🤖 https://teapawt.net/t.co/O98jouTJ7y 🚀 https://teapawt.net/t.co/fwc01hzADu 🍌 https://teapawt.net/t.co/Bjc5gxOwhK 🎙️ https://teapawt.net/t.co/93gKv6LUlJ 🧠 7 Figure Agency Founder, Marketer, Brand & Positioning Nerd, AI Developer
United States
Joined September 2024
- Tweets5.4K
- Following618
- Followers910
- Likes4.9K
Pinned Tweet
railway paused builds at 3:24am and said it was an outage for 14 hours.
by the time this notice went up, people's keys had already been used and you couldn't even redeploy to rotate them.
full breakdown below 👇
An urgent notice from the Railway team
blog.railway.com/p/security-…
When "other countries" stop trying to use open weight models to destroy the world, and stop trying to distill US intelligence to cripple us - then we can open up the bench mark.
Until then, pound sand.
Fuck you, @ArtificialAnlys
Your new index says Claude Haiku 5.5 is smarter than DeepSeek V4.1 Flash 1017 and GLM-5.3.
Haiku is Anthropic's small cheap model. DeepSeek 1017 is the model Cursor put on its cost frontier this month, and Cursor says no model scores higher at those prices. I run it every day on my 4x 3090. I dont believe for a second that Haiku beats it.
Look at how they got there. 60% of the index now runs on private tests nobody outside can see. GDPval-AA, SciCode and the long context test are gone. In comes a private coding benchmark from Cognition. And the open weights gap goes from 3 points to 12 in 1 release, 4x. Released open models are now 26 points behind the top.
1 day before v5, on the old version, they gave Ling 3.1 Flash 1019 a 44. Now the best released open model gets 32.
The weights didnt change. The tests did, and you cant see the tests.
This is a bought benchmark. A leaderboard where you cant check a single question, that widens the gap 4x overnight and puts Haiku above DeepSeek, is marketing for closed labs. Its not measurement. I hope at least you were paid enough money to publish such garbage. Liars.
Publish the questions or stop calling it an index.
If 👏 I 👏 Catch 👏 You 👏 Ripping 👏 Off 👏 Ideas, 👏 Strategies, 👏 Or 👏 Being 👏 A 👏 POS 👏 I'm 👏 Calling 👏 You 👏 Out 👏 And 👏 Clapping 👏 Them 👏 Cheeks.
Try👏Me👏
"its not about taking business"
shares a literal 1 to 1 copy of the hero from postbridge.
get outta here you unoriginal, thieving piece of shit kebab. go do something original with your time.
absolutely worthless dog shit.
Paying $99/month was getting a little too expensive...
So I built my own open source alternative that now costs me LESS THAN $1/month to run.
YES, 99x CHEAPER! 🤯
Introducing Open Post, an open source alternative to Post Bridge that lets you post to all your social media accounts from one dashboard.
I'll be open sourcing it on GitHub this coming week once I finish testing each social media integration.
And just to be clear...
This isn't about taking business away from an existing SaaS. I simply wanted to build something cheaper for myself that does the exact same thing.
If you want a ready-to-use solution and don't mind paying $40-$100/month, definitely check out the original SaaS!
But if you prefer self-hosting and saving money, Open Post is coming soon for free. 🚀
Gabe Fletcher retweeted
Five drafts of a homepage tell you what you think of your homepage. They can't tell you whether a stranger gets the product.
If the core flow works, send the link to one person who has the problem and watch what they do first.
Get on the list: vibecodelist.com
ALT A pile of five paper homepage drafts on a dark concrete wall, newest on top. Each older draft shows only its top edge and a blue pencil note: headline too long, try a darker blue, move the button left, move it right. The top draft is a sketched homepage whose button is circled in blue with the note: back to the left? Beside the pile, a fresh sheet reads THE HOMEPAGE CAN WAIT. Below: If the main thing works, a sixth draft won't teach you much. Watching one person try the product will. A blue triangle points at the next step: send the working link to someone who has the problem, watch what they do first, and say nothing. Get on the list: vibecodelist.com.
VCF - Verified Compute Fabric - SecSpec: NextGen Security Infrastructure To Stop Agent Swarms
producthunt.com/products/vcf… via @producthunt
Why are we out here, clout chasing, trying to shit on Marc?
Kebab 1 makes his a $29 LTD.
Kebab 2 just dupes it and makes it OSS.
For what? Just to waste your time and go back and fourth with one another?
These idiots out here forget that homie Yang already did a better Datafast, called Rybbit Analytics, which has been OSS for idk how long.
I swear to god you guys are absolute UNITS.
🚨New Infrastructure Security Spec🚨
This week has been an AI security nightmare. So I went down the rabbit hole.
I’ve spent a ridiculous amount of time on one question: what does secure infrastructure look like when AI is in everything?
The short answer is that we already have most of the parts. They exist in some form today. Nobody has put them together, and no single company can do it alone.
We’re watching the biggest exploit sprint I’ve ever seen. Models are being stolen through distillation. Open source models from US adversaries are being turned into weapons faster than anyone expected. And I have no doubt we’re returning the favor in some way.
The old way of doing security was to trust the network, trust the vendor, and trust the box. That stops working once attackers have AI too.
So I wrote it down.
This is my first open source white paper and spec.
It’s called Verified Compute Fabric.
It is not finished. It’s a starting point.
It needs people a lot smarter than me to poke holes in it, argue with it, and build on it.
Security engineers, infrastructure people, model builders, policy folks. If that’s you, I want your eyes on it.
Read it here: vcf.md
If you know one person who should see this, send it to them.
Ok now THIS would be wild. And also very much worth of NON COMPETE
🚨 JUST IN: SpaceXAI may have poached two KEY engineers from Anthropic's Claude Code team.
And these aren't random departures.
→ Thariq Shihipar helped scale Claude Code for over a year.
→ Anthony Morris built the Claude Code desktop app.
Both have left Anthropic, but the alleged poaching still remains unconfirmed.
If its true, SpaceXAI isn't just chasing better models.
It could be recruiting the people who know how to turn frontier AI into products developers actually love using.
And that's where this gets interesting.
You can spend billions training a frontier model.
But what happens when your competitors start recruiting the engineers who built the product around it?
Are we about to see AI labs fight just as hard over talent as they do over benchmarks?