Oliver Crest (@OlivercrestAI) header
Oliver Crest (@OlivercrestAI) profile picture

Your unfair advantage in AI | tool reviews • workflows • launches | partnered with leading AI companies | 📩 [email protected]

Scotland, United Kingdom
Joined February 2020
Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)Photo by Oliver Crest (@OlivercrestAI)
🚨 Claude can now do the Excel work most teams hire analysts for. Analyze spreadsheets. Generate formulas. Build dashboards. Find hidden insights. Create executive reports. All from a few prompts. Here are 8 prompts that turn Claude into your personal Excel expert 👇🧵
14
40
3
125
12,255
Phone calls are usually where important details get lost. A client mentions their budget, agrees to next steps, or raises an objection, and you have to piece it all together afterward. Bluedot is bringing its meeting workflow to phone calls. Call anyone from the app, and the recording, transcript, and AI notes are ready when you hang up. Less manual note-taking, and a proper record of every important conversation. The demo shows it well.
New in Bluedot: phone call recording. Call anyone from the app. Get the notes when you hang up.
1
7
99
21,109
Oliver Crest retweeted
Someone on Reddit built his wife a job search app. She is a junior software engineer, and her job search had dragged on for almost 10 months. Tailoring a resume for every posting took ages. So he built Dossier. It finds postings, then writes a tailored resume, cover letter and email for each job, using only facts from her profile. It tracks every application from draft to offer on a board, and it runs on your own server, with a local AI model option, so your data stays with you. She landed a job with it, so he open-sourced it for anyone in the same spot. It is free at github.com/AienTech/dossier
4
10
16
1,910
Your email has a search bar. Your photos have a search bar. Your house should have one too. For the adapter you bought twice. The drill you know is somewhere. The receipt you saved for exactly this moment. It is called Homebox. A free, open-source inventory system for the things you already own. You add an item. Give it a name, a photo, and a location. Next time, search your inventory before opening another shopping tab. And it goes beyond finding things. Think about your washing machine. The receipt is in a drawer. The warranty is in an email. The serial number is on the back of the machine. One appliance. Three places to look. Homebox lets you keep purchase dates, prices, serial numbers, and warranty records alongside your items. You can also share collections with your family. So the inventory does not have to live in one person's head. You do not have to catalog your entire life. The tools on one shelf. The cables in one drawer. The box you keep opening because you cannot remember what is inside. One honest note. You have to add the items and keep their records updated. Homebox is self-hosted, so setup, hosting, and backups are your responsibility. Free software. Some assembly required. Start with one drawer. Before you buy another one, find the one you already own. (Link in the comments)
12
11
81
1,379
10 free websites that feel like cheat codes for learning anything. 1. Harvard CS50 The most famous course on the internet. The best intro to computer science anywhere. Taught by David Malan. Site → cs50.harvard.edu 2. Open Yale Courses Recorded classroom lectures from 40 full Yale courses. Strongest in philosophy, history, and humanities. Site → oyc.yale.edu 3. The Odin Project The free full-stack curriculum used by career-changers landing $100,000 jobs without a CS degree. Site → theodinproject.com 4. fast.ai Jeremy Howard's free deep learning course. Every self-taught ML engineer credits it with getting them into the field. Site → fast.ai 5. Anthropic Courses Free interactive courses on prompt engineering and building with Claude. Taught by the team that built the model. Site → github.com/anthropics/cou… 6. Brilliant Interactive lessons in math, science, and CS. Learn calculus by playing with it, not memorizing it. Site → brilliant.org 7. Language Transfer Complete audio courses, no ads, fully free. Zero to intermediate Spanish in 90 lessons. Site → languagetransfer.org 8. Khan Academy Free K-12 through college. 138 million learners across 190 countries. Sal Khan explains any concept until it clicks. Site → khanacademy.org 9. Marginalia Search An indie search engine that skips SEO spam. No ads. The search Google used to be. Site → marginalia-search.com 10. Wikenigma A wiki of everything humanity does not know. 1,280+ open problems across every scientific field. Site → wikenigma.org.uk
8
24
80
1,247
Oliver Crest retweeted
unbelievable. 🤯 Emad Mostaque just showed Taalas generating at ~14,000 tokens/second speed. For context, ChatGPT generates anywhere between 50 to 150 tokens/second. ---- From "The Peter McCormack Show" YouTube channel, (full link in comment)
4
9
1
16
3,989
Oliver Crest retweeted
Reinforcement learning apparently kept this running posture over human form A person swings the arms mainly to stop the upper body from twisting with every stride, the twist that each leg puts into the torso. A robot arm is motors, gears, and wires. fast swinging will generate high torque, over and over, for a full lap. The forward lean puts the mass where speed wants it. It shifts the center of mass ahead of the feet. For a biped, that can mean less time fighting to stay upright and more of each step going into forward motion.
7
9
18
2,027
A US senator walked up to an airport checkpoint and refused to have his face scanned. His name is Jeff Merkley. He is the senator from Oregon. He has spent three years trying to get the TSA to tell travelers something the TSA already admits on its own website. The face scan at the checkpoint is optional. Almost nobody is told. There is one sentence that stops it. Here is the sentence, and everything the TSA does not want you to know about saying it. First, the TSA's own words. From the TSA website: "Travelers who do not wish to participate in the facial recognition technology process may decline the optional photo, without recourse, in favor of an alternative identity verification process, which does not use facial recognition technology to verify their identity." Optional photo. Without recourse. Those are their words, not mine. The TSA also says this on the same page: "Travelers will not lose their place in line for security screening." And using the alternative method "does not take longer." So the two things a TSA agent might tell you at the checkpoint, that you will hold up the line or take longer, are ruled out by the TSA's own website. Now the sentence. When you get to the checkpoint, before the scan, say this: "I do not consent to the scan and would like to have my ID manually verified." That is it. The officer will take your physical ID, either a driver's license or a passport, and check the photo against your face the way they did for the last twenty years. Merkley has documented himself doing this on flights between DC and Portland. Here is why almost no one knows. Senator Merkley has been pushing this since July 2023. He went to Reagan Washington National Airport that summer to publicly demonstrate the opt-out and to challenge the TSA on how the program is being run. The Washington Post covered it. In May 2025, Merkley introduced the Traveler Privacy Protection Act with Senators John Kennedy of Louisiana, Ed Markey of Massachusetts, and Roger Marshall of Kansas. Two Republicans, two Democrats. The bill is endorsed by the American Civil Liberties Union, the Electronic Privacy Information Center, Public Citizen, and three other watchdog groups. Merkley's own words from that press release: "Folks don't want a national surveillance state, but that's exactly what the TSA's unchecked expansion of facial recognition technology is leading us to." And: "Americans have the right to opt out of using TSA's facial recognition at the airport, and we need to protect that right. Our Traveler Privacy Protection Act safeguards the freedoms and privacy of all Americans by making sure no one is required to have their face scanned to travel." Here is what the TSA is planning. Facial recognition is currently at more than 84 US airports. The TSA plans to expand it to more than 430. That is nearly every commercial airport in the country. And per the bill's findings, the TSA is "eventually planning to make this technology mandatory for all travelers." Right now it is optional. That window is closing. The TSA says the technology is not being abused. Their exact wording: "Biometrics are not used for surveillance. Facial recognition technology is solely used to automate the current manual ID credential checking process and will not be used for surveillance or any law enforcement purpose." Take them at their word. The point is not that the TSA is doing something illegal. The point is you get to choose. Here is the 5-minute action you can take tonight, before your next flight. - Memorize the sentence: "I do not consent to the scan and would like to have my ID manually verified." - Have your physical ID out and ready. Driver's license or passport. Not a screenshot on your phone. - When you get to the podium, hand the agent your ID and say the sentence in a normal tone. Do not apologize.
9
22
1
59
2,403
- If the agent pushes back or tells you it will take longer, calmly repeat the sentence. If they still refuse, ask for a supervisor. - This applies to TSA checkpoints only. International arrivals into the US go through Customs and Border Protection, which is a different agency with different rules. For most US domestic flights, the sentence above is what you need. Here is the last thing. The TSA already admits, on its own website, that this is optional. A sitting US senator has spent three years asking them to put that in plain sight at the checkpoint. They have not. The signs at the podium do not tell you. The agents often do not tell you. The airline app does not tell you. Congress is being lobbied by the airlines to keep it that way. Reuters reported in July 2025 that the airline industry publicly opposed Merkley's bill. Right now, before the 430 airport rollout, you still have a one-sentence way out. Say the sentence. Show your ID. Walk through.
183
3/thought this was only a sleep-deprived thing? read the small print. the same fluid coupling to your pupil showed up in rested people during drowsy moments too. weaker, but there. every heavy-eyed afternoon at your desk. every meeting where you nodded off for a second. every long drive where the road went blank. a wave was leaving your brain. you never had an mri catch it.
1
11
Take any song on Earth. Any artist. Any era. Upload it. Click one button. The AI splits it into individual tracks. Vocals in one file. Drums in another. Bass in another. Guitar in another. The Beatles. Kendrick. Taylor Swift. Billie Eilish. A cassette your father recorded in 1974. Every instrument separated. Clean. Individual. Playable. Studios could not do this cleanly ten years ago. If you did not have the original session files from the mixing engineer, the vocals and instruments were fused together forever. Then in 2020, two pseudonymous developers who go by Anjok07 and aufr33 uploaded a Windows app to GitHub. Free. MIT licensed. No ads. No account required. It is called Ultimate Vocal Remover. UVR5. 25,833 stars on GitHub. And here is the number nobody talks about. One release, v5.6 from September 2023, has been downloaded 5,161,858 times directly from GitHub. The full release history sits at over 6 million downloads. Six million people quietly downloaded one open source tool to do what the music industry sells as a premium AI feature. Here is what it separates. Two-stem mode. Vocals and instrumental. A karaoke track from any song in minutes. Four-stem mode. Vocals, drums, bass, and other. Every rhythm section on its own layer. Six-stem mode. Vocals, drums, bass, guitar, piano, and other. Three AI engines built in. MDX-Net for clean vocal isolation. Meta's Demucs v4 for multi-stem separation. VR Architecture for post-processing to strip reverb and echo from extracted vocals. Chain them together. Extract vocals with MDX-Net. Clean them with VR. Studio-grade acapella from a YouTube rip. Here is who uses this. DJs pulling acapellas for live sets without begging labels for stems. Producers sampling drum breaks from 1970s funk records. Music teachers isolating a bass line for students to practice against. Karaoke makers turning any song into an instrumental track. Podcasters removing background music from an interview. Filmmakers extracting dialogue from a scene with a soundtrack. Guitarists learning a solo by hearing it in complete isolation. Here is what the paid alternatives charge. LALAL.AI: $9.99 to $19.99 a month, with one-time top-up packs from $50 to $300. Moises Pro: $29.99 a month on the US App Store. iZotope RX 12, the post-production industry standard, sells its Advanced edition in the four-figure range. UVR5: $0. Unlimited songs. Unlimited stems. On your machine. Runs on Windows, macOS, and Linux. Works on CPU or GPU. NVIDIA CUDA and Apple Silicon both supported. One honest note. The last GitHub push was March 2025 and v5.6 is still the current release. In most repos that would be a red flag. Here it means the tool works so well that 5 million people downloaded one release and never needed a v5.7. The upstream models it wraps around, Demucs from Meta and MDX-Net, are the state of the art in audio source separation. Six million downloads. 25,833 stars. MIT licensed. Every song is made of layers. UVR5 peels them apart. (Link in the comments)
13
13
60
1,934
Joe Rogan uploads 2 to 3 hour episodes. You see 30 second clips. Someone cut them. Creators pay $29 a month for AI to do that job. A solo developer in China gave it away for free. The repo is called AutoClip. 6,391 stars on GitHub. 1,253 forks. MIT licensed. Every clip you save from a podcast, stream, or interview was cut from something longer. Streamers do not go viral from streams. They go viral from clips of streams. A 2 hour episode holds 15 to 25 viral clips. Finding them by hand takes 4 to 6 hours. Here is what AutoClip does. Paste a YouTube link. Or a Bilibili link. Or drop in a local video file. The AI reads the whole transcript, scores every moment for viral potential, cuts the highlights, adds captions, and hands you a folder of vertical shorts ready for TikTok, Reels, and YouTube Shorts. Here is what makes it different from OpusClip. - Runs locally on your laptop. Your footage never leaves your machine. - Works with OpenAI, Google Gemini, Alibaba Qwen, or SiliconFlow. Pick whichever LLM is cheapest for you. - No credit system. No monthly minutes cap. Process a 4 hour livestream if you want. - MIT licensed. Fork it. Ship it in a product. Sell it. No permission needed. - Docker one-line install. Runs in a browser tab at localhost:3000. The stack: FastAPI backend, React frontend, Celery task queue, Redis, SQLite, yt-dlp for downloads, FFmpeg for cutting. One developer named zhouxiaoka wrote 89 commits and shipped the whole thing. Here is the interesting part. OpusClip raised $68 million in venture funding from SoftBank, Samsung Next, and DCM Ventures to build a subscription business around this exact workflow. The whole short form video AI market is priced like a SaaS product because most creators do not know an open source alternative exists. One person in Hangzhou wrote it in his spare time and put it on GitHub. Six thousand people have starred it. One thousand two hundred have forked it. One honest note. Last commit was 77 days ago on June 3. The maintainer is one person. Some features on the roadmap are still marked in development. If you want a polished product with a support team, pay OpusClip. If you want to own the tool that clips your videos, download this tonight. The people who own the software own the workflow. Your videos. Your AI. Your shorts. (Link in the comments)
18
17
70
2,562
Every robot lawn mower on the market drives in random directions. It rolls until it hits a buried boundary wire. It turns. It rolls again. Like a Roomba that escaped and found your lawn. Husqvarna Automower 305: $1,199. Random. Husqvarna Automower 450X: $3,499. Random. Worx Landroid: $800 to $2,000. Random. In 2026, these machines drive randomly on purpose. And they want you to bury a perimeter wire around your whole yard. Now meet OpenMower. A free, source-available alternative to Husqvarna Automower. Created in January 2022 by a German engineer named Clemens Elflein. He bought the cheapest robotic mower he could find, ripped out the electronics, and replaced the brain with a Raspberry Pi and an RTK GPS module. 271 commits later, he is still leading the project. 6,669 stars. 406 forks. 2,000+ people in Discord. Here is what OpenMower gives you: You walk your lawn boundary once with the mower using an RC remote or your phone. The GPS records the path. That is your map. No wire. No trenching. Press start. The mower cuts in parallel rows. Every blade of grass. No random patches missed. - RTK GPS accuracy down to a few centimeters - Multi-zone mapping in a mobile and web app - Rain sensor pauses the mower, resumes when dry - Full CAD, PCB, and firmware in the repo - €700 in parts plus a used YardForce donor mower - GPL-3.0 software, GPL-2.0 firmware, CC BY-NC-SA hardware Here is the wildest part: In May 2026, security researcher Andreas Makris demonstrated to The Verge that he could remotely take over a $5,000 Yarbo robot lawn mower from 6,000 miles away. Same hardcoded root password on roughly 11,000 devices. He pulled owner emails, Wi-Fi credentials, home GPS coordinates, and camera feeds. The reporter got run over during the demo. Your lawn map never leaves your Raspberry Pi. OOMWOO freed your floor. OpenMower frees your yard. Free and Source Available. (Link in the comments)
19
22
74
3,702
Oliver Crest retweeted
OpenAI is falling apart right now. 9 of their most important leaders have left the company recently, and Altman is about to ask the public to buy the stock. 2 of them even walked out within 72 hours of OpenAI handing its own staff $7 billion in cash... On Monday, August 10, OpenAI completed a deal letting current and former employees sell roughly $7 billion worth of their shares. The price valued the company at $852 billion, the exact same number as its March funding round. On Tuesday, August 11, Brad Lightcap announced he was leaving after 8 years. He spent 4 of them as chief financial officer, then ran the company as chief operating officer from 2022 until April. He worked alongside Sam Altman at Y Combinator before OpenAI existed. On Thursday, August 13, chief revenue officer Denise Dresser announced she was leaving. She was hired in December from Salesforce, where she had been the CEO of Slack. In April she took over most of Lightcap's responsibilities. She lasted 8 months. The cash window opened Monday. By Thursday both executives who ran the business side were gone. But what's interesting is who actually wrote the $7 billion cheque: Every previous time OpenAI let its employees cash out, an outside investor bought the shares. In October, Thrive Capital, SoftBank and others put up $6.6 billion at a valuation near $500 billion. There was a $1.5 billion version of the same deal in 2024. This time OpenAI bought the shares back itself, using its OWN money. So no outside investor put a single dollar behind that $852 billion price. The company named its own number and then paid it. This is a business generating around $2 billion a month while losing roughly $1.22 for every single dollar it earns. And it just spent $7 billion of that cash buying its own stock at a number no third party ever tested. Here is the full list of the people who left since April: - Bill Peebles, who ran the Sora video app - Kevin Weil, vice president of OpenAI for Science - Srinivas Narayanan, technology chief of B2B applications - Kate Rouch, chief marketing officer - Josh Achiam, chief futurist, after nearly nine years - Johannes Heidecke, head of Safety Systems - Chloe Bakalar, the only person at OpenAI whose entire job was ethics - Brad Lightcap - Denise Dresser Bakalar left in July. OpenAI never announced it, and NOBODY has replaced her. Fidji Simo stepped down the same month, and two thirds of the organization had been reporting to her. Greg Brockman absorbed most of her job. He also introduced Dresser's replacement this week, a Wiz executive named Dali Rajic. OpenAI filed its IPO paperwork confidentially on June 8. The full prospectus, the one with audited financials in it, still has not appeared. So the order of operations is worth sitting with... File the paperwork in June. Buy your insiders out in August at a price you set yourself. Watch the people who built the commercial side leave that same week. Then show the public the books. Retail investors will see those numbers for the first time in a document written after every one of these people had already made their decision. Sam Altman told staff in June that he expects to go public within the next year. Reporting since then has pointed at 2027 instead, and a tender offer of this size is usually what a company does when the listing is not close. Here is what I think happens next: That prospectus lands with a revenue line big enough to carry the story, and the executive turnover gets buried in the risk factors where almost nobody reads. The people who priced OpenAI at $852 billion this month were the same people who took the money out of it; and the next set of buyers will not get that arrangement.
2
10
17
3,664
Knowledge is becoming portable. A new project called book-to-skill converts any technical book PDF into a Claude Code skill in minutes. The book becomes callable context that lives inside your agent. The old way: - Read the book - Highlight the good parts - Forget most of it - Google the same concepts a year later The new way: - Drop the PDF into the repo - Get a working Claude skill - Reference it while you code - Ship faster with the author sitting next to you Every book you own becomes reusable context for your agent. Your library stops being decoration and starts being infrastructure.
21
21
73
2,692