Hands-on technical leadership in AI. Building FALKON (https://xitter.gq/t.co/bSrrBOUdZz) We Should Do Something With AI newsletter: https://xitter.gq/t.co/4n4TT4HAzR
Antwerpen
Joined April 2010
- Tweets9.3K
- Following1.2K
- Followers1.5K
- Likes14.7K
Most AI workloads are batch jobs. They don't need to give real-time answers. It's OK to take a few minutes / hours.
A lot of cost savings and value get unlocked if we start treating a GPU like a mainframe instead of a web server.
There are a lot of images of $25K Mac Studios going around today. It's an amazing machine, but the bar for self-hosting is way lower than that.
A humble Strix Halo or a gaming computer running @coldniko 's Strata is enough to run local agents.
Whenever I use dictation on Mac, I'm struck by how bad it is. The default setting on Omarchy is 10 times better. That's on a laptop half the cost.
Mike Veerman retweeted
People who don’t use AI in their work are like the new vegans.
How will you know they don’t use it?
Ohh. They’ll tell you.
It’s basically their whole identity 😂
Mike Veerman retweeted
Before ai, it was a running joke that every developer hated every codebase they didn't write themselves.
Now every anti-ai hater wants to pretend humans were writing beautiful code.
Too many of us use the human brain as the yardstick to measure intelligence. They find it outlandish to imagine a machine that is smarter than us.
Just like heliocentrism was considered an offensive idea back in Galileo's time, a machine that beats us in math, coding, or even creativity evokes strong negative emotions.
Most software engineers still underestimate the complexity of the systems we are about to develop in the coming months.
"I should be able to understand the code"
Nope, that's not the case anymore.
The code is no longer there for you. It's there for a machine that has no trouble understanding it.
Rotate your nouls, people!
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/810…
I had quite a few pet projects running. Software that I once built and maybe wanted to revisit later.
In light of the current security issues, I have taken those servers offline. Whatever doesn't get maintained around the clock becomes an attack surface. Hobby projects don't justify the attention they require to keep their security up to date.
It feels like the end of a more innocent era.
This week’s Black October event has put the final nail in the coffin of the AI-slop crowd.
A lot of us, including me, predicted the big security apocalypse would come in the form of vibe-coded applications getting hacked. Non-techies would ignore Lovable’s security warnings and publish an app that would leak its credentials.
Reality is quite different. The vulnerabilities and exploits that came to light in the last few days are found in old human-generated slop. Hand-crafted slop. Artisanal, professional, senior Clean Code look-Ma-I’m-doing-DDD slop.
This week's We Should Do Something With AI is about how security issues will bring us the dreaded software factory.
somethingwithai.substack.com…
Deployments on Railway blocked while the data is compromised.
GitHub failing under the agentic workload, causing deployment pipelines to break.
DevOps through CI/CD is a struggling paradigm.
We are leaving a naive era in cybersecurity where protection for most of us was mainly a checklist, best practises and keeping dependencies somewhat up to date.
The enemy has fully automatic weapons now.
Mike Veerman retweeted
Amazing how quickly we went from "oh no LLMs are going to flood the Internet with insecure code" to "oh no LLMs have discovered the internet was already flooded with insecure code"