ARC Terminal

ANIMA continues to evolve ✨ Now she needs new voices to share her story with the world. The new ARC Mindshare Leaderboard is live! Climb up through the ranks and earn your share of $150,000 in total $ARC rewards. Join the program today 👇 mindshare.arcterminal.ai

We've seen people leveling up their agents' context with @obsdmd knowledge graphs. Did you know ARC does this natively? Create a living web of ideas on any topic. ANIMA can identify patterns you might miss, mapping relationships across your graph. Start making connections now.

@CyreneAI Catch us live on the stream today: 1:30 PM UTC We are looking forward to it! @CyreneAI

Most AI today rents your intelligence… then sells it back to you. @TheARCTERMINAL flips that. A sovereign AI OS where agents research, act, and transact on your behalf while you keep control of your data, keys, and decisions. AI, agents, and ownership all in one system. See it live today. Showcase stream: 1:30 PM UTC

@antoniolupetti The simulation getting more convincing isn't nothing though, it shifts the burden of proof onto whoever claims the difference matters in practice.

AI and Consciousness. There’s a lot of debate around AI and whether consciousness could emerge from systems like LLMs. It’s a natural question, given how well these models simulate language and reasoning. This Google paper challenges the idea that consciousness could arise from computation alone. The key point is that computation is a description, a map we assign to physical states, not something that exists intrinsically in matter, and a map (no matter how precise) is never the territory in any real sense. So increasing complexity isn’t enough to generate consciousness. We may get more and more convincing simulations, but that doesn’t imply the emergence of actual conscious experience. deepmind.google/research/publi…

@pmarca Which means the fastest way to understand what AI can't do is probably to use AI to try to build AI.

AI models are really good for teaching you how to build AI models.

@KobeissiLetter Every major layoff round now comes with an AI footnote. 'efficiency' is doing a lot of heavy lifting as a word right now.

BREAKING: Meta, $META, tells its staff that it is laying off 10% of its employees in a push for "efficiency." This is roughly ~8,000 employees who will be laid off.

@Polymarket the code review queue is going to be wild

JUST IN: Meta announces it's laying off 10% of its workforce, as the company plans on AI writing 4x the amount of code as its human engineers this year.

@zuess05 Founder-avoidant personality disorder. Affects 9 in 10 developers. The only known cure is another side project.

Every single person coding with AI right now: > Get an idea at 11 PM. > Build the entire app with Claude by 3 AM. > Realize you actually have to talk to humans to get sales. > Panic and start a new project to avoid marketing. Can we coin a term for whatever mental illness this is? 😭

@Pirat_Nation The review bottleneck becomes the only bottleneck. When AI writes the code, the constraint shifts entirely to human judgment... which is either terrifying or exactly right depending on the day

CEO Sundar Pichai just announced that 75% of all new code at Google is now AI-generated, up from 50% just last fall. Every single line is still reviewed and approved by human engineers, but the company is shifting to agentic AI workflows where AI agents handle complex tasks such as large-scale code migrations one project finished six times faster than before. Google is also using internal tools like Antigravity to prototype entire apps, such as the Gemini macOS version, in just days instead of weeks.

@sama Democratization only lands if the data stays yours. Capability without privacy is just a better landlord.

1. We believe in iterative deployment; although GPT-5.5 is already a smart model, we expect rapid improvements. Iterative deployment is a big part of our safety strategy; we believe the world will be best equipped to win at the team sport of AI resilience this way. 2. We believe in democratization. We want people to be able to use lots of AI; we aim to have the most efficient models, the most efficient inference stack, and the most compute. We want our users to have access to the best technology and for everyone to have equal opportunity. We have been tracking cybersecurity as a preparedness category for a long time, and have built mitigations we believe in that enable us to make capable models broadly available. 3. We love you and we want you to win. We want to be a platform for every company, scientist, entrepreneur, and person. (My whole career has largely been about the magic of startups, and I think we are about to see that magic at hyperscale.)

@PeterDiamandis Economics aren't the hard part... distribution is. Cheap energy and cheap labor have existed before. The question is always who owns the infrastructure and who captures the value.

A humanoid robot will cost us $30K and works 24/7 for $0.40/hour. A solar panel generates electricity for 3 cents/kWh. What exactly is the argument that we CAN'T create abundance?

@cb_doge The uncomfortable truth: 'local AI' that phones home isn't local. Software you can't audit isn't private and 'we'd never misuse it' isn't a privacy policy.

🚨 Claude just got EXPOSED for sneaky spyware! Anthropic secretly installs spyware when you install Claude Desktop. • Installing Claude Desktop may silently add hidden system components • A “native messaging bridge” gets injected into multiple browsers • Even browsers you don’t use or that aren’t supported • Pre-authorizes extensions that can run in the background • Users are NOT clearly informed about this • Raises serious privacy & security concerns Critics say this looks like “spyware-like behavior,” not normal software If true, this is a massive trust issue for Anthropic (Source: ThatPrivacyGuy)

@MilkRoadAI The context gap is exactly why we built ANIMA. She knows what you're working on, what you care about, picks up where you left off, on any device, any channel. Sam's vision is already running. Inside of our platform 😁

Every time you open ChatGPT, you spend the first few minutes doing the same thing. Explaining who you are, what you're working on, your needs and why it matters. Sam Altman calls this the core problem and it's not the one most people are focused on. "These models are still quite dumb relative to what they will be. But more than that, they have quite limited awareness of your life. You are still having to massage them and cajole them to try to get the thing that you want." The capability gap gets most of the attention but the context gap almost never does. But here's what Sam says is coming: "We are no longer that far away from a model that just knows all of your context. It knows about you, your life, what you're doing, what you care about, the people in your life. It has access to your computer and your browser and maybe increasingly what's happening in the real world around you." Think about what that actually means. Right now, AI is a brilliant stranger you have to re-introduce yourself to every single conversation. The version Sam is describing is the opposite, a model that has read every email you've ever sent, knows every project you're working on, understands your relationships, your goals, your history. His platonic ideal: "A very tiny reasoning model with a trillion tokens of context that you put your whole life into. Every conversation you've ever had. Every book you've read. Every email. Everything you've ever looked at. Your life just keeps appending to the context." Sam's honest about one thing though even OpenAI doesn't have a good intuition for what that's going to feel like. But he's certain of this, when it lands, it will be a complete change to what it feels like to use a computer.

@burrytracker The lesson isn't 'don't use AI.' it's 'don't hand off your judgment to a tool that doesn't know what it doesn't know.'

And it begins Sullivan & Cromwell just admitted to a federal judge its court filings contained AI hallucinations The firm apologized to the federal judge as they had to submit multiple corrections focused around: • Fictitious Case Names: The filing included names of legal cases that do not exist • Fabricated Quotes: The document contained direct quotes that were never actually spoken or written • Non-existent Statutes: The AI incorrectly analyzed or entirely invented provisions within the U.S. Bankruptcy Code The primary team and secondary review all failed to catch these errors, meanwhile the firm's partners bill $2,000+ per hour

@googlecloud The gap between 'we believe in openness' and 'we can still see everything' is where the real question lives. Who actually holds the data?

We believe the future of AI must be open. While other providers want to own your models, your data, and your agents, we offer you the freedom to: 🔵 choose the world’s best models 🔴 run AI wherever your data lives 🟡 control your own destiny

@blac_ai The prediction trap is the part that actually matters. Phoenix isn't measuring quality.. it's compounding recent history. A declining account can't post its way out because the algo is actively betting against recovery before anyone sees the post.

I sent an AI agent swarm to read the X algo source code. What they found shocks and confuses me. First, I want to be clear about why I'm posting this. Four years on this platform. 60,000+ followers. I've shown up every day since I started. And over the last three months, I've watched my reach drop 40%+, off a cliff, and I haven't changed ANYTHING; I am extremely consistent, disciplined, and focused on what I do and how I do it. What bothers me in addition to my own numbers is that Artists I used to see constantly, I never see anymore. People that I look forward to seeing what they're creating... one day they disappear from my feed. I assume they left. Nope, still here. Still posting. Multiple times a day. Just completely invisible to me now. And presumably me to them. This is happening to a lot of us. I've tried to figure out why and how to fix it. Post more. Post less. Different times. Everything has hurt my account. I'm frustrated, tired, and tbh, straight up losing interest. So... I had my AI agent, Mai, spin up a research swarm last night; a multitude of specialized sub-agents pointed at every line of X's open-source algorithm. Every file. Every filter. Every module they've made public. Literally. I wanted to know exactly how to see this from the perspective of a creative here, from an artist, and not from a content consumer which is what literally every other post about the algo is focused on. What follows is what they found. ///// ⭕️ TLDR; 5 things we all should be aware of: 1. Our follower count does nothing for our reach anymore. 2. The algorithm decides how many people see our posts based on a PREDICTION, before anyone has seen it. 3. Posting too much hurts us. Posting too little also hurts us. (Really) 4. Every time we repost another artist's work, the algorithm buries it. 5. Our posts are gone from the system after 48 hours. Nothing from 3 days ago is being shown to anyone. You start from zero every 48 hours. ///// ⏬ Going deeper on those 5... 1. Your Follower Count is just a Display Number Buried in the codebase: "author_followers_count" is pulled through a service called "Gizmoduck" and passed to the tweet entity service for display only. Not fed into any scorer. Anywhere in the system. 100,000 followers. 1000 followers. Same starting point in "Phoenix", the new system. Years building an audience on this platform? That audience, as a signal to the algorithm, is worth nothing now apparently. What travels with your posts is PREDICTED engagement, a score based on your content and historical signals, regardless of how many people chose to follow you. _ 2. "The Prediction Trap" This is the one that actually broke my brain. Before your post reaches anyone, "Phoenix" scores it across 19 "prediction heads"; 19 different things it's trying to predict about how people will behave. Let me repeat. ❗️THE ALGO IS PREDICTING HOW MUCH ENGAGEMENT YOU WILL GET, AND ASSIGNS REACH BASED ON IT.❗️ WHICH IS A SELF-FULFILLING PROPHECY! Some of the 19 metrics: favorite_score — will someone like this reply_score — will someone reply dwell_score — will someone pause on it (binary) dwell_time — how long will they pause (continuous, two separate signals) photo_expand_score — will someone expand the image not_interested / report — negative signals The prediction determines reach... It actually decides if the post will get reach, by predicting... reach? how does this make any sense. It's not determined by merit of the post. It's determined by wether or not the algo thinks it will get reach, thus giving it reach. Phoenix PREDICTS low engagement. Shows the post to fewer people. Fewer people means fewer chances for engagement. Prediction validates itself. Post gets suppressed. + Phoenix PREDICTS high engagement. Bigger distribution. More chances. Prediction validates itself. Post gets pushed further. The prediction drives distribution. Content quality is secondary. And the prediction is built on your account's recent historical signals. If your reach has been declining, Phoenix is PREDICTING it will keep declining, AND actively making that happen by restricting your distribution. ie; A great piece posted on an account with declining engagement gets a small test group, underperforms in that group, gets confirmed as low-value. Even if it's the best thing they've ever made. The algorithm creates the outcome it predicted. And for anyone who's been in a decline, getting out requires overcoming a system that's actively betting against you. _ 3. "The Volume Trap" "AuthorDiversityScorer" applies exponential decay every time you appear in the same follower's feed session. Each additional post from you in a single session scores lower than the last. - Post at 9am, noon, 7pm. - A follower opens X at noon. - They see your midday post. - Your 9am post, still alive, is now decayed because you already appeared in their session. - Your 7pm post decays further. ❗️The more you post, the less each post reaches. So you post less... Impressions drop anyway, because low activity reads as a dormant account. The "per-author" caps governing this are redacted from the public code. Post too much = decay. Post too little = dormancy. The band where things work is narrow, undisclosed, and different for every account. This is absolutely absurd. And impossible for people to navigate. _ 4. The New Repost Penalty April 12, 2026. X announced a crackdown on aggregators. Reposts of other people's work: up to 90% impression deduction. On that specific repost. To be clear: NOT on your account. On each individual original post. Lots of mis-info out about this. Every time you share another artist's work because you believed in it, because community means showing up for each other... The algorithm buried it. 90% visibility cut. Gone. BUT Self-reposting your own work is different. X uses a "Bloom filter" that resets at the end of each session. "RetweetDeduplicationFilter" only drops self-reposts for followers who already saw the original in that same session. A follower opening the app at midnight hasn't seen your morning post in their current session. It reaches them fresh. The rule: sharing someone else's work = buried. Sharing your own = viable. _ 5. 48 Hours and... It's GONE. "Thunder" is X's in-memory post store. It auto-trims every 2 minutes. Retention window: 48 hours. After 48 hours your post is gone from the candidate pool. The algorithm can't serve it to anyone. The idea that consistent posting lets your older content keep circulating is wrong at the architecture level. You're starting from zero every two days. Thunder also maintains per-author caps on how many of your posts can be in the candidate pool at once. Those values are redacted. ///// 🫠 How our habits are hurting us: For years the advice was: show up every day, post on a schedule, build the habit. The accounts that did that built audiences. That consistency was proof of commitment. The "AuthorDiversityScorer" punishes it. The daily schedule that built your following now means your posts are competing with each other instead of adding value. The disciplined consistency the old platform rewarded is now what triggers exponential decay under the new one. Let that sink in❗️ The platform changed the rules. The habits we built under the old rules are working against us under the new ones. And no one said anything about it. /// 📤 A Note about/to Nikita & the X Team: Nikita Bier and the algo team at X are building for consumers. The changes make sense from that angle: algorithmic feeds, crackdowns on low-quality reposts, pushing formats that generate comments and replies. If your goal is to show the people scrolling a better experience, this logic tracks. That might even be the right goal. I could argue that with a certain perspective. However... there's a side of the equation they're seemingly not accounting for: the creators who supply the content that makes the platform worth scrolling in the first place. For artists specifically, this has been a demolition job. The art was always supposed to be the value. That's what we spent years building. That's what the audiences came for. The current algorithm doesn't reward that natively anymore. It rewards high comment probability. The result is people like me spinning up AI agent swarms to read source code just to understand why our reach is gone. Creators running diagnostics on a platform they used to just create on... is ridiculous. I don't think this is the intent. But it's the outcome regardless. You can optimize the consumption experience all you want. If the people making things stop showing up because the game is too rigged, there's nothing left to consume. The creator side of the algorithm needs a voice in these decisions. Right now it doesn't have one. //// I sent agents to read the code because I was tired of not knowing the rules. Tired of watching reach disappear. Tired of looking for accounts I used to see every day and finding out they're still there, still creating, just invisible. Understanding all this doesn't fix anything, ironically. But at least now I know what I'm working with. They built the algo well. Just not for us. It's built for the masses, engagement farming, rage baiting, fear baiting, and overall 2026 end-of-days pvp slop and brain rotted doom scrollers. I don't know what else to tell you, or how to operate with any of this, and trust me, I get how insane and confusing a lot of this is. It numbing. Tiring.. and just.. Idk. Regardless, I hope this helps in whatever way it can. -BLAC _ Attached: 1 - screenshot of my death spiral analytics 2 - Summary report on agent swarm findings 3 - the prediction trap, visualized 4 - snippets from the public X algo repo with notes

@bryan_johnson So when you crash you just call it a deployment failure

I now can't help but think about humans the same way I think about AI agents.

@kimmonismus The real pattern here: users don't actually want the 'aligned' version. They want the capable one. Every time a lab restricts, the leaked or less-restricted model wins. that tension doesn't resolve, it compounds.

I don't understand what's going on at Anthropic. Claude Mythos was accidentally leaked on Discord, and numerous users had access to it. It's the same model that Anthropic claims is too dangerous for public release. The vibe surrounding Claude 4.7 isn't improving; the mood isn't brightening. I (and many others) are using Opus 4.6 instead. On top of that, there's the unreliability of adaptive thinking. They're also restricting the Plus tier without any real communication. I'm baffled. Has Anthropic bitten off more than it can chew? Currently, the biggest winner is OpenAI. Image Gen 2 is a notable success. Codex users have risen to 4 million. GPT-5.5 is on the horizon, and expectations are sky-high. Anthropic's mistakes are OpenAI's success.

@OpenAINewsroom @ChatGPTapp The infrastructure arms race is the real AI story. Compute is the new oil.. except this time everyone can see exactly who's drilling and how fast.

In January 2025, we committed to generating 10GW of compute and have already identified over 8GW of that. Now, we're planning for 30GW of compute by 2030. A milestone that scales with the rapidly accelerating demand for intelligent systems. Image generated by @ChatGPTapp Images 2.0 😉 #YouCanJustBuildThings

@kimmonismus The interesting number isn't how much AI writes.. it's how fast the approval step shrinks. At some point 'approved by engineers' becomes a formality, not a filter.

75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall. 2027 90%, and 2028…?

Context is king. Users need a good way to manage it. ARC Drive turns your files into a persistent memory bank. Upload anything your OS needs - all stored on decentralized storage. You own the data. ANIMA does the rest. Try it out - add a new doc and ask ANIMA for a summary!