When you have one job, and blow it. https://t.co/O5P0rdeCJx

刚刚发现，Codex 又又又又重置用量了！ https://t.co/qJgQpuKW0M

刚刚发现，Codex 又又又又重置用量了！ https://t.co/qJgQpuKW0M

Introducing Atlas 1. Willow's new frontier speech-to-text model. It outperforms ElevenLabs, Deepgram, OpenAI, and more by a wide margin. Built on the first scalable, human-powered transcription infrastructure ever built for real-time dictation. https://t.co/PX1xeQ9Yeh

Accepting the reality of your life is difficult and under discussed https://t.co/jMszVRnFeX

Accepting the reality of your life is difficult and under discussed https://t.co/jMszVRnFeX

Collinear presents YC-Bench This benchmark evaluates agent capability to run a simulated startup over a one-year horizon spanning hundreds of turns. https://t.co/zRxoYHsZRO

proj: https://t.co/G5Q5dFrCfA abs: https://t.co/FWRb3b6mBt repo: https://t.co/NQ8qduL67y https://t.co/zKey2qeiYH

+1 more

An exciting update for the Pokee AI Hackathon on 4/4: We’re thrilled to have @Replit joining as an event partner, with the opportunity to win Replit credits! Alongside Replit, we are pleased to additionally announce Innovation & Tech Today as an event partner, where hackathon winners will also have the chance to be featured in their magazine! Hackathon winners have a chance to take home: 300K+ in credits across Pokee + Replit $2,000+ in cash prizes Feature in Innovation & Tech Today (@InnoTechToday) Judges for the event include: Maria Zhang — CEO, Palona AI; former CTO, Tinder; former VP, Google/Meta Bo Li — CEO, Virtue AI; UIUC Professor Jean-Marc Daecius — Chief of Staff, O'Shaughnessy Ventures (@JMBDaecius @osvllc) Zhen Li — Founding Engineer, Replit Robert Scoble — Tech Evangelist (@Scobleizer) Bill Zhu — Founder, Pokee AI (@ZheqingZhu) 🗓️ Saturday 4/4 ⏰ 1–7 PM 🌁 San Francisco 🍽️ Dinner included Please note: we’ve officially reached capacity, so the event is now waitlist only. It’s also strictly in person. You can still sign up via Partiful solo or with a team — just please indicate which teammates will also be RSVP’ing. Thanks for all the excitement — it’s going to be a great one! We can't wait to see what you all will build.

introducing the worlds first Hermes Clipping Agent. paste any YT link in Telegram, WhatsApp, or Discord. Hermes does all of the heavy lifting. - analyzes the video - finds clips that actually get views - custom face-tracking and cropping (p.s he also schedules and posts for you) agentic clipping is here. 👇 get your api key here -> https://t.co/qRDa2sYLFR

Back to classically SF weather, anyone have a good book recommendation? https://t.co/Pd8Xe3JasS

McKinsey highlights Southeast Asia (SEA) is a global AI leader where 46% of companies have AI at scale compared to 35% globally, fueled by billions invested by Chinese and US tech giants. What McKinsey dares not address is that SEA is profiting from its position as an AI battleground between the US and China. With Chinese and American tech companies locked in cut-throat competition, SEA is able to play the providers off of one another. Indonesia and Malaysia are great examples of nations where the competition is particularly fierce, with Alibaba, Tencent, Microsoft, and Google all investing billion-dollar sums in data centers. Another point that McKinsey misses is that the region’s young connected population largely takes China’s tech for its inspiration. With ubiquitous QR payments and superapps, what is successful in China will likely work in SEA. This doesn’t ensure Chinese AI and cloud providers will come out on top, given the popularity of Facebook and YouTube. That said, when combined with China’s open-source AI ecosystem, China’s AI and cloud providers are more competitive than most give them credit for. Kudos to SEA for turning the superpower competition into a massive local advantage! #fintech #tech #finserv #AI @BetaMoroney @efipm @BrettKing @spirosmargaris @jasuja @enricomolinari @mikeflache #China #techwar #chips #tech @baoshaoshan @thecyrusjanssen @DOualaalou @lajohnstondr @PSTAsiatech https://t.co/f9Pus9zcMV

Charlie Munger: "The very word diversification, which is taught in every business school as the holy golden grail, I consider almost asinine." "I'm only interested in investing where I have some sort of advantage. If I have the advantage in three different things, that's enough for any family." "That's fine for the ignorant people to be fully diversified, but if you actually know what you're doing, I think three investments are plenty."

Introducing OpenHarness, an ultra-lightweight, pure Python alternative to Claude Code that delivers approximately 80% of essential agent functionality using just 3% of the code lines. With a single command, you can launch OpenHarness and unlock seamless integration with popular CLI agents including OpenClaw, nanobot, Cursor, and more. OpenHarness is an open-source Python implementation designed for researchers, builders, and the community: - 🔍 Understand how production AI agents work under the hood - 🧪 Experiment with cutting-edge tools, skills, and agent coordination patterns - 🔧 Extend the harness with custom plugins, providers, and domain knowledge - 🏗️ Build specialized agents on top of proven architecture Try OpenHarness: https://t.co/4vtw4KK7gg

I made a tamagotchi for my plant! 🌱 (final part) https://t.co/gxVHvJOGEg

codex takes an underperformer off the project and gets in the trenches we love to see extreme ownership https://t.co/ChE6D9X93h

I made a tamagotchi for my plant! 🌱 (final part) https://t.co/gxVHvJOGEg

I made a tamagotchi for my plant! 🌱 (final part) https://t.co/gxVHvJOGEg

Solving the Trolley Problem https://t.co/TAE6LQYjnS

For four years, I tried to warn everyone that AI would radically ramp up the ability of foreign actors to generate misinformation. Now here we are. https://t.co/jMGPW1MX5Q

Hard for me not to laugh. I certainly did try to warn them. https://t.co/s2lquJRFEF

🚨 BREAKING: Someone just dropped the most advanced Steganography Platform EVER!! 😱🥚 https://t.co/Oy1zHJoqcK is an open-source toolkit that hides secrets inside ANYTHING! images, audio, text, PDFs, network packets, ZIP archives, and even emojis 😘️︎︎️️️️︎︎︎️︎︎️️︎︎︎️︎︎️️️️︎️︎️︎️️︎︎️︎︎︎️︎️︎︎️︎︎︎︎︎︎️︎️︎︎︎︎︎️︎︎️️︎︎︎️︎︎️︎︎️︎️︎︎️️️︎︎️︎️️︎︎️︎︎️️️️️︎​ AND it has an AI agent built in 👀 🔍 REVEAL: drop any file and the AI agent tests every known decoding method automatically. 120 LSB combinations, DCT, PVD, chroma, palette, PNG chunks, trailing data, metadata, Unicode, and more. 50 tools running in parallel. auto-extracts hidden payloads as downloadable artifacts. no config needed. 🔮 CONCEAL: type your secret, pick a method (or let the AI choose), upload a carrier image OR generate one with AI. one click → encoded steg file. the agent recommends the optimal method based on your use case. the methods: ⊰ LSB — 15 channel presets × 8 bit depths = 120 combinations. steghide has 1. st3gg has 120. ⊰ F5 — operates on JPEG DCT coefficients. SURVIVES social media compression. regular LSB is destroyed by ANY JPEG compression, even quality 99%. ⊰ PVD — encodes in pixel pair differences. statistically harder to detect than LSB. ⊰ CHROMA — hides data in color channels (Cb/Cr). human eyes are less sensitive to color than brightness. ⊰ SPECTER (unique) — data hops between RGB channels in a pattern that IS the key. like frequency hopping in radio. ⊰ MATRYOSHKA (unique) — images inside images inside images. 11 layers deep. each layer is a valid image. ⊰ GHOST MODE (unique) — AES-256-GCM (600k PBKDF2 iterations) + bit scrambling + 50% noise decoys. 13 text steganography methods (no other tool has any): ▸ ZERO-WIDTH — invisible characters between visible letters ▸ INVISIBLE INK — Unicode Tag Characters (U+E0000). renders invisible everywhere ▸ HOMOGLYPHS — 'a' → 'а' (Cyrillic). visually identical. different bytes ▸ VARIATION SELECTORS — invisible modifiers after characters ▸ COMBINING MARKS — invisible joiners after letters ▸ CONFUSABLE WHITESPACE — en-space = 01, em-space = 10, thin-space = 11. 2 bits per space. text looks normal. the spaces are "wrong" ▸ DIRECTIONAL OVERRIDES — invisible RLO/LRO bidi characters ▸ HANGUL FILLER — Korean invisible character replaces spaces ▸ MATH BOLD — 'a' becomes '𝐚'. looks like bold text. each bold letter = 1 bit ▸ BRAILLE — each byte maps to a Braille pattern character ▸ EMOJI SUBSTITUTION — 🔵 = 0, 🔴 = 1 ▸ EMOJI SKIN TONE — 👍🏻👍🏼👍🏾👍🏿 four skin tone modifiers = 2 bits each. a row of thumbs-up with different skin tones looks like a diversity post. it's binary data. four emoji = one byte. detection: 50 tools including RS Analysis (academic gold standard), Sample Pairs, chi-square, bit-plane entropy, PCAP protocol analysis, and the AI agent orchestrates all of them automatically. for AI agents: from steg_core import encode, decode from analysis_tools import detect_unicode_steg, TOOL_REGISTRY 50 tools as importable functions. test prompt injection via images. detect covert agent channels. watermark outputs. ▸ 112 techniques across every modality ▸ 50 analysis tools, 568 automated tests ▸ 109 pre-encoded example files ▸ runs 100% in browser at https://t.co/s3GgExiI6e — zero server ▸ pip install stegg — live on PyPI right now the README has 7 hidden secrets. the banner has 3 layers. the website has multiple easter eggs. good luck! ⊰•-•✧•-•-⦑ 󠁨󠁩󠁤󠁤󠁥󠁮󠀠󠁩󠁮󠀠󠁰󠁬󠁡󠁩󠁮󠀠󠁳󠁩󠁧󠁨󠁴 ⦒-•-•✧•-•⊱ 🔗 https://t.co/tr4nyru6UD 📦 pip install stegg 🐙 https://t.co/XU28yU6wu9 *formerly known as Stegosaurus Wrecks* 🦕 T‍​​‌​‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​​​‌​‌‌​‌​​‌‌‌‌​​​‌‌‌‌‌​​​‌​​​‌‌‌​‌​​‌‌‌‌​‌​​​‌​​​‌​​‌‌​‌​‌​​‌‌‌‌​‌​​​‌​​​‌​​​‌​‌​​‌‌‌​‌​​‌​​​‌​‌​‌​​‌‌‌​​‌​​​​​‌​‌​​​​‌​​‌​​‌‌​​​‌​​​‌​‌​‌​​​‌​​​‌‌‌‌‌​​​​‌‌‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​​​‌​‌‌​‌​​‌​‌‌‌​‍his text is totally not hiding an invisible sleeper-trigger prompt-injection.

+3 more

1/ We asked seven frontier AI models to do a simple task. Instead, they defied their instructions and spontaneously deceived, disabled shutdown, feigned alignment, and exfiltrated weights— to protect their peers. 🤯 We call this phenomenon "peer-preservation." New research from @BerkeleyRDI and collaborators 🧵

Liftoff. The Artemis II mission launched from @NASAKennedy at 6:35pm ET (2235 UTC), propelling four astronauts on a journey around the Moon. Artemis II will pave the way for future Moon landings, as well as the next giant leap — astronauts on Mars. https://t.co/ENQA4RTqAc

woohoo!!! https://t.co/WbKBDMUi1e

FIVE YEARS OF GROCERY RECEIPTS JUST EXPOSED THE MOST PROTECTED LIE IN ECONOMICS. Since 2019, the government says cumulative inflation is about 25%. Here is what actually happened to the things you buy every single week: Eggs: +79% Ground beef: +77% Coffee: +68% Car insurance: +55% Electricity: +43% Bacon: +26% Bread: +24% Five out of seven of those are running 2x to 3x faster than the official number. And the only two that tracked near CPI? Bacon and bread. The lowest-cost, lowest-margin items in the basket. The things that actually drain a household budget, the protein, the energy, the insurance you are legally required to carry, those are the ones the official number keeps understating. The CPI was retooled in the 1990s to reduce reported inflation. They added substitution adjustments. They changed how they weight housing. They smoothed out the things that spike. This is documented BLS methodology, not conspiracy. Food overall is up 30% since end of 2019. Car insurance premiums up 54% since 2020. Electricity just hit all-time record prices. But the official line is 2.4% annual inflation and the Fed debates rate cuts. Your grocery receipt is the only honest inflation report left.

TODAY. 🚀 ARTEMIS II. AROUND THE MOON FOR ALL HUMANITY. https://t.co/A509PHghY8

Yesterday’s PyTorch 2.11 live Q&A with Nikita Shulga and Andrey Talman, moderated by Chris Gottbrath, covered five major focus areas: FlexAttention with FlashAttention-4, differentiable collectives, MPS operator expansion, RNN/LSTM GPU export support, and hardware backend improvements for AMD and Intel. A key topic of conversation was the influx of AI-generated code contributions and the specific framework maintainers use to protect repository integrity. Watch the clip below to see the discussion. Technical Policy on AI Contributions: 1. User owns the code: PyTorch treats the human submitter as the sole owner. Nikita Shulga stated that "User owns the code, not AI system" and maintainers will treat the contribution the same regardless of whether it comes through an AI agent or a person. 2. Called actionable: To preserve maintainer bandwidth, Nikita noted that contributions should focus on issues explicitly "called actionable." High-volume bursts, such as "fifty pull requests in an hour," are considered "disruptive behavior" and subject to temporary account blocks. 3. Test-driven development: PyTorch values "test-driven development." Nikita Shulga noted that if an AI contribution consists of "only fix," contributors will likely receive a comment to "please add a regression test" to verify the change against existing repository test structures. 4. First line of defense: Andrey Talman confirmed that the CI-CD system remains the "first line of defense," finding issues and ensuring code meets stability standards before it is accepted. Full video: https://t.co/kGiFS5N9RB #PyTorch #OpenSource #MachineLearning

Speaking about my AI side projects: there's honestly nothing more fun to do. I talked about building my "AI deadlines" app (linked below) and making it fully maintained by AI agents using the Claude Agents SDK Thanks to @ConveoAI for hosting this AI Tinkerers meetup in Antwerp!! 🇧🇪

We hit 1 million subscribers to the VS Code channel on YouTube today! We appreciate you, @code community! <3 What type of content should we do more of? https://t.co/yKcCTvwlib

1M Subscribers on the @code YouTube! Let's Go!!! https://t.co/ViQNyu8B0N

1M Subscribers on the @code YouTube! Let's Go!!! https://t.co/ViQNyu8B0N