snippai + fluidvoice = seamless
a terminal screenshot tool enhanced with an open source local only voice-to-text model. capture -> speak -> straight into terminal
shoutout to @AndrewWarner for popping up on my timeline talking about FluidVoice
extremely unprofessional. if kimi wants to make it as a frontier lab, they need to act like one: perhaps silently route people to worse models, and maybe write a blog post about the collapse of humanity
China open-sourced a peanut-sized OCR that parses entire 100-page PDFs in one shot..
It's called Unlimited-OCR. Only 3B params. Runs locally.
Every other OCR tool chops your doc into pages and loses the thread. this one reads the whole thing in a single pass.
→ One-shot "long-horizon" parsing (32K context window)
→ Multilingual, out of the box
→ 93% on the standard parsing benchmark (+6 over baseline)
→ <0.11 error rate past 40 pages
→ Runs 100% locally on your own hardware
→ Works with Transformers, vLLM, SGLang, Docker, Ollama, llama.cpp
Traditional cloud OCR (Textract, Google Vision, Azure Doc Intelligence) costs $1.50–$15 per 1,000 pages.
This runs on your machine. For free. Forever.
Baidu built it explicitly to push DeepSeek-OCR one step further. Already at 1.9M downloads on Hugging Face and most people have no idea it exists yet.
100% open source.
Guys for the 10th time please setup your AI systems in a way that meets fundamental TCPA compliance.
23 business owners that hired AI agencies and they all got hit with $50k in TCPA lawsuits EACH.
Shoutout AI agencies that are scammers... you're somehow the best and worst.
Despite being a “successful” founder, I really am poor at several things.
These aren’t fake weaknesses that one would share in job interviews. They are things I genuinely stink at and wish I stunk less.
I’m disorganized. I’m impulsive. I procrastinate. I’m highly distractible. I have trouble sticking to a timeline.
So how have I had one 9-figure outcome & another hypergrowth startup, in spite of my deficiencies?
1. I embrace the cringe.
I make audacious asks to customers. I’ll look like an idiot on social media. I’ll push through the discomfort of asking innumerable questions and keep going until I actually understand something. I’ll make an ass out of myself if it means getting more exposure for my business.
2. I’m a one-trick pony.
And I don’t feel bad about it. There are a few things I’m world class at. Building organic distribution. Relationship building with execs. And telling a compelling story. I spend 90% of my time sharpening these tools while letting others around me cook in their zones of genius.
3. I’m like a kid.
If there’s one thing I’m good at it’s honoring my 5-year-old self. I’m insatiably curious. I probably ask 100+ questions/day to employees, customers, and people in my network. I’m goofy and playful. I can be serious, but most of the time I’m not, because most of what we do in knowledge work, just isn’t that serious. You can find me shooting nerf darts across the office or building a Lego set in between calls.
Why do I share all of this?
Because first time founder Alex wanted to be great at everything and enjoy every part of building a business. And if he didn’t, he thought something was wrong with him.
Nothing was wrong with him. He was exactly what he needed to be and where he needed to be. He just didn’t believe it yet.
i had to learn the same thing for tendonitis in the knees. in fact, tendonitis is a misnomer. after a while and with too much rest, the tendons end up in a state of disrepair and degenerate, and further rest doesn't help them heal. you have to bring in steady blood flow and it takes a long time for the tendons to start breaking down and rebuilding healthy fibers
@karpathy@btaylor also, a weird loop where constantly reading AI slop, or even if it's not slop, but AI responses, you start to pick up patterns of speech and use them yourself because they actually work. this then reinforces AI being trained on more recent text
bitcoin is much closer to the end of its bear market than the beginning. ETFs are puking their brains out on top of treasury company selling, who's been buying?
while long term holders now hold record supply, btc has traded into lower bound of on-chain valuation methodologies.
i'm perfectly happy to accept that i'm just a bad prompter, but I asked gpt 5.6 sol to render a cute, fluffy pixar style kitten, and rig it up for me with realistic movements using blender and this is the nightmare fuel it came up with