RudiResearch (@_floormodel) header
RudiResearch (@_floormodel) profile picture

Focused on tools, creative workflows & internet experiments Built and shipped Rudi-Sim open source

New York, USA
Joined December 2017
Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)Photo by RudiResearch (@_floormodel)
I’m 32. Solo dev. Founder. Building my own thing. Looking to connect with more marketers, indie hackers, and developers. Let's grow together. 🤝🚀
I'm 22. Solo dev. Founder. Building my own thing. Looking to connect with more marketers, indie hackers, and developers. Let's grow together. 🤝🚀
35
1
33
842
From KOLs to Binance employees, no one is safe from social engineering
12
Do I know anyone mining pearlhash? #PRL
88
So how are we feeling about this name change?
We’re renaming our Max plans: Max 5x → Max Max 20x → Ultra Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do. We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until they’re verified, which helps us prevent fraud and abuse. You'll get an email when it's removed. Weekly limits remain unchanged. View our plans here: claude.com/pricing
24
Qwen3.8-Flash-Next runs on one RTX 5090. Kimi K3.1 Pro needs 80. Same "open weights" label, wildly different reality. Open weights sounds great but for Kimi K3.1 Pro that means over $600K of gaming GPUs. Qwen3.8-Flash-Next, for comparison, runs on one 5090 and 128 GB of RAM at about 48 tokens a second. It's 125B parameters, but only touches 6B per word, so the RAM can carry most of it. Kimi is 2.8 trillion parameters and touches 104B per word. Even squeezed down to 2-bit you'd need about 30 cards. The standard version needs 50 just to load. Someone actually did it: @totheagi ran 80 RTX 5090s across about 10 servers at 20 tokens a second, roughly $665K to build (maybe they can confirm that cost)
China’s latest model Kimi K3.1 Pro came out this week, and it's the first open-weight model I'd honestly put next to Claude Opus 5.5 and GPT-6 Astra. The quick version: - Kimi won 7 of the 14 tests I compared - Opus 5.5 still has the best independent overall score and wins on real job tasks. It's the most consistent of the three. - Astra leads on hard science and security testing, and it's the priciest at $10 in / $50 out per million tokens. (I will say that most of these scores come from Moonshot's own launch chart) Kimi's weights are due in November, so anyone will be able to download it and run it themselves which sounds enticing until you look at the hardware required to run it. It's a 2.8T parameter model. Compared to the 128B open weight front running Qwen3.8 flash next that can run on a 5090 and 64gb of ram, Kimi needs about 1.5 TB of GPU memory just to load. The whole model would require 2 ET900N machine at $125k each. Even a heavily compressed Kimi3.1 pro would require 30-80 5090s and substantial ram depending on your set up.. Unless I’m missing something, it seems the API is still the move and will be for a long time. Open weights make sense if your data has to stay on your own machines, you need to fine-tune the model, or you're running at huge scale. Will local frontier inference ever be financially viable for home set ups?
1
1
80
China’s latest model Kimi K3.1 Pro came out this week, and it's the first open-weight model I'd honestly put next to Claude Opus 5.5 and GPT-6 Astra. The quick version: - Kimi won 7 of the 14 tests I compared - Opus 5.5 still has the best independent overall score and wins on real job tasks. It's the most consistent of the three. - Astra leads on hard science and security testing, and it's the priciest at $10 in / $50 out per million tokens. (I will say that most of these scores come from Moonshot's own launch chart) Kimi's weights are due in November, so anyone will be able to download it and run it themselves which sounds enticing until you look at the hardware required to run it. It's a 2.8T parameter model. Compared to the 128B open weight front running Qwen3.8 flash next that can run on a 5090 and 64gb of ram, Kimi needs about 1.5 TB of GPU memory just to load. The whole model would require 2 ET900N machine at $125k each. Even a heavily compressed Kimi3.1 pro would require 30-80 5090s and substantial ram depending on your set up.. Unless I’m missing something, it seems the API is still the move and will be for a long time. Open weights make sense if your data has to stay on your own machines, you need to fine-tune the model, or you're running at huge scale. Will local frontier inference ever be financially viable for home set ups?
2
2
478
Google just dropped ARTEMIS An AI tool that lets your agent operate Android phones, test app workflows, and investigate failures. Naturally, I had a brief “wait, did Google just build my thing?” moment. While there’s obvious overlap with Rudi-Sim, Artemis goes deeper into Android automation (naturally). Giving an AI assistant a way to interact with what you’re building and check its work seems to be a growing theme around here. Everyone is jealous of Mac’s simulator. This space never ceases to amaze me and now Google of all things has given me a few more things to add to the list.
Rudi-Sim has 39 Apple devices built in, and for anything else you type a size. I check small laptops, odd tablets and half-width browser windows with one line in Claude Code: "Show my site at 390x844, 1280x800 and 768x1024."
1
50
Every time Claude walks through your site in Rudi-Sim, it saves a video with a dot for each tap and a caption for each step. That makes your next product demo for X or a client a by-product of testing.
1
2
60
This is the most serious 48 hours in cybersecurity history. "Black October" they're calling it It seems every *hour* there's a new incident, that in a typical year, would be one of the biggest incidents of the *entire year* SUMMARY OF THE LAST 48 HOURS: YEAR-LEVEL INCIDENTS (In a typical year, these would be among the biggest incidents of the year): 1. PHP2Shell: a hole in the software behind ~70% of websites. Governments took sites offline. 2. eMudhra: the master keys that prove websites are real were cracked. One was the number 4. 3. California: data on every public school student since 2023 stolen, including disability status 4. Railway: hackers stole customers' secret keys from a hosting platform used by millions of developers 5. Zeabur: 321,810 accounts breached. The company is shutting down. 6. LiteLLM: an AI tool downloaded 3.4 million times a day was poisoned with a password stealer 7. GitLab: a 10-out-of-10 severity hole in every version since 2014 8. Cairn: criminals used an AI hacking tool to break into 100+ companies 9. Ahold Delhaize: hackers knocked the owner of Food Lion and Stop & Shop offline 10. Onamae: malware planted on up to 26,739 customer servers in Japan MONTH-LEVEL INCIDENTS (In a typical month, these would be among the biggest incidents): 11. Signal: a bug could leave your mic and camera on after you hang up 12. NGINX: 3 critical holes in software running ~1/3 of websites 13. HTTP/3: a single connection can crash 7 of the 10 web server types tested 14. Plex: hackers breaking into home media servers 15. SafePay ransomware: FBI warns it has hit 600+ critical infrastructure organizations 16. Palo Alto: ransomware gangs breaking in through company firewalls 17. Bullhorn: platform used by 10,000+ companies knocked offline 18. Medi-Cal: 970,000 patients' data stolen 19. WordPress: hackers stole a plugin maker's keys and used them to plant backdoors on websites 20. AOMEI: backup software's update server handed out a backdoor 21. Meta Muse: Meta's AI agent exposed users' personal and business secrets 22. Tor: 333 relays hijacked to steal logins 23. TiDB Cloud: database company's cloud breached Also serious: 24. State Bank of India: India's biggest bank confirms a data breach 25. OVHcloud: hackers slipped a password stealer into a cloud company's official tool 26. Siemens: a software hole being used to spread malware 27. North Korea: state hackers launch a new attack campaign on South Korea
Guys I’m scared the takeoff is starting. I swear any one of the ~10 CVEs that dropped in the last 24 hours would’ve been “one of the worst in recent years” except a new one is dropping every hour
603
5,649
1,106
35,181
4,042,520
Railway is part of my stack. Wish me luck If you’re using railway for your backend immediately roll all environment variables, including any API keys stored in Railway, as well as any private registry credentials. Review logs for unusual activity. Check your billing on your OpenAI/claude or stripe. Ensure nothing looks out of the ordinary.
1
44
Claude Duet flips the script as Anthropic’s new app takes aim at Meta’s Muse and OpenAI’s Dot. Working in the cloud without local access to your computer by using a virtual machine with its own browser, Duet is geared to tackle what needs your attention most. Input plain text or voice to assign tasks and it keeps working behind the scenes. Setup is minimal, built on Claude Opus 5.5 it carries over the memory, skills, and connected apps from your existing Claude account. In practice, Duet can send a morning brief, flag what needs you in your inbox and calendar, and draft replies for you to approve. Who is it for? Anthropic says the aim is to serve professionals and small business owners with more on their plates than they can track. Right now it's a waitlisted research preview for Claude Pro and Max subscribers in the US, available as a standalone iOS and Android app or on the web at claude.ai. There's no extra fee: it shares your plan's usage limits, and its background checks count toward them. Anthropic is candid that it can still misjudge what's urgent, and it will keep tuning it based on early feedback. Duet enters a crowded field. It's Anthropic's answer to Meta's Muse agent, OpenAI's Dots agent XAI Grok bot, and Google’s Gemini Spark. Are you signing up?
We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.
1
87
I went to the @ASUS ProArt Creator Connect at @Pier59Studios in Chelsea, NY tonight. After a glorious sunset tonight and some drinks at the bar, it was time to get down to business. The event had an open bar inside and the food was catered, nice touch Asus. Pier59 Studios and @Nvidia kicked things off, with a look at @ComfyUI. Then @digitalix (Alex Ziskind) showed off the monster ET900N, which he uses to transcribe all his videos in seconds. He also ran a live demo with local AI models that added a cat holding a purse walking through the crowd. I got to talk with him afterward about where his work overlaps with Rudi Research. Next, Brett Danton showed how he uses ProArt monitors on set for consistent color. His team 3D-mapped an entire forest, rendered a Range Rover driving through it, and still uses the footage from the forest mapping. Finally, @sirwadefx talked about filming and editing his social content on ProArt gear, and using AI in DaVinci Resolve to cut his videos down before posting. All in all, I was really excited to get to check out some of the new Asus ProArt tech in person. The spark and et900n are amazing machines. I wish we'd had more time to talk with the presenters, but the space was closing and we had to head out. If you get the chance to go to a future event, I definitely recommend it.
1
1
251
My release checklist used to be "click around and hope." Now Rudi-Sim replays a saved tour of my site on iPhone, iPad and Mac after every update and marks each step OK or FAILED. Any walkthrough Claude does can be saved as a tour.
2
18
Windows will restart my PC at 3am because one pixel needs updating, then use this as their profile picture.
2
16
When you owe the bank $80k, you can’t sleep at night When you owe the bank $80B, they can’t sleep at night
JUST IN: Anthropic recorded a staggering $80,000,000,000.00 net loss in the first half of 2026.
2
16
AI offers a way to reclaim the time lost during a traditional job Once you glimpse the potential, it becomes difficult to resist This is why many new devs find themselves juggling multiple unfinished projects, each contributing to their journey towards financial freedom
2
13
Everyday is just one epic adventure to get back to my bed
2
18
MCPs are the connective tissue. Claude Code talks to Claude. MCPs talk to everything else... GitHub, databases, APIs, your entire tool stack. Skip them and you're just building in a sandbox.
2
15
Trending: AI cold emails are bombing. Why? Over-personalization sounds fake. AI learned to write emails that follow every best practice and ignored the one that matters: sounding human. The irony: AI is better at rough work than polish.
8