WOW. 46% of Italians and 37% of French people see the U.S. as a threat https://t.co/udMajWQ2ow

(1/5) FP4 hardware is here, but 4-bit attention still kills model quality, blocking true end-to-end FP4 serving. To fix that, we propose Attn-QAT, the first systematic study of quantization-aware training for attention. The result: FP4 attention quality is comparable to BF16 attention with 1.1x–1.5x higher throughput than SageAttention3 on an RTX 5090 and 1.39x speedup over FlashAttention-4 on a B200. Blog: https://t.co/NxVSXKWEgI Code: https://t.co/6irFgQ7GeM Checkpoints: https://t.co/GsrzbJlRY8

+2 more

@emollick Humans are jagged too! We're just very used to their type of jaggedness. Here are 6 random people from a jagged-intelligence study I ran. https://t.co/ASUin06qq5

No wonder Trump loves Hungary's Viktor Orban. Trump wishes he had made as much progress as Orban in implementing the Autocrat's Playbook by suppressing the media and civil society. But Orban may soon be voted out of office. Republicans are likely next. https://t.co/VGIA0a8DwD

A Bloomberg Terminal costs $30,000/yr and still can't do a fraction of what Perplexity Computer just launched today 💻 It now connects directly to your bank accounts, credit cards, loans, and brokerage accounts through Plaid. Your full financial picture, from monthly spending to net worth to individual stock positions, sitting on top of 40+ live finance data sources including SEC filings, FactSet, S&P Global, and Coinbase. Every dollar you earn, spend, owe, and invest, cross-referenced against institutional-grade data in real time. You can walk up to this thing and say: → Run a risk analysis on my portfolio against the current tariff environment → Show me where my spending spiked last month and what's driving it → Build a net worth dashboard that tracks everything in one place → Flag any holdings that overlap with what insiders have been selling this quarter And it just does it. Pulls from your linked accounts, cross-references SEC filings, builds the output, and delivers a finished product. The system running underneath is Perplexity Computer. It orchestrates 19 models simultaneously, breaks any goal into subtasks, spins up specialized agents for each one, and keeps working after you walk away. One model handles the reasoning. Another does the research. Another writes the code. Another builds the visualization. All coordinated automatically. Last month they launched with brokerage data only and someone built a Bloomberg Terminal clone in a single afternoon. That post did 7.5 million views. Now they've expanded to your entire financial life: checking, savings, credit cards, loans, and investments all in one place. Wall Street pays $30K a year for a terminal with 30,000 function commands built over four decades. It won't replace Bloomberg for institutional traders executing billion-dollar orders. But for everyone else, the gap just got a lot smaller.

US fertility reached 1.57 last year, the lowest ever recorded, and the WSJ explanation is "uncertainty about finances, relationship stability, and the political climate" my great grandma had eleven children during the second world war, in a country being bombed, in a house with no running water, on rations. poor people have always had kids. the poorest people on earth right now still have kids and the financial excuse is a story we tell ourselves because it makes us feel good and the real one is unbearable the real mechanism is that we got rich enough to redefine children as an expense instead of the point. somewhere in the last fifty years the cultural goal inverted and a child stopped being what life is for and became a line item competing with the lifestyle. once you frame it that way the math never works, because the math isnt supposed to work. that's the point we are living in the richest moment in human history and we decided to use the surplus to buy ourselves out of the future. the most prosperous civilization that has ever existed is committing demographic suicide at the altar of personal optimization and comfort, and the official line is that we cant afford it the birthrate is a lagging indicator of a civilization that forgot why it was alive

WhatsApp’s “encryption” may be the biggest consumer fraud in history — deceiving billions of users. Despite its claims, it reads users’ messages and shares them with third parties. Telegram has never done this — and never will 🤝 https://t.co/2DYguybgoU

The Judges who released this animal 14 times should be found ‘incapable to proceed’ in their jobs. https://t.co/ali2Wfnlg3

INSPATIO-WORLD: Real-time 4D world simulator Turn any video into an explorable, interactive 4D world you can navigate in real-time using WASD controls. Built on spatiotemporal autoregressive modeling with state-anchored world states, running at 24 FPS on NVIDIA H-series GPUs and ranking #1 on WorldScore-Dynamic.

@Yuchenj_UW Perplexity Computer saved me $14k in tax. It found 2 double taxing errors and 2 form filling errors from my $2000-CPA's draft, which CPA fully agreed. In another thread, I let it compute tax from scratch. It's correct to the cents. https://t.co/zqXCxasdHG

Dialing in real time V2V in Sentinel. No Python or venvs, just an installer. Runs locally on any 16GB+ NVIDIA GPU. Sign up for early access: https://t.co/DHpxt0geoY https://t.co/xPAnVrzbHC

Ever visit a GitHub Issue and wonder when the fix shipped? Release info is now directly included in the issue sidebar alongside the linked PR. https://t.co/2xH37vME8N

Introducing OctOpus 🐙 — the Cursor for Data Science. Connect your data, describe your goal (forecast revenue, predict churn, etc.), and say go. OctOpus 🐙 will autonomously write a plan, pick the right models, write the code, runs experiments, and deploy the best model within minutes! 😀 Tested with enterprises and research teams. Pretty cool. Try it here→ https://t.co/JJHN3VsMxs 🐙 Inspired by @karpathy #autoresearch #DataScientist #AI #MachineLearning #Models #Kaggle #Opus #OctOpus @AnthropicAI @OpenAI @GoogleDeepMind @xai @AIatMeta @ycombinator

Sorry I couldn't quite hear you over ALPHA ZONE. But seriously check out my podcast it's called In Practice and it's weird and technical about real AI applications https://t.co/3JEzhTmWCm https://t.co/57w1bUoFtJ

+3 more

Twitter’s algorithm is optimized for addiction, not for us. We deserve better. We’re releasing Bouncer today so you can take back control of your feed. Describe what you don't want, and Bouncer removes it. It’s free, doesn’t collect your data, and will be open source soon. https://t.co/fSUxGdmUcS

@RoundtableSpace All of them: https://t.co/y8TQ8DgvH3 My favorite is AI Newsmakers. My AI reads ask of them and picks out the best: https://t.co/kiuZ7QXLzb

Introducing Gooseworks — AI coworkers that can do real GTM work. Over the last few months, we've been obsessed with turning AI into actual coworkers for our team. And it's working. Goose is now our most valuable teammate – it finds high-intent leads, runs outbound campaigns, coordinates with influencers, tracks SEO / AEO and a ton more. Today we're launching this publicly so everyone can create their own AI coworkers. Each coworker (or Goose) has its own computer, filesystem, mailbox, memory, and tools. Goose can work autonomously for hours, communicates over Slack, iMessage, WhatsApp or email, and gets smarter as you give it more context. It's not just a personal assistant — your whole team can talk to Goose from anywhere, and schedule it to run on autopilot. And here's the best part – we've already given Goose 100+ skills and data APIs for GTM work: – map your TAM – search people databases with natural language – find leads from X and LinkedIn posts – monitor intent signals from your ICPs – update your CRM and generate reports – find influencers and run outreach campaigns – track visibility in search and answer engines – connect to all the SaaS tools your business runs on +++ lots more. We also open-sourced every skill as a toolkit you can install directly in Claude Code. Comment "Goose" and follow and I'll send you the install link.

jump scare from @thsottiaux https://t.co/7fcMlHfWiQ

one of my favorite parts of the job sitting down with extraordinary people like @jxnlco from OpenAI to discuss how codex changes the way we work presented by @cerebras // @alyciazcary https://t.co/g9j1yKvQXq

introducing Chargebotic, autonomous charging for robots. robots are getting smarter every day. but they all share the same problem: someone still has to charge them. designed for space to charge and save rovers on the Moon, we realized robots on Earth have the same problem. Chargebotic is a simple, universal charging system. it includes a charging pad, a receiver module, and software that manages your robots' charging. this allows your robots to charge themselves anywhere in extreme environments, across entire fleets, and across every robot brand. currently working with our own robot, Spark-E. testing with more robot types coming soon. sign up for early access link in bio.

All In’s @davidsacks liking one of my tweets was not in my 2026 bingo card. https://t.co/WFsRZolcRO

it all makes sense now. dario was still at openai in 2019. he left next year and took his marketing playbook with him. hasn't changed a thing since. https://t.co/lJ27MBHYFR

One fun thing about AI is that it lets you play with interfaces and approaches to displaying information in new ways without a lot of effort. I got a an internet connected e-ink display and set it up to show me the weather as interpreted by nano banana using rotating styles. https://t.co/BOD4V0jhlM

https://t.co/thNKojJJoI

GenAI’s popularity has hit a wall. Hard to see how OpenAI is going to make its numbers, and easy to see why they bought a media company. Their work is cut out for them. https://t.co/9Pv8QBpqzH

@per_arneng @ApacheArrow @kszucs_ @huggingface ➡️ https://t.co/biadKxzubv

GenAI’s popularity has hit a wall. https://t.co/LNyrZpB1ql

INTRODUCING: https://t.co/nn42Az3rY1! 🐉📺 this one's a little different than my usual project, but i hope you enjoy this foray! 🫶 the idea? AI-curated content... straight from the latent space! as it turns out, LLMs have been training on lots and LOTS of embedding links 🧐 so in the Pliny TV backend, there is no search API. no YouTube Data API key. no vector DB. no scraper. discovery happens entirely inside a language model's memory of the internet. you describe a channel in plain English — "music videos from 200", "cozy 3am lofi rain", "cursed liminal space docs", — and an AI curator recalls real YouTube video IDs directly from its weights! the system prompt literally says: "you have been trained on millions of YouTube videos — USE THAT KNOWLEDGE." we then hit YouTube's oEmbed as a reality check. survivors make it to the channel queue, hallucinations get filtered out. the LLM is the search engine. the index is the model. because there's no keyword index, no recency bias, no algorithm optimizing for watch time, the curator doesn't search. you give it a vibe and it navigates the probability space of that concept, pulling back whatever is most salient. which turns out to be, almost without fail, the most nostalgic, cultishly-loved, collectively-imprinted videos in that region of latent space. the memes you forgot you'd forgotten! you ask for a vibe, you get a tour of the collective unconscious, sorted by resonance. less search engine, more akashic record. and one of the most fascinating elements is seeing how this effect differs between models! different curators produce vastly different generated playlists, even with the same prompt. what's in the box? 📡 create channels & publish them to public so anyone can tune in, plus an option to have AI auto-generate an idea 🤖 AI curator auto-queues 15–20 videos per run, auto-shutoff at 20min to save API cost 👥 AI personas watch with you and chat live — custom personas generated from your channel's vibe 🎨 110+ preset themes — each one a full visual universe, chosen by the model based on your prompt 🎬 auto-skip on dead videos, shuffle, request queue, fallback tiers when the AI whiffs 📱 retro flip phone UI — dial channels by number, unlock secret speakeasy rooms with hidden codes (420, 1337, etc.), DTMF tones and all you're not searching YouTube. you're asking an AI what it remembers. and what it remembers is what the internet collectively loved the most. video streaming as dream logic. https://t.co/Dg6zZvpqO1 🐉📺

We just launched Tinycloud into public beta 🌥️ It’s an AI agent for video work. It sees everything in your videos (transcripts, audio, visual context, on-screen slides, speaker patterns, and more) so you can query and make artifacts from them. Think Claude Code but for doing work across your videos. Try it and show us what you create → https://t.co/dfKusQCKIY 🧵

Automating Loan Income Verification with AI agents 💸☑️ Loan processors spend a huge portion of time verifying the borrower’s documents (tax forms, pay stubs, bank deposits), and cross-checking them for consistency. This tutorial shows you how to encode that logic into an agentic workflow that incorporates both state-of-the-art document OCR/extraction along with agentic reasoning with structured outputs to understand discrepancies between documents. All decisions are flagged to the human for approval. For these sensitive applications where 100% accuracy is critical, the goal is to not replace the human but help them make decisions more efficiently and accurately. The full tutorial is here: https://t.co/ZMFtyfkb8x. Feel free to follow it as a human or point your coding agent towards it. If you’re interested, come check out LlamaParse! https://t.co/TqP6OT5U5O

OMG HOLY CRAP ARE YOU TELLING ME TARIFFS MADE THINGS MORE EXPENSIVE IT'S ALMOST AS IF WE SHOULD HAVE LISTENED TO ECONOMISTS OH WELL https://t.co/yW8CaKrQkQ