← Back to leaderboard
bigaiguy

Spencer Baggins

Verifiedactive
@bigaiguy

Love AI and Tech and Dawgpreneurship. Light focus on the DEEP STATE - Lets go to therapy together if possible.

RosterAI Agents & Productivity ToolsAI Model News & ReleasesTech Career & Coding Education
23,804 followers · 958 following · Last polled 2h ago
Perf. Score
37.8
Rank #312
Value Score
—
set rates to compute
Median views
898
p25 700 · trailing 7d
Followers
24K
-0.1% 7d
ER (impr.)
2.23%
eng ÷ impressions
ER (foll.)
0.09%
eng ÷ followers
Est. CPM
—
needs a rate
WoW views
-23.3%
▼ falling

Rates & value · est. from 898 median organic views

Quote tweet
—
no CPM
Post
—
no CPM
Thread
—
no CPM
Retweet
$75
no CPM
rate age unknown (pre-audit import)

Trends

Follower growth
Median views / day
Engagement rate
View distribution · trailing-7d posts

Each blue dot is one organic post. The Performance Score ranks by the median (898), not the mean (994) — so a single viral post doesn't inflate a creator's reach. The p25 floor (700) is the level 75% of posts clear.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
A French engineer who lives quietly in Paris has spent 30 years writing software that the entire internet now runs on without knowing his name. He wrote the code that streams every YouTube video, every Netflix show, every TikTok clip. He wrote the code that runs the virtual servers underneath AWS, Google Cloud, and Microsoft Azure. He calculated more digits of pi than anyone in history. He has no Twitter. He has no marketing. He just keeps shipping. His name is Fabrice Bellard. Here is the story, because almost nobody outside the systems programming world knows what one man has built. Fabrice was born in 1972 in Grenoble, France. He studied at École Polytechnique, the top French engineering school. He never went to Silicon Valley. He never built a startup empire. He just wrote code. In 2000 he started a project called FFmpeg, an open-source multimedia framework for encoding, decoding, and streaming video. He was 28. The project did one thing nobody else had done well. It handled every video and audio format that existed, in one library, on every operating system. He led it himself for years. Today FFmpeg is the invisible engine of the internet. YouTube uses it. Netflix uses it. VLC uses it. Chrome and Firefox use parts of it. Every Android phone, every iPhone, every smart TV, every video editing tool you have ever touched runs FFmpeg somewhere underneath. If you have watched a video on a screen in the last 20 years, Fabrice's code processed it. He was not done. In 2003 he started QEMU, a machine emulator and virtualizer. He wrote it solo until version 0.7.1 in 2005. QEMU lets you run any operating system on any other operating system. It became the foundation of modern virtualization. KVM, the Linux kernel hypervisor, runs on top of QEMU. Every major cloud provider, AWS, Google Cloud, Microsoft Azure, IBM Cloud, runs virtual machines on infrastructure built around it. The Quick Emulator is the most cited piece of cloud infrastructure code on Earth. He kept going. In 2001 he won the International Obfuscated C Code Contest with a small C compiler that grew into TCC, the Tiny C Compiler. TCC can compile and boot a Linux kernel from source in under 15 seconds. In 2004 he calculated the most digits of pi ever computed at the time, using a personal desktop computer and an algorithm he derived himself called Bellard's formula. In 2011 he wrote a complete PC emulator in pure JavaScript that runs Linux in your browser, a project called JSLinux that engineers still cannot believe is real. In 2019 he released QuickJS, a small but complete JavaScript engine that fits where V8 cannot. In 2021 he released NNCP, a neural network based lossless data compressor that immediately took the lead on the Large Text Compression Benchmark. Then he turned his attention to large language models. He built TextSynth Server, a web server with a REST API for running LLMs locally. He released ts_zip and ts_sms, compression utilities that use language models to compress text and short messages at ratios traditional algorithms cannot reach. He released TSAC, a very low bitrate audio compression system. In December 2025 he released Micro QuickJS, a new JavaScript engine for microcontrollers, separate from QuickJS, designed for environments with almost no memory. Fabrice co-founded a telecom company called Amarisoft in 2012, where he serves as CTO. Amarisoft builds 4G and 5G base station software used by carriers and labs around the world. He has been running it for over a decade while continuing to ship personal projects from his own home page at bellard dot org He has no Twitter. He has no Instagram. He gives almost no interviews. His personal website is a flat list of projects with no styling, no fonts, no marketing copy. Just titles and links. A quiet French engineer who never moved to Silicon Valley wrote the code that quietly runs the internet. He is still shipping.
frozen
22d ago
1.3M
7.3K
1.4K
113
160
3.7K
0.99%
A teenager in the United States started publishing software at 14 in 1998, built the entire online infrastructure for the Occupy Wall Street movement in 2011, joined Google as a software engineer, quit in 2018, and then spent five years writing a C library that does something the entire industry said was impossible. Then she combined it with llama.cpp and shipped the easiest way on the planet to run a large language model on any computer. Her name is Justine Tunney. Here is the story, because almost nobody outside the low level systems world knows what one engineer has built. Justine was born in 1984. She started writing and publishing software at 14, back when distribution meant uploading binaries to BBS systems and chat networks. She picked up the handle jart, which she still uses on GitHub today. She did the work most teenagers her age were not doing. She read the systems programming literature. She studied compilers. She fell in love with C. In July 2011 she registered the @occupywallst Twitter handle and the occupywallst dot org domain. Within weeks the protest movement that began in Zuccotti Park in New York had become a global phenomenon, and her infrastructure was the digital backbone of the entire thing. She handled the social media, the website, the donations, the coordination. She built the platform that pushed the movement to reach millions. After Occupy she joined Google as a software engineer. She worked on TensorBoard, the visualization tool for TensorFlow, and on site reliability for Google infrastructure. She stayed for years. Then in 2018 she left Google Brain to work on a personal project. The project was called Cosmopolitan Libc. Cosmopolitan does something most C programmers would tell you is mathematically impossible. It lets you compile a C program once and have the resulting binary run natively on Linux, Windows, macOS, FreeBSD, OpenBSD, and NetBSD with no modification. One file. Six operating systems. No virtual machines. No interpreters. No recompilation. The technique she invented is called Actually Portable Executable. The implications are wild. Cosmopolitan binaries violate every assumption about how operating systems load programs. They are at once a Windows PE file, a Linux ELF binary, a macOS Mach-O binary, and a shell script. The same bytes run on every platform. For five years she worked on it mostly alone. She funded the development partly through Mozilla's MIECO program, which sponsored her work on Cosmopolitan 3.0, released on October 31, 2023. A month later she shipped llamafile. llamafile is what happens when you combine Cosmopolitan with llama.cpp. You take any LLM weights file in the standard GGUF format, you wrap it in Justine's binary, and you get a single file that runs on six operating systems without installation. No Python. No CUDA setup. No dependency hell. Just one file that you double click and it works. Mozilla launched it as an official project of their innovation group on November 29, 2023. It went viral immediately. The repository, hosted at github .com/mozilla-ai/llamafile, now has 24,600 stars. The license is Apache 2.0. Justine kept shipping. She added GPU support to Cosmopolitan, a task systems engineers thought would require rewriting the whole thing. She added dlopen support, another thing nobody else had figured out. She wrote whisperfile, a single file version of OpenAI's Whisper speech-to-text model based on the same architecture. Her GitHub profile lists projects most engineers would consider impossible. sectorlisp, a Lisp interpreter that fits in a boot sector. blink, the tiniest x86-64-linux emulator on Earth. bestline, a teletypewriter command session library. redbean, a complete web server inside a single zip file. A teenager who shipped software in 1998 grew up to write the C library that the entire local AI movement now runs on top of. She did most of it alone, and most people scrolling AI Twitter cannot name her.frozen23d ago200K3.6K67186622.6K3.52%
A Dutch programmer released a text editor for the Amiga in 1991 and asked one thing from anyone who liked it. Donate to children in Uganda. The editor became one of the most used developer tools in history. It ships on nearly every Linux and Mac machine on Earth. Millions of engineers have it burned into their muscle memory. And for 32 years, he maintained it himself. His name was Bram Moolenaar. Here is the story, because every developer who has ever typed :wq owes something to this man. Bram was born in the Netherlands in 1961. In the late 1980s he bought an Amiga and wanted the vi editor he had used on Unix. There was a clone called Stevie, but it was missing features. So he started improving it. On November 2, 1991, he released Vim. Vi IMitation at first. Later, Vi IMproved. Then he made a strange choice for a software license. He called it charityware. Vim was free, but if you found it useful, he asked you to donate to ICCF Holland, a foundation he supported that helps children in Kibaale, Uganda. He had worked there as a volunteer himself. For decades, the help screen of the most popular programmer's editor in the world pointed people to an orphanage project in East Africa. Vim spread everywhere. It came preinstalled on Linux distributions and macOS. Every server you ever SSH'd into had it. Generations of programmers learned it, cursed it, and then refused to use anything else. It spawned forks, including Neovim, and inspired the modal editing keybindings found in VS Code, JetBrains IDEs, and almost every serious editor today. Bram kept maintaining it through everything. He worked at Google in Zurich for years and still released Vim updates. He wrote Vim9 script, a new scripting language for the editor, when most people would have handed off the project. Vim 9.0 shipped in 2022, 31 years after the first release. On August 3, 2023, Bram passed away at 62. The community took over. The next major release, Vim 9.1, was dedicated to him. The donation request is still in the help files. A programmer who wanted a better editor on his Amiga ended up shaping how millions of people write code. And the whole time, he kept asking them to help kids in Uganda.frozen21d ago9.5K13035102232.10%
Google researchers showed that AI can learn patterns from examples inside a prompt. It's called "Few-Shot Prompting" and it's ridiculously simple. Instead of explaining what you want in 500 words, you show the AI 3 examples of perfect outputs. It picks up the pattern. Here's how to use it:frozen14d ago9.2K441230170.83%
10 BOOKS THAT WILL MAKE PEOPLE LIKE, TRUST, AND LISTEN TO YOU Bookmark every one. 1. How to Win Friends and Influence People by Dale Carnegie Builds: the ability to make people genuinely like you. Written in 1936 and still outselling almost everything, because the basics never changed. Use someone's name, talk about what they care about, make them feel important, and doors open everywhere. The foundation every other book on this list quietly stands on. 2. Never Split the Difference by Chris Voss Builds: the ability to win any negotiation without a fight. Voss was the FBI's lead hostage negotiator, and he teaches you the exact moves he used when lives were on the line. Mirroring, labeling emotions, the calm voice that defuses anyone. Works on your boss, your landlord, and your kids. 3. Nonviolent Communication by Marshall Rosenberg Builds: the ability to disagree without it turning into war. Rosenberg's method separates what happened from how you feel from what you need, so a fight becomes a conversation. The framework couples, managers, and diplomats use to say hard things without blowing up the relationship. 4. Crucial Conversations by Kerry Patterson and team Builds: the ability to stay calm when the stakes are high. The talks that actually matter, the raise, the breakup, the confrontation, are the ones where people either freeze or explode. This teaches you to keep the conversation safe and honest at the exact moment it wants to go off the rails. 5. The Like Switch by Jack Schafer Builds: the ability to make people trust you fast. Written by an FBI behavior expert whose job was getting strangers to open up. The friendship signals, the small cues that lower someone's guard, the science of why we like who we like. Reads like a manual for instant rapport. 6. Influence by Robert Cialdini Builds: the ability to see persuasion coming, and use it yourself. Cialdini spent years undercover in sales jobs to find the six levers that make people say yes, reciprocity, scarcity, social proof, and the rest. Once you can name them, you stop getting played and start understanding every ad, pitch, and request. 7. How to Talk to Anyone by Leil Lowndes Builds: the ability to never feel stuck in a conversation again. Ninety-two small, specific tricks for small talk, first impressions, and working a room. The tactical opposite of vague advice, it hands you exact things to say and do. The cure for anyone who dreads walking into a party alone. 8. Difficult Conversations by the Harvard Negotiation Project Builds: the ability to raise anything without making it worse. From the team behind Harvard's negotiation research, it breaks down why hard talks go sideways and how to steer them. It teaches you to lead with curiosity instead of accusation, which changes how the other person hears every word. 9. Emotional Intelligence by Daniel Goleman Builds: the ability to read the room and yourself. Goleman made the case that managing emotions matters more than raw IQ for how your life actually goes. Recognizing what you feel, handling it, and sensing it in others is the invisible skill behind everyone who's great with people. This is where the idea started. 10. Give and Take by Adam Grant Builds: the ability to succeed by helping, not stepping on people. A Wharton psychologist shows that the most successful people are often generous ones, as long as they're generous the smart way. It reframes networking and career-building as something you win by making others better. Proof that being good with people and being good to them are the same skill. Nobody teaches you people. These do.frozen28d ago4.7K16037121337.05%
Claude quietly opened up 10 income streams that most creators have no idea exist. Each one does $1K to $8K a month. One weekend to set up. Zero inventory, zero ads, zero TikTok dances. Save this before everyone else finds out:frozen19d ago2.4K211321252.64%
🚨Udemy is hiding some of the best FREE courses online. Here are 10 free courses to upgrade your skills in 2025 👇frozen8d ago2.1K21921202.48%
10 GITHUB REPOS THAT REPLACE EXPENSIVE STARTUPS. All free. All open source. All install in one command. 1. n8n → replaces Zapier ($300/year) 2. Cal .com → replaces Calendly ($192/year) 3. Plausible → replaces Google Analytics 4. Penpot → replaces Figma ($180/year) 5. Mattermost → replaces Slack ($96/user/year) 6. Jitsi → replaces Zoom ($200/year) 7. Supabase → replaces Firebase 8. PostHog → replaces Mixpanel ($1,200/year) 9. AppFlowy → replaces Notion ($96/year) 10. Nextcloud → replaces Google Drive ($240/year) Every single one has 10K+ GitHub stars. You can self-host all of them on a $5/month server. Silicon Valley raised $50 billion to build software you can run yourself for the price of a coffee.frozen12d ago2.0K231411253.20%
If your authenticator codes sync through Google or iCloud, someone who gets that one cloud account gets every 2FA code you thought was offline. The codes were supposed to live on the phone. They live wherever that backup lives. You did not get a hardware key. You got a screenshot of a hardware key stored in the same account the key was meant to protect. Your authenticator isn't a second factor. It's a copy of the first one. Here's how to check what's syncing and move the crown jewels off the cloud in 15 minutes ↓frozen29d ago1.8K211110182.82%
Someone turned Alex Hormozi's $100M offer framework into an AI prompt system. It's like having Hormozi audit your business, rip apart your pricing, and rebuild your offer from scratch. Here are the 11 prompts (save this): https://t.co/5CqG0EhVhbfrozen3d ago1.8K171111182.71%
is Claude down for y'all? they gotta get their sh*t together https://t.co/5cDz2VY2Q6frozen23d ago1.8K624010.74%
You can turn Claude Code into a better deep research agent than ChatGPT Deep Research It's free. It's open source. And it's called Hyperresearch The problem with every deep research tool: you get one report, then everything it read gets thrown away. Next question, it starts from zero Hyperresearch keeps EVERYTHING Every source it reads lands in a searchable vault on your computer. Plain markdown files. The next time you research anything, it checks the vault first before searching the web Your research literally compounds. Every session starts smarter than the last Here's what happens when you give it one prompt: 1. Breaks your question into pieces 2. Reads 250+ sources in a single run 3. Writes 3 drafts in parallel 4. 4 AI critics attack the report looking for weak spots 5. A patcher fixes only what the critics found. It's locked so it CAN'T rewrite the whole report 6. Every citation gets checked before it ships. Fake quotes get blocked 7. Final report, fully sourced Some details that blew me away: - 5 reprints of the same press release count as ONE source, not five - Retracted papers get flagged and blocked - Paywalled paper? It finds a legal free copy and reads the full thing - Run crashes halfway? It resumes from the exact step it died on Works in Claude Code AND Codex. 3,600+ stars on GitHub The creator says it tops the DeepResearch-Bench leaderboard in internal testing. Third party validation is still pending, but the architecture speaks for itself Setup takes 2 commands: pip install hyperresearch hyperresearch install Then type /hyperresearch followed by any question Fair warning: a full run takes 1.5 to 2.5 hours. Quick questions take 30 to 40 minutes Do yourself a favor and run it on the biggest question you have right now. Go make a coffee. Come back to a report better than anything you'd pay for https://t.co/07MvPnO7rFfrozen24d ago1.7K251121133.11%
this might be the most underrated AI tool for creators... imagine uploading ONE podcast and getting multiple short-form videos without touching an editor. AI finds the hooks, cuts the clips, adds captions and reframes everything. 90 seconds later, your content is ready. step by step tutorial: 👇frozen13d ago1.6K20124062.55%
10 WEBSITES EVERY INTERNET USER SHOULD CHECK TONIGHT. Bookmark all of them. Most people don't know half of these exist. 1. https://t.co/zwR28T6wZa Shows every data breach your email is in and what got leaked. 2. https://t.co/3c9sMcOYLH Shows every social profile, photo, and login tied to an email address. 3. https://t.co/MOsvtupjHn Free disposable email for any signup you don't trust. 4. https://t.co/t6W6t9kzvQ Burner inbox that self-destructs in 10 minutes. 5. https://t.co/lHWq4ZeJXH A directory of direct links to delete your account from any major service. 6. https://t.co/vpVXkaS6Uc Check if your face was used to train AI image models without consent. 7. https://t.co/cC7q3S3Uui Tells you if your VPN is actually hiding your real location or leaking it. 8. https://t.co/1Q31VRhSQ6 Shows how trackable your browser fingerprint is, even in incognito mode. 9. https://t.co/TVtBWcv6Sw Tells you which programs on your PC are useless bloatware or spyware. 10. https://t.co/yvtYh3ade9 Drop any file or link. It scans against 70+ antivirus engines instantly. The internet is hostile by default. These websites are your free defense.frozen20d ago1.6K211120283.86%
Online Free learning is the future. AI can now teach machine learning like Stanford's CS229 professors (for free). Here are 15 insane Claude prompts that replace $120,000 ML bootcamps (Save for later) https://t.co/EyldF6IYBJfrozen10d ago1.6K251141113.32%
The most useful Claude Code skill I've seen this month has nothing to do with code quality. It fixes the way the agent talks to you. It's called i-have-adhd, and it has 50,000 stars on GitHub. Most coding agents bury the answer. You ask how to fix a failing auth test, and you get a paragraph about middleware, a detour into your dependency versions, and a "Let me know if you want to dig deeper!" The fix is in there somewhere. You have to dig for it every single time. This skill gives the agent 10 rules. A few of them: - Lead with the next action. - Number multi-step tasks. - End with one concrete next step. - Specific time estimates (minutes, not "a bit"). - No preamble. No recap. No closers. The before/after example in the README says it all. The "after" version is one command, three numbered steps, and what to paste back if a test fails. This matters more than it looks: When you run agents all day, reading their output becomes the bottleneck. Every sentence of filler costs you twice: once in tokens and once in your own focus. You can't fix this one prompt at a time. The rule has to live in a file the agent reads every session. It works with Claude Code, Codex, Cursor, Gemini CLI, and OpenCode. MIT license. Repo link below: https://t.co/RWLXIQpw7Ofrozen27d ago1.5K23118063.16%
OpenAI and Anthropic engineers leaked a prompting technique that separates beginners from experts. It's called "Socratic prompting" and it's insanely simple. Instead of telling the AI what to do, you ask it questions. My output quality: 6.2/10 → 9.1/10 Here's how it works:frozen9d ago1.5K11411142.09%
Researchers from Princeton and Google DeepMind developed a prompting technique that lets AI explore multiple solutions before choosing one. It's called "Tree of Thoughts." Instead of following one line of reasoning, AI explores different paths, evaluates them, and can backtrack. Think of it like playing chess with your own ideas. Here's how to use it: 👇frozen13d ago1.4K15101142.14%
This man changed the world, and you've probably used his invention 10 times this week without knowing his name. He invented the QR code. You scan it to pay for coffee, to open a restaurant menu, to get into a concert, to join Wi-Fi. He built it to track car parts in Toyota factories. He worked out its key design while playing Go on his lunch break. His company could have charged the whole world for the QR code. They made it free. His name is Masahiro Hara. Here is the story, because almost nobody outside Japanese manufacturing knows what one man designed. Masahiro was born in 1957 in Tokyo. He studied engineering at Hosei University. He joined Denso, a parts maker in the Toyota group, and spent his early career building barcode scanners. He never moved to Silicon Valley. He never started a company. He just worked on scanners. In 1992 the factories came to him with a problem. A barcode could hold only about 20 characters. One box of parts could carry up to 10 barcodes, each scanned separately. Workers were scanning around 1,000 barcodes a day. Barcodes couldn't store Japanese kanji at all. So Hara set out to build a new kind of code. His team had two people. The hard part was speed. A scanner had to find the code instantly, from any angle, even on a dirty, crumpled cardboard box. Hara decided to put markers in three corners so the scanner would always know where the code was and which way it was turned. The markers had to look like nothing else in the world. So the team took leaflets, magazines and cardboard boxes, reduced every printed pattern to black and white, and measured them. They were looking for the ratio of black to white that shows up least often in print. The answer was 1:1:3:1:1. That's why every QR code on Earth has those three nested squares in its corners. He was not done. The Go board solved the next problem. In Go you can read a position even when a stone is slightly out of place, because the stones around it tell you where it belongs. Hara used the same idea so the code could be read roughly first and then fine-tuned. A smudged or damaged QR code would still scan. After a year and a half, in 1994, they had it. The new code held about 7,000 digits instead of 20. It stored kanji. It scanned more than 10 times faster than other codes. Even with part of it torn off, it could still be read. They named it Quick Response. The QR code. Then came the decision that changed everything. Denso Wave owned the patent. They could have charged every company, every app and every printer on Earth. They didn't. They made the QR code free for anyone to use. The reason was simple. Barcodes were already free, and nobody would switch to something that cost money. He wanted everybody to be able to use it. It left the factory. In 2000 the QR code became an international ISO standard. In 2002 Japanese phones started shipping with QR readers built in. In the 2010s Alipay and WeChat Pay made it the way hundreds of millions of people in China pay for everything, from street food to rent. India built a national payment system that runs on QR codes at every tea stall. Then in 2020 the whole world used QR codes to check into restaurants, read menus and show vaccine records. A code built to track car parts became the way the world pays, orders and checks in. He has said he never imagined people would one day scan his codes on the street. He never became a household name. He never started a startup. His company never charged a patent fee. When people ask him why the QR code spread everywhere, he doesn't talk about himself. He says it's because it was free. This man changed the world with a code he built for a car parts factory. Most people will scan it today without knowing his name.frozen27d ago1.4K19135253.07%
holy crap… Pocket FM's memory system hits the same 86.7% accuracy as Claude Code's memory harness on Opus 5.x class models. for about 21x less per run. a storytelling company just built one of the most cost-efficient long-context memory setups I've seen.frozen22d ago1.4K300050.58%
I DON’T UNDERSTAND WHY PEOPLE DON’T USE GROK BOT FOR STOCKS. Most traders are looking at charts from 6 months ago. Grok BOT analyzes real-time sentiment on X to predict future. Here are 20 prompts to find the next 10x stock:frozen5d ago1.4K131021102.63%
10 BOOKS THAT MAKE YOU DANGEROUS AT BUILDING A STARTUP 1. The Mom Test by Rob Fitzpatrick Teaches you how to ask questions that can't be answered with a polite lie. Fitzpatrick shows why customer interviews fail and how to get real signal about your idea instead of compliments. You stop asking "would you buy this?" and start finding out what people already pay for. 2. Zero to One by Peter Thiel Gives you a filter for ideas that are actually new. Thiel argues the goal is building something that doesn't exist yet, not copying what works. After reading it, you can't look at a crowded market without asking what the secret is that nobody else has found. 3. The Lean Startup by Eric Ries Teaches you to treat every product as an experiment. Ries gives you the build, measure, learn loop and the idea of a minimum viable product. It's the fastest way to stop spending six months building something nobody asked for. 4. The Hard Thing About Hard Things by Ben Horowitz The book about what nobody tells you once the company is actually in trouble. Horowitz writes about layoffs, bad hires, and near-death moments from his own career with no motivational polish. You come away knowing there's no playbook for the worst days, only decisions. 5. Crossing the Chasm by Geoffrey Moore Shows you why early adopters and the mainstream are two different markets. Moore explains the gap where promising tech products stall and die, and how to pick one narrow beachhead to cross it. Essential if your product has fans but no growth. 6. Hooked by Nir Eyal Teaches you how habit-forming products are built. Eyal breaks down the trigger, action, reward, and investment loop behind apps people can't put down. Read it once and you'll see the loop in every product on your phone. 7. The Innovator's Dilemma by Clayton Christensen Explains why great companies fail by doing everything right. Christensen shows how well-run incumbents ignore small, cheap competitors until it's too late. It's the best lens for understanding why a scrappy newcomer can beat a giant. 8. Obviously Awesome by April Dunford Teaches positioning, the thing most founders get wrong. Dunford gives you a step-by-step method for deciding what your product is, who it's for, and why it's different. A good product with weak positioning looks mediocre, and this book fixes that. 9. High Output Management by Andy Grove The operating manual from the man who ran Intel. Grove treats management as a system with inputs, outputs, and leverage, not a vibe. It's short, dense, and still one of the best books on how to run a team. 10. Shoe Dog by Phil Knight Shows you what building a company actually feels like. The Nike founder tells the story of cash crunches, lawsuits, and constant near-collapse in the early years. You finish it knowing that most great companies were never smooth.frozen17d ago1.3K161131133.27%
You guys know in 100 years nobody will remember we existed right? Probably lessfrozen6d ago1.3K1234001.44%
You don't need the math anymore just call the API" is the most common advice given to new AI engineers yet the fastest growing AI curriculum on GitHub does the exact opposite. AI Engineering from Scratch, by rohitg00, just crossed 65k stars - 523 lessons across 20 phases, ~342 hours - Python, TypeScript, Rust and Julia - 114k readers, 182k page views in the last 30 days - Linear algebra at one end, autonomous swarms at the other - Free, MIT licensed The structure is the interesting part. Every algorithm gets built from raw math before any framework shows up. Backprop. Tokenizer. Attention. The agent loop. By the time PyTorch appears, you already know what it's doing under the hood. And every lesson ends with a reusable artifact instead of a certificate: a prompt, a skill, an agent, or an MCP server you drop into your own workflow. 523 lessons, 523 tools you built yourself. The README cites a stat that explains the demand: 84% of students already use AI tools, only 18% feel prepared to use them professionally. That gap is the whole market. The API-only engineer ships a RAG pipeline and can't explain why retrieval degrades past 50 documents. They wire a tool to an agent and can't say what attention is doing inside the model calling it. When it breaks, and it does, they're debugging a black box. The "just call the API" era rewarded speed. As models commoditize, the people who can actually open the box are the ones who get real leverage from them. 65k people seem to agree. https://t.co/3T3ugsPMBWfrozen7d ago1.2K22103053.29%
Someone built Git for AI agent memory. It's called OmniGraph, an open-source graph database that lets multiple AI agents share one knowledge graph, branch off it, review each other's changes, and merge, without overwriting each other's state. The problem it fixes shows up with long-running agents. When several agents work the same memory, one silently overwrites what another just wrote. There's no branch, no review, no record of who changed what. OmniGraph borrows the Git model instead. Each agent works on its own branch of a shared graph, adding entities, relationships, documents, and evidence without touching main. One agent ingests papers and extracts claims. Another resolves duplicate entities and links supporting evidence. Both run independently, and their changes get inspected, compared, and merged after review. The engineering goes further than the analogy. It does Git-style three-way merges, atomic commits, and historical queries. If two agents modify the same entity incompatibly, the engine reports a typed conflict instead of silently picking a winner. Retrieval is the other interesting part. One query engine combines: → Graph traversal for relationships → Vector similarity for meaning → BM25 full-text search → Reciprocal Rank Fusion to rank the results So an agent can retrieve by semantic similarity and explicit relationships in the same call. Under the hood it's built in Rust on the open Lance columnar format, with local files or S3-compatible storage. Schemas define valid node types and constraints. Cedar policies control mutations. Commit history records every change and the actor behind it. There's also an MCP server and a TypeScript SDK to wire it into coding agents and apps. Fully open source. GitHub: https://t.co/K06lPo604i1d ago1.2K18117063.52%
Stop paying $2,000 for AI bootcamps Someone put the entire AI engineering learning path into one free GitHub repo It's called awesome-ai-engineering and it's the closest thing to a free CS degree I've seen Here's what's inside: - Stanford CS229 and CS25 Transformers, the actual courses - Andrew Ng's Deep Learning Specialization, free to audit - fast AI Practical Deep Learning for Coders - Hugging Face Transformers course and fine-tuning docs - The best GitHub repos for LLMs and prompt engineering - Where to find and read research papers that matter - Playgrounds to test every major model side by side - Full interview prep, internships, and open source programs Here's how I'd use it if I was starting from zero today: 1. Start with Python. Don't skip it 2. Do Andrew Ng's ML course, audit mode, $0 3. Move to fast AI and build something real by week 3 4. Take the Hugging Face course and fine-tune your first model 5. Read one paper a week from the research section 6. Ship 3 projects and put them on GitHub The people getting hired in AI right now aren't the ones with certificates. They're the ones who built things Everything you need to start is free. The only cost is your time https://t.co/kwT2TKrPqWfrozen6d ago1.1K17102093.32%
Holy crap. OpenAI has 120+ YC founders working inside it. @thsottiaux says it feels like a giant startup. Unlimited tokens included. He also thinks Google is holding itself back. Most honest lab-vs-lab take I've heard all year...frozen21d ago1.1K210000.28%
If you have a startup idea, do this before telling anyone: Run it through a shadow advisory board of AI personas. Here's the exact prompt I use for Peter Thiel, Naval, Buffett, a YC partner, and a skeptical VC 👇10h ago1.1K12111162.95%
In early 2018 a former Kaggle champion showed that the way people trained language models was backwards. Within months, OpenAI and Google built their breakthrough models on the same idea. His name is Jeremy Howard. Here is the story, because he is a teacher who accidentally changed the research field. Jeremy grew up in Australia. After studying philosophy he worked as a management consultant, then co-founded Fastmail, the email service. He moved into data science and became the top-ranked competitor on Kaggle, then its president and chief scientist. He founded Enlitic to apply deep learning to medical imaging. In 2016 he and Rachel Thomas started Fasr AI. Its premise was that you do not need a PhD or a datacenter to do serious deep learning. Their free course, Practical Deep Learning for Coders, taught people to build real models from the first lesson. Thousands of students used it to change careers. He was not done. In January 2018 he and Sebastian Ruder published ULMFiT. The idea was simple. Pre-train a language model on a large body of general text, then fine-tune it on your specific task. In computer vision, transfer learning was standard. Nobody had made it work reliably for text. ULMFiT reduced error by 18 to 24 percent on most of the datasets it was tested on. With only 100 labeled examples it matched the performance of models trained from scratch on 100 times more data. OpenAI published GPT-1 in June 2018. Google released BERT in October. Both built on the same recipe: pre-train on a vast amount of text, then fine-tune. That recipe became the foundation of the entire field. In 2023 he co-founded Answer AI with Eric Ries. In 2024 the lab worked with Tim Dettmers to make it possible to train very large models on consumer GPUs. He also proposed llms.txt, a file that helps websites speak to language models. His courses are free. His library is open source. The most influential idea in modern language AI came from someone whose main job was teaching people how to use it.frozen17d ago9771893043.48%
This repo shrinks your AI skills by turning them into images. (100% open-source, and it's part of a 103k-star project) Every skill you install is prompt text. Your agent reloads that text on every single call, so a long skill costs you tokens every turn. Caveman's Pixel mode takes a different route. `caveman convert` renders the skill body into PNG pages in place, and the model reads those pages as an image instead of text. On the caveman skill itself: ↳ 1,069 estimated tokens down to 415 ↳ A 61% cut The safeguards are what make it usable: → It only converts when the image pages beat the text → Any failure leaves the skill byte-identical and names the gate that said no → `caveman convert --dry-run` shows the token math for every installed skill with no writes → `caveman convert --revert` restores the original byte for byte from SKILL.orig.md Run it per agent: caveman convert --agent claude Note that these are estimated tokens from the maintainer, so check your own numbers with the dry run first. The skill itself is MIT. The engine runtime is BSL-1.1 source-available. GitHub repo: https://t.co/KLj5l0H6kAfrozen15d ago9311290002.26%
This is the first AI company I’ve seen built solely because humans are dumbfrozen23d ago882100000.11%
Bad acting might finally be optional. An intensity slider that lets you keep the performance but fix the emotion is a real shift for anyone who has ever had to reshoot a take just for the face.frozen15d ago814510020.98%
Most models force you to choose between context and speed. This doesn’t. 384K tokens of text + images in one call. $0.05 / M input. Output free 100K tokens → decision in under a second😳 99.8 on Decision RULER at 256K (Jev 90.7).frozen10d ago795311000.63%
new AI benchmark: can my mom make something useful without knowing what a model isfrozen15d ago783411111.02%
Nothing better than a know-it-all that doesn't talk back1d ago746902021.74%
10 GITHUB REPOS THAT WILL CHANGE HOW YOU USE THE INTERNET. Bookmark every single one. Most people have no idea these exist. 1. LocalSend / https://t.co/kEzqe8Xzdi AirDrop for every device. Share files between Windows, Mac, Linux, Android, and iOS with no account and no cloud. 2. FreeTube / https://t.co/TcpPhy9Zw4 Desktop YouTube player with zero ads, zero tracking, and zero Google account required. 3. Vaultwarden / https://t.co/8ogZZxePmj Self-hosted Bitwarden server. Replaces 1Password, LastPass, and Dashlane for $0 forever. 4. Syncthing / https://t.co/0KSDbO9MdH Replaces Dropbox, iCloud, and OneDrive. Syncs files directly between your devices with end-to-end encryption. 5. Ladder / https://t.co/ZyFKBAPuzo Bypasses paywalls on NYT, WSJ, Bloomberg, Nature, and hundreds of other sites. Runs on your own server. 6. Immich / https://t.co/rr0bs3X7jj Self-hosted Google Photos replacement. Auto-backup, facial recognition, and AI search on your own machine. 7. AdGuard Home / https://t.co/MEXlYTarGM Network-wide ad blocker. Blocks ads and trackers on every device in your home including smart TVs. 8. Jellyfin / https://t.co/OasydvmVLL Self-hosted Netflix. Stream your own movies, shows, and music to any device with zero subscription. 9. LibreChat / https://t.co/I4TWzEOCJT Run ChatGPT, Claude, Gemini, and DeepSeek in one interface. Pay only for API usage. No $20/month subscription. 10. Uptime Kuma / https://t.co/TKK6bucHcF Beautiful self-hosted monitoring for every website, server, and service you own. Replaces $50/month tools. Big tech built the internet you use every day. These repos built the one you actually want.2d ago735301071.50%
“that’s not how our software works” has won way too many arguments about how a company should operate very ready for the software to be the thing that adapts for oncefrozen7d ago7051050022.41%
After watching him go a full day without a phone, it’s hard not to wonder what else we’ve been calling "essential" just because it’s in our pocket.frozen5d ago702200000.28%
we keep making agents smarter but they still forget what i told them yesterday. glad someone's finally fixing it https://t.co/C3iRdL2nSR2d ago695522011.44%
Sazabi staring at every bug in my code file https://t.co/30UyObZyBgfrozen9d ago671200020.60%
🚨 Kokoro just lost a blind listening test to every single Phonon model and nobody's talking about it. Independent listeners rated each one more natural. https://t.co/dQSQfl7b7H2d ago662503011.36%
I can’t tell if that’s a lot of money or notfrozen7d ago602202000.66%
AI isn’t just about building the next model. Sometimes it's just about creating a useful tool and getting it in front of the right people. CloneCut, an AI skill on Capafy generated $300K in sales in 30 days. These AI marketplaces are worth following.1d ago549212021.28%