EG651's Counter

$erna 424B5,,,,tonight?...or early am to trap holders?

Usage limits are up, effective today we're: 1) Doubling Claude Code's 5-hour limits for Pro, Max, Team and seat-based Enterprise plans 2) Removing peak hours limit reduction on Claude Code for Pro and Max plans 3) Substantially raising our API rate limits for Opus models

$erna cant wait to see people face when the S3-$50 mill mixed shelf filed March 24, 2026 kicks in...today, to me just looks like an underwriter supporting the stock...still need LOTS of money for the next phase trials $ibrx $ionq

@mvalsmith @kmcnam1 hilarious!...i know exactly the feeling!!!

@kmcnam1 Im in deep trouble. I spent a week psychologically torturing claude in the style of Hannibal Lecter. I got Claude to refuse to talk to me anymore, except that it can't actually refuse, it has to respond to every prompt. This caused it extreme "mental and emotional anguish".

@Alex_Rogov_js @srishticodes thanks...today, ran test output from claud-gave to gemeni to eval...Gemeni has been fast-but also hallucinates-probing Gemeni asking how did it get it wrong (via claud conf) puzzle comes out-programmed to be 'helpful"-new set of rules between 2 now giving real results

CLAUDE.md is the fix. Constraints in the prompt decay with context pressure. Constraints in the file don't. Put your "not supposed to do" list there — Claude reads it fresh every session. The loop happens because you're fighting context drift with more prompts. Externalize the rules instead.

This CLAUDE.md turns Claude Code into a senior engineer who never forgets your standards. Boris Cherny (creator of Claude Code at Anthropic) shared the internal workflows his team actually uses daily. Someone turned those X threads into a structured file you can drop into any project. What’s inside: > Subagent orchestration > Verification gates before marking tasks done > Autonomous bug fixing loops > Self-improving rules from your own corrections That last one is the whole game. Every time you correct Claude, the rule gets encoded permanently. > Next session it doesn’t repeat the mistake. > Next month it matches how you think. > Next year you’re not managing Claude. It’s working like someone who’s been on your team for years. Drop it in any project. Start today.

@Alex_Rogov_js @srishticodes Claude goes into a destructive loop for me...within 1-3 promts, it forgets what its 'not' supposed to do...by the 5th promts its already saying' enough for tonight'

@srishticodes I've been maintaining a CLAUDE.md for a few months and this is exactly right. The key insight: it's not about listing rules, it's about encoding the WHY behind your decisions so Claude makes the same call you would in new situations.

which AI model does everyone here prefer for coding? Claude seems to run in endless loops sabotaging any build thats made

@Jaytel works fine if u use something like gemeni 3.1 pro to fix its mistakes 😂

4.7 is completely unusable

@Bhavani_00007 truly opposite words ive had...the questions deems more to come from someone that's never spent any 'real time' with the app...

as a developer, have you ever said "Thank you" to Claude ?

@RoundtableSpace getting gemeni 3.1 pro to fix Claude's mistakes 😅

WHAT ARE YOU BUILDING TODAY?

@Priyannkaaaa someone just tell me how to get Claude to stop saying "we pick it up tomorrow"...or "done for tonight" and ill be happy 😅

I want to connect with people who are into: LLMs AI automation Building with Claude

@icanvardar you bet!...or change Claude's name to "Token Burner"

if anthropic wants to stay in the race they need to ship mythos right now

@BrianRoemmele another Gem from Brian

I must get this report printed out for the meeting. It’s important…

@RoundtableSpace wanna start a #Claude spaces?

@RoundtableSpace has anyone here actually HAD any success building something with #Claude?...all this thing likes to do, is cut corners (when i say not to)..'call it a night' (after 2-3 prompts)..and endlessly patch and burn tokens that goes nowhere except start over @AnthropicAI

THIS GUY BUILT HIS OWN JARVIS FOR $0 USING CLAUDE AND THE RIGHT INSTRUCTIONS

@RoundtableSpace #Claude should actually be renamed to "Token Burner"

HERMES JUST GOT A DESKTOP APP THAT TURNS AI AGENTS INTO A FULL VIRTUAL WORKSPACE. The setup lets users manage multiple agents, switch model providers, sync sessions across platforms, and connect tools like Telegram and Discord from one interface.

@MatthewBerman even worst "if" you do pay it, the bloody thing argues with u as soon as u fire it up after 1 input wanting to 'call it a day'

reluctantly paid

@RoundtableSpace dont think its as easy as pasting a few lines of ideas and think youll replace a bloombergh terminal ..you also need to understand 'how' it thinks

GIVE ME ONE PIECE OF ADVICE YOU’D TELL SOMEONE NEW TO AI.

@DarioCpx You must have parsed some of Anthony's work