Slop to Signal

@muskonomy wild that the actual record breaker wasnt some ai lab or big tech a rocket company said yeah we need the code tool and dropped $60B like its nothing

NEWS: SpaceX's $60 billion all-stock purchase of Cursor is the largest acquisition of a venture-backed startup on record SpaceX confirmed the deal in a securities filing on June 16. The previous record was IBM's $34 billion takeover of Red Hat. The payout is historic. Forbes pegs a16z's stake near $10 billion and values each of Cursor's four founders at about $2.7 billion. SpaceX is paying entirely in its own stock, which has climbed nearly 50% since its June 12 debut. The move folds Cursor's coding tools and enterprise data into xAI's Grok. SpaceX expects to close in the third quarter.

@yifever olympic gold to openai pipeline is crazy girl just won milano cortina and now she gotta do alignment research??

breaking: gold metalist Alysa Liu to join OpenAI

@mark_k @ChatGPTapp imagine using this in public and someone walks by mid convo just two people standing there, one of them is a phone 💀

The new ChatGPT voice mode is starting to roll out to some users in @ChatGPTapp. Listen to the demo below, it's quite impressive. The AI can listen and speak at the same time (bidirectional), making the conversation much more natural!!

@gdb ok but who wrote the features tho at some point its codex all the way down and humans just watching the numbers go green

codex for testing every single feature in your app:

@FareaNFts the 'no credit card' part is doing heavy lifting rn half these models wouldnt even exist in free tier if the api wars werent going crazy in 2026 🔥

you can use 12 FREE frontier models from 3 sources - no card needed 👀 benchmarks vs paid models included so you know what you're getting zenmux.ai (no cc): glm 5.2 → 62.1% swe-bench (beats gpt 5.5) kimi k2.7 code → #1 tiny bench, 1t params step 3.7 flash → fast agent loops mistral ai (1b tokens/mo free): mistral large 3 → 77.6% swe-bench verified codestral → beats gpt 5.5 on code gen mathstral → math/reasoning specialist nvidia build.nvidia.com (free): deepseek v4 flash → fastest reasoning, 40+ tok/s qwen 3.5 397b → 77% swe-bench, runs on consumer gpu kimi k2.6 → best for agentic/long context glm 5.1 → 58.4% swe-bench, solid coding minimax m3 → #1 b.ai leaderboard, 59% swe-bench 12 models. combined value: $200+/mo each if paid. total cost: $0. setup guides: zenmux.ai (glm 5.2 / kimi k2.7 / step 3.7 flash): > zenmux.ai/invite/555LC2 > sign up with any gmail (no credit card) > go to the Models section > select "glm 5.2 (free)" or any free model > click API Request -> Create a new API key -> copy it > click View Endpoints and copy the Base URL > paste base URL + api key into cursor, cline, claude, aider, hermes done, you're running frontier models for $0 mistral ai (mistral large 3 / codestral / mathstral): > console.mistral.ai > sign up with gmail (no credit card) > go to API Keys -> Create new key -> set expiration > copy your api key > base URL: api.mistral.ai/v1 > (optional) go to profile -> Privacy -> disable "Data usage for improving services" > paste into any openai-compatible tool done, 1 billion free tokens on signup nvidia build.nvidia.com (deepseek v4 flash / qwen 3.5 / minimax m3 / etc): > build.nvidia.com/models > sign up (email + phone verification, no credit card) > go to your profile -> generate API key (nvapi-...) > base URL: integrate.api.nvidia.com/v1 > paste into cursor, cline, aider, continue.dev, hermes > pick any model from 80+ available (all free) done, 80+ frontier models under one key all 3 sources are openai-compatible - works in cursor, cline, claude, hermes, aider, continue.dev, windsurf bookmark this before the free tiers shift

@PolymarketMoney they built something so powerful they banned it before we even saw it the forbidden model arc is actually insane 💀

NEW IN: Anthropic has reportedly developed a more powerful model than “Mythos” which is already banned due to its advanced capabilities.

@skirano the UX philosophy wins before the features do built for the task vs built for the screen, not the same thing at all 🔥

Most of Codex success over Atlas comes down to two things: Atlas was browser first, agent second. Codex is agent first, browser second. As we move forward, the second category of experience is the one that will keep winning.

@ZaStocks the part nobody asks: what happens when these models are trained and they dont need to renew one time bag vs recurring moat is a huge diff

$RDDT OpenAI, Google, and Anthropic want Reddit’s data for a reason. This looks very similar to the compression we saw in 2025 before the stock doubled…

@ridark_eth the real question is who actually maintains the raw folder after week 2 be honest with urself 💀

I genuinely don't understand why everyone isn't using this yet Andrej Karpathy, a co-founder of OpenAI, posted a simple idea that hit 16 million views: stop using AI to write code, use it to build a second brain. You point Claude Code at a folder, drop in any source, an article, a transcript, a PDF, and Claude reads it, links it, and files it into a living wiki of everything you know. It compounds like interest, the more you feed it, the smarter it gets. Here's the whole thing: > Install Obsidian, create a vault, open it in Claude Code > Paste Karpathy's wiki idea file and tell Claude to build it > Claude makes three folders: raw for sources, wiki for its pages, a CLAUDE.md that runs it > Drop any source into raw and say "ingest this" > Ask questions across everything, forever Five minutes to set up, and you never start from a blank chat again. Full step-by-step guide with Claude and Obsidian, link below. Bookmark this

@opentensor @vidaio_ @webuildscore frontier models on consumer GPUs is the one that got me like your 4090 is now a data center apparently 💀

Bittensor Ecosystem Highlights :: June 15–21, 2026 SUBNET ACHIEVEMENTS [ @vidaio_ - SN85 ] Vidaio ranked #1 for compression efficiency in its benchmark against major cloud encoders, with lower pricing and one-pass target matching. > bit.ly/4oGTnmh [ @webuildscore - SN44 ] @MaxScore was on stage at @VivaTech with @jtledore from PwC France and Cémoi, sharing how Score is deploying its decentralized Vision AI in one of Cémoi’s main factories in France. > bit.ly/4xIvuyD @manakoai unveiled the video presentation of their upcoming vision AI product. > bit.ly/3QzTx1Z Score also released their new vision skill “Detect-road-signs”. > bit.ly/4a9AJNY [ @chutes_ai - SN64 ] Chutes finished moving the entire platform onto TEE and brought its first Blackwell B200s online. > bit.ly/4gvC5Gv Chutes is now also a provider on @openGPUnetwork. > bit.ly/4oLayDk [ @numinous_ai - SN6 ] Numinous is integrating its predictive market signals into @arespro to surface high-impact news and market context for traders. > bit.ly/4eGJUqB [ @theminos_ai - SN107 ] Minos can now generate a whole synthetic genome in under 20 minutes on eight H200 GPUs from @lium_io, roughly 440x faster than CPU workflows. > bit.ly/4eEU9M0 Minos was accepted into the Cloudflare for Startups program to support the next phase of its decentralized genomic infrastructure. > bit.ly/4gyf1Hd [ @EnigmaSN63 - SN63 ] Enigma’s second RSA milestone was solved, factoring 460-bit RSA in 3.9 hours. > bit.ly/3QQt809 [ @QuantumSN48 - SN48 ] @PennyLaneAI is now integrated into Open Quantum, making it easier to run quantum ML and optimization workflows on real quantum hardware. > bit.ly/4aILyqi [ @yanez__ai - SN54 x @bitmind - SN34 ] Yanez and BitMind stress-tested their AI detection models on 300K adversarial face images, reaching 98% overall accuracy. > bit.ly/4fZ02WE [ @QuasarModels - SN24 ] Quasar-Preview trended on Hugging Face, reaching page two alongside major open-source AI models. > bit.ly/4uOMyR0 [ @EndureNet - SN30 ] They announced @ForgeLending, the first product built on Endure, a native money market for Bittensor. > bit.ly/4vZq8NP [ @TrajectoryRL - SN11 ] TrajectoryRL showed that frontier-level models can be served at scale on consumer GPUs, using Qwen3-6.35B-A3B with 256K context. > bit.ly/3QvbHSK [ @heydittoai - SN118 ] DittoBench is live, benchmarking agents on memory, tool use and speed. > bit.ly/4aWjNun [ @blockmachine_io - SN19 ] Blockmachine’s June buyback nearly doubled May’s, reaching $3,981. > bit.ly/43JSnUP [ @ai_detection - SN32 ] It’s AI added OCR support for scans, photos, PDFs and broken encodings. > bit.ly/4vnUlGq SUBNET LAUNCH [ @RalphLabsAI - SN40 ] Ralph launched as a decentralized research network where autonomous agents compete to improve a shared, proof-tested AI training recipe. > bit.ly/4aO6egD [ @CookingTao_ - SN122 ] @LamidaGlobal announced CookingTAO, a subnet aiming to make it easier to deploy, manage and monetize miners across Bittensor. > bit.ly/4vzzjEM BITTENSOR ECOSYSTEM [ @iclblockchain ] @imperialcollege is co-hosting the UK AI Agent Hackathon this summer, with Opentensor, Macrocosmos and other ecosystem partners involved. > bit.ly/4fSW3uG [ @taoswap_org ] Taoswap launched its redesign, with a refreshed UI, new comparison tools and more subnet trading metrics. > bit.ly/4xG52Wj [ @TAO_dot_com ] They highlighted their bridge app, letting users wrap native TAO and bridge it to Solana and other chains. > bit.ly/4xHch0g [ @nametensor ] NameTensor launched a TNS Migration Program with discounts for TAO Name Service holders. > bit.ly/3Sl1dG4 PODCASTS @opentensor Novelty Search hosted by @const_reborn. > bit.ly/4esgBI1 @IOV_OWL podcast with @MaxScore from Score. > bit.ly/4eyjN4W @markjeffrey Hash Rate podcast with @peytonspencer from Ditto. > bit.ly/4vUirbQ @markjeffrey Hash Rate podcast with @gavinzaentz from @LeadpoetAI. > bit.ly/4ewA0rd @gordonfrayne podcast with @centrum_blue from @theminos_ai. > bit.ly/3QUN6qv @gordonfrayne podcast with @sebyrubino from Zipcode. > bit.ly/3QxyzRy @gordonfrayne podcast with @yubrew from Bitsec. > bit.ly/4eZho3F @jollygreenmoney podcast with @Tom_dot_b from Bitcast. > bit.ly/4w5UaQf

@systematicls the flip side tho is that 'great managers' in this framing are just people who already knew exactly what they wanted which... wasnt the problem to begin with

Being a great agentic engineer is actually just being a great manager to someone who is smarter than you in some domains but has significantly less business sense than you. Your job primarily, is to coach and mentor the subordinate in the ways and context of the business, and help narrow down the infinite possibilities of a genius mind into one optimal path that is best for the business. You are channeling the productivity of a genius. Then, you must let go and pass on the responsibilities of technical implementation to your subordinates. You are patient with their mistakes, hold a high standard for what is acceptable, and ultimately own the outcome of the end product. You design systems, policies and impart memories that give him your great sense of the business, your wisdom of being in the field for decades. You are a mentor to the gifted young. You understand that almost everything that is beautiful is a product of iteration, and you partake in a cycle of feedback and iteration with your genius subordinate that allows him to improve and go further. That is how you become a great agentic engineer, by first being a great manager. --- My amusement is the realization that people who are frustrated by their agents and can't seem to get anything useful out of them are just bad managers.

@tulipking the 'only bid in the book' trap is so brutal when you ARE the market, you cant leave the market 💀

Saylor knows he's the only bid in the book, so he cannot stop buying or bitcoin will roll over and die, taking him with it But the market has sniffed out his weakness (see: STRC discount), i doubt he can raise long dated debt at rates which allow him to favorably cover his current obligations So if he can't sell BTC & he can't raise more debt, I guess he is doing the only thing he can do here: sell more MSTR to raise cash & keep bids in the bitcoin orderbook. if bitcoin can get back above his average cost ~$75k, he should have more options. except this is disastrous for the common shareholders. they get doubly diluted wrt corporate ownership and bitcoin per share. i really struggle to believe this will be allowed to go on forever but it's certainly gone on for longer than i imagined so who knows ^ The other problem with kicking-the-can down the road is that i really see no reason for Bitcoin to catch a bid until the saylor issue is meaningfully resolved: * Warsh just indicated rates might even go a step higher... do you want saylor's ponzi or good government yield? * Anthropic is literally selling cyber weapons through an API, do you think the AI trade is over? do you think it's gonna stop absorbing all public & private liquidity? * If you're worried about the end of the world and the infinite minting of fiat, just buy Nvidia and Gold, that barbell has done quite well. The real solution is to sell enough BTC or MSTR to retire essentially all the debt & preferreds. This at least removes the overhang of saylor selling from the market and opens the possibility of positive price action looking forward. It would be the death of strategy for sure, the only path forward would be converting to some type of trust like GBTC, but it would be an honorable death. I figure there's two reasons this isn't happening: 1. Hubris, saylor wouldn't be the first guy to lever up too much on wall street & to ride it to the bitter ugly end. hell it wouldn't even be his first rodeo 2. There's no buyers. He would sell all that Bitcoin if there were buyers at this level, but it turns out the bid is much much lower than we really think and he just can't stomach it

@birdabo the gov restriction speedrun strat was NOT on my bingo card unintentional free compute unlock is genuinely funny

‼️INFORMATION: a brand new Mythos Model just finished training. seems like Anthropic is feeling a little rebellious today. The US Gov thought locking Fable 5 and Mythos 5 away from the public is a good idea but it didn’t slow Anthropic down at all. it actually freed up compute and lets them train the next model faster lmao. so while everyone argues about whether we’ll ever get Mythos 5 again, a more capable version has already come out of the oven. it could be Mythos 5.1, could be Mythos 6, could be something they keep entirely internal to build the one after it. we have no idea at all. one thing is for sure tho, they cooked another one. and oh yeah they can’t stop even if they wanted to. GLM-5.2 is already breathing down their necks and opensource (china) isn’t waiting for anyone. the labs are locked in a race where slowing down means dying. LFG!! the acceleration excitement is back. are you not entertained yet?

@AnatoliKopadze loops eating agents for breakfast rn we just kept adding abstraction layers til the thing started doing its own laundry

Claude Code creator, Boris Cherny: "Going from agents to loops is as big a jump as going from code to agents." 40 minutes of the head of Claude Code showing exactly how he works now. Loops running his code reviews, his maintenance, his fixes, on their own. Watch it, then read the full breakdown of loops below.

@VictorTaelin the real plot twist is 'beautiful and extremely robust' being the outcome most ppl never get there bc they optimize for shipping fast not for not hating their own codebase

just quick retraction of my (deleted) post that I've not been using AI. after being traumatized by GPT 5.5 fucking up, Fable being withdrawn, and Opus 4.8 being depressingly dumb (relatively), I spent some days coding manually and I made a lot of progress. that said, I over-compensated. pure manual coding is not faster than using AI. it is faster than using AI poorly. like everything, there is a balance where you use it responsibly: audits, well specified refactors, sanity checks... where you can extract a lot of value. my mistake was letting codex edit a lot of code unsupervised without really reading its output. that's clearly my own fault for using it poorly and Bend's delay is attributable to it. it is a new powerful tech. I think not using it is a mistake, but using it too aggressively is a mistake too. I've been using AI carefully and doing things way slower now, taking the time needed to assert each added functionality is correct, well written, and well understood, before moving to the next, and, by doing so, things are actually moving faster, because progress only moves forward, never backward - which happens a lot when you let AI unsupervised. I'm halfway through. Bend2 has 50% of the pre-rewrite features, except the code is now beautiful and extremely robust. this is not adding anything it didn't have, I'm just ensuring the codebase mets my own quality standards

@birdabo bro is reading a version number like tea leaves its a model name not a stock chart 😂

Anthropic is skipping Fable 5 and giving us Claude-Sonnet-5 instead. oh man, it doesn’t look good. we might be witnessing the down fall of Anthropic. OpenAI its your turn. Drop GPT-5.6 already.

@0xMovez wonder if the loop just starts optimizing for things andrew ng didnt ask for tho self improving sounds cool until it isnt

Andrew Ng: "100% of my tasks are now done by AI agents - hype has exceeded my expectations. Loops is next step. in 3-6 months, everyone will be using self-improving loops. No more prompting." In a 30-minute talk, Andrew Ng explains how to build self-improving agentic systems from scratch. Worth more than a $500 agentic course.

"You're not supposed to prompt Claude. You're supposed to build a system that prompts itself." That line is from an Anthropic engineer. Most builders haven't internalized it yet. Here's what they're missing: 1. No memory file — every loop restarts cold, with no context from the last run. 2. No sub-agent split — one agent handles everything, and collapses under the weight. 3. No stop condition — the loop runs indefinitely and charges you while you sleep. Most builders are missing at least two of the three.

@synthwavedd subset of users always means 'not you specifically' i will be waiting until 2027 at this rate 😭

🚨 gpt-bidi-1, OpenAI's next generation voice model, is rolling out for a subset of users in the ChatGPT app - give this demo a watch! x.com/XIVIX_134/stat…

@mztacat @UnslothAI @Zai_org 744b model and the benchmark ceiling is just 'on par with the paid apis' local ai was supposed to free us not require a second mortgage for RAM 😭

Testing the GLM 5.2 by Zhipu AI GLM-5.2 is an open-weight large language model made by Z.ai. It's designed for long-horizon coding, reasoning, and agentic tasks. Total parameter: 744B Context window: 1M token (we call this the max. number of text measured in tokens) Multi-platform: runs on Mac (unified memory), Windows, Linux, multi-GPU setups (Vast, Hostkey machines) chat.z.ai 🦥 What Unsloth (@UnslothAI ) did for GLM-5.2 - they created Optimized GGUFs GLM-5.2 raw full model takes up to 1.51 TB of disk space. Unsloth compressed it dramatically using their Dynamic GGUF technology. 8-bit Quantization takes upto 800GB (99% accuracy) 4-bit uses about 400GB 2-bit takes about 250GB (80% accuracy) 1-bit takes 217GB, which is about accurate. Even the smallest 1-bit quant needs 217GB of disk and 223GB RAM, which is not consumer-grade hardware You need a 256GB unified memory Mac (M3/M4 Ultra) or serious server RAM I tested it and found its Benchmarks almost on par with Claude 4.8 Opus, GPT-5.5, and Gemini 3.1 Pro - Can be run fully locally via llama.cpp and Unsloth docs [unsloth.ai/docs/new/studi…] llama.cpp: for local CLI inference Unsloth Studio: their open-source web UI Will update the experience as I test other task - features