adlrocha

So it goes. > zerohedge: *SPACEX SHARES INDICATED TO OPEN 29% ABOVE IPO PRICE IN DEBUT

Post written for this week's newsletter.

i am the only one that hates people that do not use capital letter when they write. even after full stops. i hate this more than typos and bad grammar. because I do the latter a lot. it would be hypocritical. but is use capital letters. so there you go!

I can confirm that is pretty good!

Time to test this guy to see what all this fuss is about

While everyone is praising Fable I am falling in love again with Kimi K2.6

/goal mode is the incarnation of Karpathy's autoresearch.

Getting this kind of feedback is the extra motivation that one needs to consistently write a new post every week.

I think setting Claude Sonnet's default effort as "Low" is a product mistake.

It ended up longer than I planned, but I’m pretty happy with how this one turned out.

An adoption crisis and the end of an AI cycle are aligning...

I agree.

It ended up becoming longer than originally planed, but I am genuinely proud of this week's post :) I had a lot of fun writing it.

This is good news for my strix halo.

And I know no one will read this tweet because now I know how the algorithm works, but that’s how my brain works and I am a “burst tweeter”.

I know what I am going to write about this Sunday.

I wasn't aware of this "hardware compatibility" feature from @huggingface.

I’ve been thinking a lot lately about the right way to work with AI agents.

The spectrum of agent relationships: interns, contractors and (swarm of) peers....

Saw it somewhere in Twitter, sorry for not giving attribution.

"Even if it is a time sink, it is usually more rewarding to just switch on the AI slop slot machine and start vibing with it."

Co-worker: "wait 2 mins, my computer is going slow" Me: "stop running that local model" (a few decades ago we would have blamed porn instead)

It's been half a decade since I last went to South Summit Madrid.

I am too optimistic to see here a European awakening?

We may start seeing more and more orange here

72 tok/s on my Strix Halo.

This is a perfect way to illustrate the alignment problem.

After the last few weeks going down the local AI inference rabbit hole, I wanted to come back to something I’ve been thinking about (and honestly struggling with) for a while now.

From Coder to Agent Manager: how I stopped writing code by hand after more than a decade of daily practice....

Productive week, but now I need to pay myself and write at some point the Sunday post.

Not me. > Python Programming: Claude Code vs Developer 🤣🤣🤣

I had this same idea the other day as I was listening to @lexfridman's podcast on this.

I am too simple.

Closing tabs is the best way to learning how to let go!

To my surprise, I am enjoying Grok lately.

Last week I wrote about the hardware side of running AI locally: why memory bandwidth beats raw compute, which machines are worth building, and where the market is heading.

Local LLM inference optimisations: from attention mechanisms to predictive decoding and software-model-hardware implementations....

After going through the notes from @javilop about the X algorithm, it is clear to me that is not the best channel for my content (and my personal behaviour).

Periodic reminder that I share updates to my posts in Substack notes.

This is also my experience.

This!

Month after month I need to allocate my savings somewhere.

This just happened, thank you all for your support.

And this is what I am going to build towards with the time I have!

More info about the talk :) The topic of the talk may ring a bell to myh newsletter subscribers. > adlrocha: I'll be speaking at CONVEX Summit 2026 in Madrid!

I'll be speaking at CONVEX Summit 2026 in Madrid!

GitHub Copilot’s recent shift to usage-based billing isn't a surprise, it's a wake-up call.

Why the AI bubble is a dependency trap: A guide to local LLM inference hardware...

I think X is consistently hiding my tweets.

Market timing!