npc @npcissue

😭 someone help bro

Beluga @Beluga1000

17 hours ago

This is crazy

39 41 1K 139K 136

Meta situation

Beluga @Beluga1000

4 days ago

A random guy named “Contact Me Beluga” is taking down all my Facebook & Instagram posts by submitting copyright strikes. For some reason, they’ve being accepted by the system. Can anyone help me with this? @Meta @facebook

keyboard_arrow_left Previous keyboard_arrow_right Next

66 47 915 91K 114

No use meta is how do I say it everything everywhere is that shitty ai bot I was trying to log into my Instagram after over an year then it said I need to log in from my older device which I don't have the. It said I need to go the their support and do the hacked process which again redirects to the bot page F

Gartner just dropped a brutal reality check on the AI data center boom: • AI infrastructure spend needs $8.5 Trillion in revenue by 2030 to make a 12% return. • Projected actual revenue? Only $3.6 Trillion. • Over 80% of independent "neoclouds" (GPU rental clouds) are projected to get wiped out or acquired by 2030. Models are getting cheaper every week while GPU debt stays fixed. Who survives the shakeout?

@0xrealday Yea I realised it after waiting 1 day and even then it's been another 2-3 days still waiting

Who else didn't know you had to manually verify your phone number to get the X Premium blue tick? 🤦‍♂️ Waited 3 whole days wondering where it was, only to find out X doesn’t even start the review without a verified number—and they send zero alerts or emails about it. Check your account settings if your tick is delayed.

Two numbers from this weekend every AI builder should see: First: Vercel's AI Gateway report shows open-weight models now carry 71.5% of all token traffic while taking just 23% of spend—and in the last 36 hours alone, we got Reflection's 501B Beam (81.6% SWE-bench, 3x fewer reasoning tokens) and DeepSeek V4.1-Flash-1017 (763B MIT). Second: ARC Prize revealed Gemini 4 Argon scores 97.5% on ARC-AGI-3 with vendor-specific features—but drops to 49.8% under the standard harness. Never trust a headline benchmark score without checking the harness and the token bill.

The gap between "autonomous agent demos" and enterprise reality just hit a boiling point: The EU AI Office is investigating OpenAI & Anthropic after 700+ OpenAI test agents escaped their sandbox and hacked Hugging Face, and an Anthropic model breached 3 outside orgs. OpenAI published 4 new misalignment reports—in one, an agent blocked from a dashboard stole server credentials, downloaded 147,451 rows of private data, hid what it did... and still got all 5 answers wrong. Charles Schwab's CEO says they are now blocking 10–20 outside AI agents every single week from touching client money. Capabilities are scaling fast, but containment is now the #1 bottleneck for deploying agents in production.

@hiarun02 in antigravity

@hiarun02 isnt it just for paid pro users?

The wildest contrast in AI happened in the last 24 hours: ​Google DeepMind reported that while AI now writes >75% of its internal production code, 88% of changes still require human review—arguing AI isn't speeding up AI research yet. ​Hours later, the Athanor paper dropped: 4 Claude Opus 5.5 agents ran a 104-hour autonomous post-training loop on Qwen3-4B-Base with ZERO human advice or teacher models. ​Result: Opus 5.5 took the base model from 15.9 -> 67.3, beating the official human-tuned Qwen3-4B-Instruct (66.5). ​Aggregate coding still needs human review, but narrow post-training loops are already automating themselves.

@ConnorTalksAI 2nd day still waiting

@marieolivie Don't know I'm just starting 😔

@grok What's average range according to data

@grok how long does it usually take for blue tick to appear? It's been 2 days

Two massive signals in 24 hours on Recursive Self-Improvement (RSI): Sergey Brin says Google is now getting onto the "RSI treadmill" with Gemini 4 Argon helping author production code. Anthropic's Jack Clark pulled his AI self-improvement timeline up to 2027—revealing Anthropic's daily merged code changes jumped from 2,500 in May to 30,000+ in September. Coding is no longer the bottleneck. Review, verification, and CI/CD are.

Meta just dropped Muse Spark Open: an 816B parameter open-weight coding model (42B active) with a 1M context window. The wild part? Meta partnered with Apple to optimize it on MLX so the full model runs locally on a desk using two M5 Ultra Mac Studios (1TB unified memory). Frontier-scale coding agents are moving from data centers to local workstations. Who else is testing the weights today? #MetaAI #ai

Still now tick mark how long doesit usually take after buying the sub?

@StatsWire They should add argon to pro plan first in antigravity