Maximilian Scholz

Astra... Show me on this Mac mini where the ai agent touched you.

I feel like over the last 2 months I've slowly slipped into a place where I just do things rather than trying to get others to see why it would be a good thing to do things. It's liberating.

Richard, what did you do to the owls?

If you bitshift this data.frame just right, the serialised binary can be interpreted as NN weights

┏━━━━━━━━━━━━━━━┳━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━┓ ┃ Model ┃ Tasks ┃ Passed ┃ Pass Rate ┃ Avg Latency ┃ Input (M) ┃ Output (M) ┃ Cache (M) ┃ Total Cost ┃ ┡━━━━━━━━━━━━━━━╇━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━┩ │ glm-5 │ 206 │ 27 │ 13.1% │ 91.1s │ 0.00M │ 0.00M │ 0.00M │ $0.00 │ │ gpt-oss-120b │ 205 │ 32 │ 15.6% │ 68.3s │ 2.35M │ 0.20M │ 0.00M │ $0.33 │ │ minimax-m2.5 │ 205 │ 30 │ 14.6% │ 96.3s │ 14.11M │ 0.15M │ 0.84M │ $4.43 │ │ nemotron-3-n… │ 206 │ 27 │ 13.1% │ 90.7s │ 0.00M │ 0.00M │ 0.00M │ $0.00 │ │ step-3.5-fla… │ 206 │ 27 │ 13.1% │ 61.5s │ 0.00M │ 0.00M │ 0.00M │ $0.00 │ │ trinity-larg… │ 170 │ 78 │ 45.9% │ 367.3s │ 53.65M │ 1.73M │ 0.00M │ $15.14 │ │ trinity-mini │ 204 │ 57 │ 27.9% │ 115.7s │ 1.58M │ 1.83M │ 0.00M │ $0.35 │ └───────────────┴───────┴────────┴───────────┴─────────────┴───────────┴────────────┴───────────┴────────────

Well, some parts are working and some not so much but we are measuring *something*. Now on to make it better and figure out how to fund running this on non-free models...

Because I stumbled over o/u stuff again. My friggin pipe!

Randomly stumbled over this again and am reminded again why "30 days of bleed" is the best title for any blog post written about #over/under

Cover for Pumpkin Spice a TTRPG showing a witch pouring coffee in a pumpkin spice coffee cup.

So proud to announce my collaboration on "Pumpkin Spice" a cozy, magical TTRPG, published by #AcheronGames

Check it out on BackerKit (link in bio) and discover all the amazing perks of backing!

Pour yourself a cup of tea and enjoy the vibe!🪴🍵

#PumpkinSpiceRPG

Screenshot of a terminal window showing benchmark results: ────────────────────────── Model Performance ────────────────────────── ┏━━━━━━━━━━━━━━━━┳━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━┓ ┃ Model ┃ Tasks ┃ Passed ┃ Pass Rate ┃ Avg Latency ┃ ┡━━━━━━━━━━━━━━━━╇━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━┩ │ minimax-m2.5 │ 106 │ 15 │ 14.2% │ 102.9s │ │ step-3.5-flash │ 194 │ 26 │ 13.4% │ 60.5s │ └────────────────┴───────┴────────┴───────────┴─────────────┘

I was supposed to submit a Stan design proposal but got sidetracked... Also, it's a good thing that api keys can have limits... so I've heard.

porngrind never left

Turns out unpaid overtime costs us around 400k jobs in Germany.

Started working on my own R-focused llm benchmark and all of a sudden the zai pro plan didn't feel infinite anymore >.>

Thinking that more companies need a jester. Someone who is allowed to make fun of the king and point at all the ugly and dumb things everyone else silently agreed on not talking about. Someone who can call out the invisible clothes of the emperor and not be lynched but taken serious instead.

Don't be shy.

@hadley.nz I have something in mind I want to tinker with and need some showcase packages for that :)

What R packages have the highest code quality and should serve as examples to learn from in your opinion? #rstats

I'd have a couch to crash on for 2 ;)

You should just come visit Hamburg :*

Not to throw shade but not being exclusively surrounded by high-achievers for the first time since high-school is doing wonders to my mental health. Also, many recurring comments from the past make much more sense now.

If the pen is mightier than the sword, the eraser is by extension also mightier. Few understand this.

Mam, this is an eraser.

It's also kinda slow and they quietly removed the access to new frontier models point from the plus plan.

TIL about the narcissist-aspie trap 👀

That's why I only work with raw commit hashes.

We have the industry to see methods of times past in the wild. It's like Jurassic Park.

What kind of crazy are talking? Do I get boons?

Is it?
github.com/discourse/di...

Screenshot of Claude writing: "why is my PATH fucked?"

mood

Well you found me, congratulations.
Was it worth it?
The only thing you've managed to break so far, is my heart
This isn't brave.
It's murder.
What did I ever do to you?
You don't even care, do you?
Please proceed into android hell.

A black and white drawing of a corpse that reads “If I just try harder, then maybe it won’t kill me.”