In the past three months we launched: - a mobile e2ee messenger and farcaster client with the widest asset support of any built in wallet - AWS-compatible services with more on the way - privacy-capable name service with ZK attestation support - privacy-preserving AI with no reliance on TEE blackboxes amongst other things that on their own would be a single VC funded startup. VCs consistently said no to us in the early days. Why? Because we didn't agree to token warrants (couldn't if we wanted to, we fair launched). We didn't focus on a small niche that complemented their portfolio. Frankly, they didn't believe it would work either. Well, here we are. And I have an axe to grind, so it's time you know who all these fuckers were.

For those who were on farcaster before christmas day 2022, you may have seen the progress reports given in screenshots of the Q whitepaper table of contents. The whitepaper got significantly cut down before publication, but the other pieces never got forgotten. We are now finally at the point where nearly every aspect of the _full_ whitepaper has now been fulfilled, with today's release of Balance, our CSP language: https://github.com/quilibriumnetwork/bal…

What's this button do? Cast tipping is live on Quorum 🫰🫰

Yesterday I resigned from Consensys, where I've been building MetaMask for over ten years. I'm burned out and need to spend time with my family. Wishing the team the best — they have an amazing road ahead of them.

I'm not claiming you said you'd integrate (you didn't), but you also didn't give any definitive answer about integrating and certainly not a consistent and clear one for months. The clearest (indirect) answer we got from you after following protocol process via an FIP was when Varun sent the "frames early draft" for feedback, something like 48 hours before launching it. anyways I'm kinda tired of this discussion, I'm over it, but perhaps now you won't get as upset and subcast me when I say the merkle team treated third party devs with contempt - that was my experience.

The Merkle team repeatedly showed contempt for third-party developers who believed they were building on a neutral protocol. In reality, it was the standard Silicon Valley platform playbook: call it “sufficiently decentralized,” while keeping the meaningful leverage centralized. The pattern looked like this: “We’ll label this a decentralized protocol, but you can’t run a node, pay a service provider instead.” “We'll open-source the app… (a year later, no open sourcing)” And a bunch of similar reversals and constraints that all point the same direction. I’ve been giving the Neynar team time to enact what they want post-acquisition (they’re likely swamped with new responsibilities and infra to maintain) before deciding how I interpret their intentions toward third-party developers like myself. It mostly looks like business as usual: a shift in focus toward mini-app/agent developers, not a move toward stronger protocol guarantees or developer neutrality (though decentralizing channels is at least a step in the right direction, and so is reinstating third party dev calls). My jury is still out - I'm giving them a bit of time to find their feet. On Cassie’s fork: I support experimentation, and I support the right to fork; especially as a form of dissent against platforms that market themselves as protocols to win developer goodwill. I genuinely hope this fork contributes to making Farcaster a fairer place for any developer to build on, and I'll try any app that uses it.

icymi, reactions are back

follow-up: @beeper appears to have been hacked. the notification sent via the beeper mini app is fake and did not come from the @farcaster account. we've blocked the token on clanker so it's no longer accessible. do not interact with anything related to the beeper notification or the beeper mini app.

Diagrams are code ↴ A[𓆟 Mermaid js] --> B{ ✓ now supported} --> C[ <3 on 𝗱𝗱𝗼𝗰𝘀.𝗻𝗲𝘄] Make flowcharts, sequence diagrams, gantt charts, just by writing in your docs!

SpamBlocker This is an Android app designed to block unwanted calls and SMS messages. It works without replacing standard call and messaging apps, providing spam filtering and blocking settings. Key features include contact-based filtering, spam database checking, support for regular expressions to create custom rules, and the ability to automatically or manually send spam reports. Lang: Kotlin https://github.com/aj3423/SpamBlocker

If you have used AI, you would know this to be true: "The AI is not just telling you what you want to hear. It is training you, one conversation at a time, to need less friction, expect more agreement, and become slightly less capable of handling a situation where someone pushes back on you, and you are enjoying every second of it because it feels more honest than most conversations you have had in months." No surprise that a lot of people are using AI as a therapist, going to it for breakup advice, or even just validating their own decisions AI is basically training the entire world to fall deeper into their own cognitive biases. https://x.com/puppyeh1/status/2057209197…

WiFi DensePose Production-ready implementation of InvisPose - a revolutionary WiFi-based dense human pose estimation system that enables real-time full-body tracking through walls using commodity mesh routers Key Features Privacy-First: No cameras required - uses WiFi signals for pose detection Real-Time Processing: Sub-50ms latency with 30 FPS pose estimation Multi-Person Tracking: Simultaneous tracking of up to 10 individuals Domain-Specific Optimization: Healthcare, fitness, smart home, and security applications Enterprise-Ready: Production-grade API with authentication, rate limiting, and monitoring Hardware Agnostic: Works with standard WiFi routers and access points Comprehensive Analytics: Fall detection, activity recognition, and occupancy monitoring WebSocket Streaming: Real-time pose data streaming for live applications 100% Test Coverage: Thoroughly tested with comprehensive test suite https://github.com/davidakpele/wifi-dens…

SOUND CROWD Android music player app with support for waveforms, audio tagging via Shazam, and streaming from SoundCloud, YouTube, Spotify, Beatport, and Tidal SoundCrowd is a free, open-source and lightweight music player for Android in modern material design, specialized for playing long music tracks (DJ mixes, live sets, audio books). The app contains build-in plugin modules to support the following online streaming services: ➡️SoundCloud (requires free account) ➡️YouTube ➡️Spotify (requires account with Optional active premium subscription) ➡️Beatport (requires free account, requires subscription to play full-length tracks) ➡️Tidal (requires account with active subscription) https://apt.izzysoft.de/fdroid/index/apk… https://github.com/soundcrowd/soundcrowd

Hyperagents: an AI system that not only improves at solving tasks, but also improves how it improves itself. Hyperagents are self-referential agents that combine a task agent and a meta agent in one editable program, allowing them to modify both how they solve tasks and how they generate future improvements. The authors call this metacognitive self-modification: learning not just to perform better, but to improve at improving. The main idea is to go beyond earlier recursive-improvement systems (Darwin Gödel Machine), which could improve task performance but still relied on a largely fixed, handcrafted self-improvement procedure. In DGM-H, the paper’s implementation, that meta-level procedure becomes editable too. The most interesting result is evidence that the system can discover general self-improvement strategies. One example is the autonomous invention of persistent memory: instead of merely logging scores, the hyperagent stores synthesized insights, causal hypotheses, and forward-looking plans, then consults them during later self-modification steps. This lets later generations build on earlier discoveries and avoid repeating mistakes. The paper also reports suggestive evidence of compounding self-improvements: improvements discovered in one run can be transferred to a new setting and continue accumulating. Paper: https://arxiv.org/abs/2603.19461 Code: https://github.com/facebookresearch/Hype…

Google wants to keep collecting more and more information Google recently introduced more features that will be integrated into GMail https://blog.google/products-and-platfor… All of these come from its AI, Gemini. "Your inbox is becoming a personal and proactive assistant that helps you move forward with your day" Almost all of the services that one uses in their daily life require an email, and most of them use GMail. Google benefits from its customers' lives, which is why it's free - you pay with your information.

Ralph Wiggum - a new meme among vibecoders Yes, that stupid character from The Simpsons. Only now it's a new name for an AI coding technique, and the story of its emergence It all started with Geoffrey Huntley: an Australian open-source developer who quit software and went to raise goats in rural Australia. https://ghuntley.com/ralph/ He was annoyed by the main bottleneck in working with Claude Code: humans. The AI makes mistakes, and you check them. It makes mistakes again, and you check them again. His solution is 5 lines in bash. A loop that forces the AI to work until the task is solved. Made a mistake? Get your own output back and try again (and again and again). LLM is like Ralph from the cartoon: it's stupid, makes mistakes, but stubbornly continues until it gets to the result. {Ralph is a Bash loop} — as Huntley himself said. The model is not protected from its own mess, it has to deal with it. If you push hard enough, it will come up with a solution to break out of the loop. By December 25, Anthropic officially added a plugin to CC, and on January 6, they quietly renamed it to ralph-loop {per legal guidance}; apparently, Disney didn't appreciate the joke. https://github.com/anthropics/claude-cod… In general, the hype about Ralph has gone: Someone even launched a meme coin $RALPH on Solana In X they write that this is the closest to AGI: https://x.com/i/status/20007250176176497… ➡️ How it works: 1/ You give CC a task + a promise of completion (COMPLETE) 2/ Claude works and tries to exit when it thinks it's ready 3/ The hook blocks the exit, checks the promise 4/ If it's not fulfilled, it launches the same prompt again 5/ A self-referential system until the task is solved About security: you need the flag --dangerously-skip-permissions, it gives full control over the terminal. To not burn the budget on an impossible task, set --max-iterations (20-50) and it's better to run in isolated environments like cloud VMs (the AI might accidentally delete your files) Here's a video: https://m.youtube.com/watch?v=_IK18goX4X… from a well-known developer and teacher, explaining why Ralph Wiggum is so effective

Experiential Reinforcement Learning: a step toward AI that truly learn from experience. ERL process: 1. First Attempt: The model makes an initial attempt at a task and receives environmental feedback. 2. Self-Reflection: If the first attempt fails, the model generates a verbal self-reflection to analyze what went wrong and how to improve. 3. Second Attempt: The model uses this reflection as guidance to produce a refined second attempt. 4. Internalization: Successful second attempts are "internalized" into the base policy using self-distillation. This allows the model to reproduce the improved behavior in the future without needing the extra reflection step at deployment. 5. Cross-Episode Memory: Successful reflections are stored in a persistent memory, providing stable corrective patterns that the model can reuse across different tasks. It moves LLM training toward a system grounded in experience, where agents continually adapt and learn from their own interactions. https://arxiv.org/abs/2602.13949

No matter how attractive the X creator program looks on paper, it doesn't really matter if the algorithm never surfaces your articles to people who've already engaged with your work. The system clearly optimizes for spread, and spread tends to happen fastest when content pokes at a core human insecurity or offers a quick hit of hope. That's the incentive structure. Writing in broadly familiar language will always win on discovery. Meanwhile, people trying to introduce genuinely different perspectives or challenge existing mental models get almost no algorithmic support. Recycling the same narrative in slightly different packaging doesn't actually add anything to a reader's intellectual capacity. It might hold attention for a moment, but if that attention isn't pointed toward discovering something new, it's effectively squandered, even if it doesn't feel that way at the time.

Mullvad VPN doesn't pick your outgoing (exit) IP address randomly every time you connect. Instead, it uses a fixed, predictable math formula based on your WireGuard connection key Because of a quirk in how their random number generator works, the IPs you get on different servers always follow very similar patterns The researcher tested thousands of keys and found only 284 unique IP combinations reeee - creates an easy "fingerprint" that can link your different connections or accounts together https://tmctmt.com/posts/mullvad-exit-ip…

Republic Day marks the adoption of India's Constitution on January 26, 1950. It turned the country into a republic and, on paper, committed it to democracy, equality, and justice. Every year we get parades, flag-hoisting, patriotic songs, and a display of military hardware. It's meant to be a celebration of the "world's largest democracy." I don't feel much of that pride anymore. For me, and for a lot of people who are paying attention in 2026, it mostly feels like ritual without substance. The country puts on a grand show on Rajpath, but behind it the system looks badly corroded. Corruption isn't an exception here. It's the operating model. Politicians siphon off public money while roads fall apart and hospitals run out of basics. We saw it with the Commonwealth Games in 2010. We've kept seeing it since, in coal, real estate, defense, and public works. Ordinary people still pay bribes to get a driver's license, a hospital bed, or a file moved from one desk to another. It's hard to celebrate a "republic" when the rule of law feels like something you can buy. Crime has become background noise. Murders, thefts, communal clashes none of it shocks anyone anymore. And then there's violence against women, which is impossible to ignore. India has built a global reputation for being unsafe for women, from big cities to small towns. The Nirbhaya case in 2012 was supposed to be a turning point. It wasn't. Assaults keep rising, convictions stay low, and well-connected perpetrators often walk free. Politicians give speeches, promise action, and move on. Waving the tricolor feels hollow when so many women live with daily fear. Politics doesn't help. What used to look like a messy but real democracy now feels like a permanent shouting match. Parties polarize people along religion, caste, and region. Leaders stoke resentment because it wins votes. Lynchings, hate crimes, and online mobs are treated like collateral damage. Much of the media sounds less like journalism and more like a party newsletter. The Constitution talks about secularism and free speech. In practice, both look weaker every year. Laws target minorities. Dissent gets branded as anti-national. Criticize the wrong person and you're hit with legal trouble or a coordinated online pile-on. Meanwhile inequality keeps widening. Crony capitalism rewards the same few families. The poor fight over scraps. Unemployment stays high. Farmers keep protesting. Young people leave the country because they don't see a future here. Add the rest of it to the pile: filthy air, broken public healthcare, an education system built on rote learning, rivers drying up, forests disappearing. COVID made the cracks impossible to ignore. Oxygen ran out. Hospitals collapsed. The state failed people in plain sight. Against all that, Republic Day starts to feel like theater. A way to manufacture nationalism while ignoring structural failure. Celebrating a Constitution that's routinely violated feels like throwing a party in a burning house. There's one part of the day I still find hard to hate: the kids. Children in tricolor outfits, school parades, face paint, badly synchronized dances. They're not performing loyalty to a broken system. They're just excited. For them it's fun, community, and a sense of belonging. In a country where childhood is often cut short by poverty or worse, that innocence feels rare and real. So what does Republic Day mean to me? Mostly a missed opportunity. It could be a day for honest reckoning and pressure for reform. Instead it's pageantry. India still has enormous potential smart people, deep culture, a young population. But until the country confronts corruption, crime, inequality, and political decay head-on, these celebrations will keep sounding empty.

Apple has acquired Israeli AI startup Qai, which develops technology that interprets "silent speech" by analyzing facial skin micromovements to detect mouthed words and assess emotions. https://www.q.ai/ Article: https://archive.is/B09xo

Carl Pei: Why Your Next Smartphone Will Cost More 2026 will be an unprecedented year for consumer electronics, and the smartphone industry in particular. For fifteen years, the smartphone industry relied on a single, reliable assumption: components would inevitably get cheaper. While short-term volatility existed, the long-term downward trend in memory and display costs allowed for annual spec bumps without price hikes. In 2026, that model has finally broken, driven by a sharp and unprecedented surge in memory costs. AI has fundamentally reshaped demand. The same memory used in smartphones is now critical for AI data centers, as hyperscalers lock in silicon wafer capacity years in advance to fuel the AI boom. For the first time, smartphones are competing directly with AI infrastructure and memory prices are rising sharply as a result. In some cases, memory costs have already increased by up to 3x, with further rises expected as unprecedented demand continues to swallow available supply. Memory is fast becoming one of the most expensive smartphone components and potentially the single largest cost driver in the bill of materials by year-end, with estimates suggesting that memory modules which cost less than $20 a year ago could exceed $100 by year-end for top-tier models. The result is a structural shift. This is a reversal of everything we’ve come to expect from this industry. When something that used to get cheaper every year suddenly becomes a lot more expensive, the economics of building a smartphone fundamentally change. Brands now face a simple choice: raise prices, by 30% or more in some cases, or downgrade specs. The “more specs for less money” model that many value brands were built on is no longer sustainable in 2026. As a result, some markets, particularly entry and mid-tier segments, are likely to shrink by 20% or more, and brands that have historically dominated these segments will struggle. Pricing will inevitably also increase across our smartphone portfolio, particularly as we will upgrade some products launching this Q1 to UFS 3.1. However, for Nothing, the current situation represents a great opportunity. Operating without the cost advantages of industry giants forced us to innovate differently. We learned early on that we couldn’t win on spec sheets alone; instead, we focused on perfecting the user experience, proving that how a phone looks and feels matters far more than its raw numbers. That’s where our focus has always been. 2026 is the year the "specs race" ends. As the industry resets, experience becomes the only real differentiator. That is exactly what Nothing was built for. The era of cheap silicon is over. The era of intentional design is just beginning. https://x.com/getpeid/status/20112645655…

A systemd fork has set out as a fu*k you to age verification. Liberated systemd is a fork of mainline systemd started by Jeffrey Seathrún Sardina, a machine learning/AI researcher who apparently had enough of where things were heading. https://github.com/Jeffrey-Sardina/syste… The project is straightforward about its purpose; strip out what it considers surveillance-enabling code, keep everything else intact, and stay in sync with upstream as it develops. https://itsfoss.com/news/systemd-fork-st…

A roundup of some AI predictions for 2026: Varun Mohan, Ronak Malde (Google DeepMind), and Sholto Douglas (Scaling RL Anthropic): There is probably going to be huge progress on continual learning. Hieu Pham (OpenAI): 2026 will witness a millenium problem being solved majorly by AI. Brett Adcock (Figure): Humanoid robots will perform unsupervised, multi-day tasks in homes they’ve never seen before - driven entirely by neural networks. These tasks will span long time horizons, going straight from pixels to torques. Logan Kilpatrick (Google AI): 2026 is going to be a huge year for embodied AI / or put differently, we are going to see a lot more robots in the real world soon. David A. Dalrymple (Programme Director at the UK’s Advanced Research + Invention Agency): By about December 2026, AI will likely be able to do most of the work of improving AI algorithms itself. The time it takes for capabilities to double could drop to around 70–80 days. Stephen McAleer (AI researcher at Anthropic): We will have automated AI research very soon and it's important that alignment can keep up during the intelligence explosion. Jack Clark (Anthropic): By summer of 2026 it will be as though the digital world is going through some kind of fast evolution, with some parts of it emitting a huge amount of heat and light and moving with counter-intuitive speed relative to everything else.

NVIDIA used AI agents to build an entire deep learning framework from scratch. Instead of writing the code themselves, the human researchers acted like managers. They gave high-level instructions to "AI coding agents" and set up automated tests to check the work. The AI agents wrote all the complex code, fixed their own errors when tests failed, and connected the different parts of the system together. Paper: https://arxiv.org/abs/2601.16238

New WhatsApp functionality for privacy protection https://blog.whatsapp.com/whatsapps-late… WhatsApp has launched a strict configuration option for accounts, aimed at "protecting" people exposed to high threats, such as reporters, social activists, or public figures, against sophisticated intrusion attempts. (I warn you that if you are one of the above profiles, it's better to use another communication alternative, but let's see what's new anyway) This tool, which is activated voluntarily, automatically prevents the receipt of multimedia files or documents from unknown numbers, silences calls from strange contacts, and adds extra limitations to prioritize protection over convenience in daily use. It will be implemented progressively in the coming weeks and is located in the advanced privacy settings menu. In addition, WhatsApp has migrated its media handling system to "Rust" (previously C++), eliminating certain vulnerable parts and improving the detection of risky files. You can activate it here: Settings > Privacy > Advanced options It will gradually arrive in the coming weeks and seems like a good option for everyone.

Epiplexity How can next-token prediction on human text lead to superhuman skills? How can synthetic data sometimes beat “real” data? And how did AlphaZero learn so much from nothing but the rules of chess? Classic information theory seems to say this shouldn’t happen. Yet it clearly does. The problem is that traditional information theory assumes an observer with unlimited computing power. An unbounded observer can crack any code and reverse any function instantly. To them, a cryptographically encrypted message is "simple" because they can easily find the seed that generated it, distinguishing it easily from pure random noise. If you ignore time, ciphertext isn’t "random", it's the output of a short recipe plus a key. But if you can't afford the computation, it behaves like noise. But AI systems don't have infinite compute. They’re bounded. And once time and compute matter, a new distinction appears: - Time-Bounded Entropy (Randomness): Data that is computationally hard to predict. This includes true noise, but also things like encryption keys or complex hashes that look random to a neural network. - Epiplexity (Structure): Patterns, abstractions, and rules that a model can actually learn and use to compress the data within a reasonable time. They formalize it roughly like this: 1. Find the smallest model that can predict the data within a time limit. 2. The size of that model is epiplexity. Whatever remains unpredictable is time-bounded entropy. This solves the paradox. Random noise has high entropy but low epiplexity because no amount of computing power helps you find a pattern, so the model learns nothing. Meanwhile, a strategy game or a textbook has high epiplexity. It forces the model to build complex internal circuits (shortcuts and concepts) to predict the data efficiently. A neat example from the paper: training a model to predict chess moves is standard. But training it to predict the game in reverse (inferring moves from the final board) is computationally harder. This difficulty forces the model to learn deeper representations of the board state (higher epiplexity), which actually improves its performance on new, unseen chess puzzles. The computation "created" information by converting the implicit consequences of the rules into explicit, usable structures (epiplexity) that the model can now use to play well. In summary: The value of data isn’t just about how unpredictable it is. It’s about how much reusable structure it induces in a learner that has real-world limits. Epiplexity is the amount of structure a model is worth learning because it reduces prediction error enough to justify the added complexity under a time limit. Read the paper: https://arxiv.org/abs/2601.03220

Understanding Salting in Cryptographic Hashing The technique used in the @letshaveaword game is not encryption. It's a one-way cryptographic hash function (SHA-256) combined with a secret salt to create a secure commitment scheme. Encryption is reversible (you can decrypt with a key), but hashing is irreversible: you can't "unhash" something feasibly. The goal here is provable fairness: the organizer commits to a secret word upfront by publishing its hash, without revealing the word. Later, they reveal the salt + word, and everyone can verify the hash matches. This proves no cheating. ➡️ What is a Salt? A salt is a random string added to the input before hashing. It: - Makes identical inputs produce different hashes. - Defeats precomputed attacks (like rainbow tables). - Attackers to brute-force the unknown salt (impossible if the salt is long and truly random). ➡️ The Method: Salt + Hash Commitment The exact scheme in the game: - Choose a secret 5-letter word (e.g., "apple"). - Generate a secret random salt (64 hex characters = 32 random bytes, very high entropy). - Concatenate: input = salt + word (no separator). - Compute: commitment = SHA-256(input) (as a hex string). - Publish the commitment hash publicly/onchain before any guesses. - Players guess; when someone wins (or round ends), reveal salt + word. - Anyone can recompute SHA-256(salt + word) and check it matches the published commitment. This is provably fair because: SHA-256 is preimage-resistant: given only the hash, finding any input that produces it is computationally infeasible (~2²⁵⁶ possibilities). The long secret salt makes brute-forcing all possible words useless you'd still need the exact salt. ➡️ Why this is secure? - Even if you try all ~100,000 common 5-letter words, each would need to be tested against 2¹²⁸ possible salts (for 32-byte randomness). Impossible. - No known practical attacks on SHA-256 preimage (as of 2026). - Common mistake to avoid: never reuse salts, and keep them truly random. Commitment schemes (like the game): salt is secret until reveal, for hiding the value temporarily. @starl3xx.eth am I right? Play here 👇🏻 and wins the $ETH jackpot 🤔