, right?

@BigWum @braelyn_ai And how to the models know how to code things? Weren't they trained on existing code, thus they are in a way copying it?

@paraddox @braelyn_ai People actually use LLMs to assist with coding original things too, not just copying things that already exist.

you probably dont fully understand how serious this Claude Code leak is > code copyright law has been stable since the 80s > you cannot legally copy or fork someone's code without permission (licenses) the caveat is a "clean room" build - start from scratch by people who have never seen the code. it takes too long to build a product from scratch, so it wasn't usually a problem > enter codegen > anyone with a Claude Max plan can "clean room" rebuild any repo just by pointing the agent at the tests > a few months ago, opensource repos started close-sourcing their tests > a codegen rebuild has never been challenged in court > the legal precedent would have the potential to decimate the concept of Open Source forever SO > last night > claude code leaks > Sigrid Jin "clean room" rebuilds claude code in Python > names it Claw-Code > Reminder: Anthropic sued the last open source tool that used a funny "claw" name Now? > Anthropic has the potential to sue this project and put all of software licensing and copyright law into question > Anthropic might allow corporate greed to effectively put an end to opensource software > Anthropic's corporate greed might forever kill the software community as we know it today ... now that you know how serious it could be, the best thing we can do is pressure anthropic to let it go

@SIGKITTEN If you add an extra 3 stop slider there for permissions it might work nicely. But then it might be a bit confusing because you don't necessarily expect permissions via the /model command.

@paraddox its actually not at all, i was thinking of adding it to the model picker ui

i've got a lot of interesting ideas that i want to explore, so Litter may transform, we're just getting started! All feedback is welcome, both here or on the hub

@SIGKITTEN Obviously 🤦

@paraddox yeah, /permissions

@SIGKITTEN Cool. Stupid question : Is there a way to start Yolo sessions?

@paraddox ty. i think i have this fixed, ill push another update tonight after i test a little bit

once you experience kimi k2.5 turbo at 250 tps using @FireworksAI_HQ, you can't go back to using other models running at 50 or 100 tps🥲

@SIGKITTEN If I stopped android from optimizing the app for battery, it's working fine even after 1h in the background

@paraddox thanks, i havent added the background stuff and notifications for android yet, coming soon. will look into the crash in the meantime!

Good way to start exploration: zread.ai/instructkr/cla…

@JsonBasedman Easy fix. Just use Codex for non UI work.

It's.... Actually kinda wild how retarded opus is after having used codex for a bit now. Like even getting Claude code to subcall codex for design reviews is bad. It just ignores important architectural advice it gets from codex.

@Resorcinolworks Open-source, nah. Can we extract some nice enjoyable bits out it, for sure.

Is the leaked codebase enough to make Claude code open source ? Or we again deadass enjoying that we just found a broken headlight of a car ?

i can't believe more people aren't talking about this part of the claude code leak there's a hidden feature in the source code called KAIROS, and it basically shows you anthropic's endgame KAIROS is an always-on, *proactive* Claude that does things without you asking it to. it runs in the background 24/7 while you work (or sleep) anthropic hasn't turned it on to the public yet, but the code is fully built here's how it works: every few seconds, KAIROS gets a heartbeat. basically a prompt that says "anything worth doing right now?" it looks at what's happening and makes a call: do something, or stay quiet if it acts, it can fix errors in your code, respond to messages, update files, run tasks... basically anything claude code can already do, just without you telling it to but here's what makes KAIROS different from regular claude code: it has (at least) 3 exclusive tools that regular claude code doesn't get: 1. push notifications, so it can reach you on your phone or desktop even when you're not in the terminal 2. file delivery, so it can send you things it created without you asking for them 3. pull request subscriptions, so it can watch your github and react to code changes on its own regular claude code can only talk to you when you talk to it. KAIROS can tap you on the shoulder and it keeps daily logs of everything. > what it noticed > what it decided > what it did append-only, meaning it can't erase its own history (you can read everything) at night it runs something the code literally calls "autoDream." where it consolidates what it learned during the day and reorganizes its memory while you sleep and it persists across sessions. close your laptop friday, open it monday, it's been working the whole time think about what this means in practice: > you're asleep and your website goes down. KAIROS detects it, restarts the server, and sends you a notification. by the time you see it, it's already back up > you get a customer complaint email at 2am. KAIROS reads it, sends the reply, and logs what it did. you wake up and it's already resolved > your stripe subscription page has a typo that's been live for 3 days. KAIROS spots it, fixes it, and logs the change endless use-cases, it's essentially a co-founder who never sleeps the codebase has this fully built and gated behind internal feature flags called PROACTIVE and KAIROS i think this is probably the clearest signal yet for where all ai tools are going. we are heading into the "post-prompting" era where the ai just works for you in the background like an all-knowing teammate who notices and handles everything, before you even think to ask

@unclebobmartin You must be the only one that still hasn't hit his limits :))

I need to plan a major upgrade. Back to Claude to do the brainstorming.

🤣🤣🤣

@flavioAd Is this for real @jxnlco ? Codex CLI just became open source? 🤭

big if true

@FakePsyho oh well. I'll try harder :))

@paraddox You actually have 3 in my calculations for the last 200 tweets 😅, but: - I only count top-level replies - you did your burst of replies after posting here and I cache the results

I got tired of engagement-farming AI reply bots, and a few hours later I had a firefox extension that correctly tags (and hides/collapses if enabled) about 90% of them. Turns out you can detect most of these accounts just by analyzing their posting patterns. Take their last 200 tweets and find the maximum number of reply tweets within any 5-minute window. If it's 4 or higher, it's an automated reply account. I manually tagged 100 suspicious accounts and 50 terminally online accounts. This single if matched 86/100 of sus accounts and none of the terminally online ones. I've been running this for a few days now and I'm still surprised by the accuracy. I'm pretty sure you can detect almost all of this AI slop without even looking at the content of tweets. I manually checked "false positives" I got and all of them were either: (a) accounts spamming gifs/single words/emoji replies, which is more or less the same behavior just without AI (b) accounts that used AI replies in the past. tl;dr: fixing majority of the current slop apocalypse requires a 0.1x engineer and a minimal effort

@SIGKITTEN I only do that for PR's I open on open-source projects. Haven't read my own code in 6 months :))

many such cases

@Shpigford @claudeai Come over to Codex. with /fast and 1M context, I'm struggling to hit the limits :))

note: NOT my 1M context. I disabled that everywhere and I'm still hemorrhaging usage. @claudeai i know you guys are aware of this but...please fix. 😭

@steipete @benitoz We're gonna blame the Codex CLI leak on you though :))))

@benitoz hey don't bring me into this, I didn't leak it xD

bun just cracked open Claude Code for everyone. agentic AI is going mainstream. the next inflection point is agents. hope we have enough compute 🙏🏽 @steipete 🦞🦞🦞

@SIGKITTEN Hey! Don't tell them. How are we going to have fun if they do it?

pst anthropic, you can use the npm prepublishOnly hook to check for source maps