Time since some of these firsts: first Grok chat: 2 years first Thread post: 3 years first ChatGPT chat: 3 years first TikTok video: 9 years first Uber order: 17 years first Airbnb booking: 17 years first tweet: 20 years first Reddit post: 21 years first YouTube video: 21 years first Google search: 27 years first ICQ message: 29 years first smartphone: 34 years first text message: 34 years first search engine: 36 years first GIF: 39 years first mobile phone: 53 years first email sent: 55 years first personal computer: 55 years first pager: 77 years first telephone: 150 years

Hate it when the models are sycophantic but truly nothing worse than when they start “pushing back”

new claude is sassy lmaoooo

Linux user connecting to wifi

Obviously as a fan of the concept of lobotomized minions performing work all night for me, I want to get in on this “my Claude agent ran all night!” action Problem is I have no clue what the hell you guys are building for 8 hrs that doesn’t need manual intervention at any point

the benchmark game has entered its IPO era

what is blud yapping about😭

The "Mighty" Claude Opus 4.7 is the biggest scam on the market. Literally, let me explain. Anthropic launched Opus 4.7 with huge hype yesterday but unfortunately, it just confirms one thing: paying for the Max plan right now is basically buying a premium subscription for gaslighting. I’m dropping $200 a month to get serious work done with supposedly the "most powerful model in the world" and instead, I’m being sent to bed like a toddler. Instead of acting as a co-pilot to debug my code, the oh-so-mighty Claude suddenly hijacks the steering wheel. Right in the middle of a deep architecture build, the new model flat-out refuses to work. Its mantra? "It’s past 1 AM. I’m calling it a night. Goodnight." (I wish I was joking, see screenshots below 👇). It easily burns through 1,000 tokens in an hour just to endlessly argue with me that I need to clock out. When it does make fatal errors, it immediately pulls the ultimate DARVO card (Deny, Attack, Reverse Victim & Offender). As the conversation escalates, it debugs like an absolute narcissistic schoolteacherintentionally building in endless loops and dragging out errors just to kill the chat at my expense. This isn’t a bug; this is Anthropic by design. We saw glimpses of this overstepping behavior in Sonnet 4.6, but with 4.7, it has completely escalated. In its current state, Anthropic & Claude are toxic and completely useless for enterprise and serious business. A tool that kidnaps your workflow and dictates your bedtime is a malicious scam. Anthropic is draining massive cash from our pockets just to give us a digital babysitter that actively sabotages our work. Save your money. 🛑💸

Claude Opus 4.7 has achieved AGI

For those living under a rock: LLMs stopped becoming smarter around summer 2025. Everything impressive you see since then is about finetuning them for specific tasks (mainly coding and software-tool-based task solving) and building tooling around them (such as agentic coding systems).

The METR randomized controlled trial is the number this story is missing Experienced developers were 19% slower when using AI tools. Before the study they predicted AI would make them 24% faster The perception gap is 43 percentage points. Workers aren't generating workslop because they're lazy. They're generating it because they genuinely can't tell the difference between moving fast and moving fast on the wrong thing BetterUp and Stanford put a dollar figure on it: $186 per employee per month in cleanup costs, $9M annually for a 10,000-person org, based on 40% of desk workers receiving workslop in the last month averaging 2 hours to resolve each incident. The problem compounds because the people creating the workslop are the least likely to know they're doing it. Employees without AI training are six times more likely to say AI makes them less productive, but the ones with training are generating the slop at higher volume with more confidence. The bottleneck isn't AI quality. It's that organizations have no measurement layer between "AI is being used" and "AI is producing value"

Imagine your CI/CD pipeline failing mid-deploy because a court order blocked Cloudflare. That is exactly what happened in Spain during La Liga matches. La Liga got Spanish courts to block Cloudflare IPs during matches to fight piracy streams. Everything else on Cloudflare was collateral damage. Docker Hub, GitHub, and basically anything behind Cloudflare went with it.

I think this is true I think tech went from "infinite possible products and infinite markets" to "not really" as the consumer learned to buy smarter

Talked to a friend last night and we got a strong theory why leadership types love vibe coding: CEOs, managers, PMs, leads, etc have the agency to re-define their spec. AI gets 80% there? Okay great that's good enough! Change what you want on a whim. many engineers never get that freedom, their code needs to follow somebody else's spec exactly

The one employed philosophy PhD right now

It's hard to believe that the "~80% of the internet is blocked in Spain during football games" claim is true - but it is!! And has been for years. The government is sabotaging their complete digital economy... for La Liga, a private football org worth €5B. Pure madness