recovering data scientist building ai that actually works severely allergic to "this will change everything" posts
Canada
Joined January 2026
- Tweets352
- Following96
- Followers368
- Likes187
If someone tells you that video editing is fully solved by AI, either they're editing very simplistic videos with obvious cut-off points, or they're lying
Unbelievably helpful, but I'm here on my seventh iteration with Claude for a video I've been working on for a few days, and I've had to spoon-feed it.
Very basic errors that a 10-year-old would be able to hear and say that there's a problem
It is absolutely not solved.
It has just increasingly become easier.
Matrix multiplication is not and will never be ‘conscious’
It will always be an illusion of intelligence.
Pure publicity stunts to prove otherwise.
I've seen 18 videos with these exact thumbnails and similar titles in the past 24 hours...
As soon as one video hits, you have tons of creators copycat it within hours with zero change whatsoever
I think this is an ominous warning of what's to come, especially with AI content creation
Any good idea will be knocked off, recreated and shipped within hours
Your next cold email will get read by an AI first
A year from now, practically everyone will have their own version of an assistant gatekeeping their inbox (whether it's a DIY agent or something like Grokbot)
You're going to have more senders, fewer humans reading, and an agent deciding what gets through
Cold 'anything' is about to get exponentially harder than it already is
"Claude 2.0 just came out" keeps coming out
I get it, that's the content game (and people still click every time)
But you literally can just set one scheduled task that tags the 'claude code guide agent' every week to walk through every new feature
Then it scans your own conversations and shortlists the ones you'd actually use
And if you really want a video, it builds you a personalized one with hyperframes (even spoken in your voice) or claude motion
The whole breakdown lands in under 3 minutes, versus 15 minutes of video for 1 minute of compressed sauce
I'd rather manufacture a better information diet than sit through fluff videos
🤖 Made with AI
Claude has gotten so restrictive that it refused to edit this video where I try to help you use uncensored models for good instead of evil
But you have to be careful how and where you use them...
I'll show you 3 ways to run uncensored AI, with the prompts to set it up
Even though GPT-6.1 Sol and Astra are very similar in terms of their core benchmarks, Astra is infinitely better at image generation and precision with GPT-2.5 Image.
Especially if you're into content creation, it's not even close.
Give the exact same brief to both models, and put them both on high. You'll see a night-and-day difference on average.
Anthropic just donated Claude Skills to the Linux Foundation.
Most skills are becoming unnecessary.
They were written for models that needed more hand-holding. Now those old, overlapping instructions can get in Claude’s way.
Here’s what’s still worth keeping ↓
Mark Kashef retweeted
Here are some of the people that I have learned from over the last few months/years.
This isn't everyone, but it's a great starting point if you are interested in AI, software engineering, or local hardware.
Later, I want to share more underrated people I missed. Please comment anyone you think I can learn from, and I will be sure to follow them! 🙂
Software Engineers (Super technical)
@webdevcody (Super technical in-the-weeds engineer)
@poteto (Unlimited AI usage but very cool skills)
@mattpocockuk (Very technical engineer and architect)
General AI knowledge and amazing tips
@MarkKashef (Amazing skills, tools and workflows)
@MatthewBerman (Very consumer AI tips but no BS)
@BetterStackHQ (More software than AI but some very good lessons)
Local Hardware and AI engines
@0xSero (Pushing local hardware)
@ashxhart (TensorFold)
@RaidOwlTweets (Homelab and networking)
Local AI
@bertholomusai (Very underrated local recipes)
@plotarmordev (Very cool niche models and config)
@MiaAI_lab (The Local AI Goddess)
One of the biggest perks of dictation apps is that they force you to enunciate.
When they mess up, it’s usually because I got verbally lazy, so they double as passive speech training.
They also made me realize I can optimize my “value per token spoken.”
If you ever saved an API key in opencode's console, do two things tonight.
Delete it at the provider (claude, openai, google) and make a fresh key for each app
Then set a monthly cap and a usage alert, so a stolen key tops out at $20 to $30
I'd rather lose $20 than wake up to $400.