He/him. Dad and husband. I work at https://xitter.cf/t.co/aVBqOzwwD1. My main hobbies are sailing, mountain/road biking, and creating an ultra low power personal computer.

Trondheim, Norway
Joined November 2009
Can someone start training frontier models to not write pkill commands that match themselves
1
1
91
> The L100 summary never printed because the pkill command killed its own shell before it could finish
1
48
> The deploy succeeded, but the sampler restart command killed its own SSH session again since pkill -f sim_sampler.py matched the shell itself.
28
Combining business and pleasure, likely the last time this season.
4
79
Hard but true
great post from @exedev team - the same thing that has always been true of software engineering is still true: Do fewer things in parallel, do them very well, optimize for getting incremental value out to users, FOCUS It's been 30 years since the XP and Lean folks said "limit work in progress (50+ if you count the eli goldratt thing) - its still true blog.exe.dev/etoomanythings
1
1
81
And I'm terrible at it 🫥
14
Please help, every cpu you can spare counts. I'm getting tired of this
🤖 Made with AI
2
66
My wifi/cloud connected panel heaters stopped working with the app a few years ago. Me and Opus 4.8 just reverse engineered it all, found a takeover, flashing custom modified firmware, full local control with no cloud required. Hell yeah.
2
5
240
Mill gen 2 heaters, previous fixes required soldering on new controllers. Will publish after giving them 90 days to react
43
yeah why force all the data to flow through the lossiest, most expensive layer
slowly we're all realizing that tools should be called from code, not from within the llm api
1
163
Andreas Eriksen retweeted
I’m excited to join the speaker lineup at Vespa.ai Live! My session will explore Nuances of Binarized Embeddings-Based Retrieval If you’re attending, let me know — would love to connect at the event.
2
2
2
9
553
Andreas Eriksen retweeted
My opinion on tokenmaxxing is companies shouldn’t mandate/constrain any tools at all and then evaluate software developers by output / (salary + token use)
32
9
5
284
29,928
Andreas Eriksen retweeted
Vespa isn't only powering American AI, now it's also @MistralAI's retrieval solution
1
2
6
540
Andreas Eriksen retweeted
bm25 is nice and all, but you won't believe how easy it is to improve upon it with and how much more you can squeeze from lexical features in @vespaengine
3
5
2
26
2,007
Bound to happen, very happy to see it!
Claude Code is finally an RLM (oct 2025), congrats to Anthropic :-)
2
269
Yeah this was great. Also feel a deep connection with wanting to do stuff but not feeling qualified lol
I loved the Project Hail Mary movie (and the book). Science, technology, competence, and openness – this is my culture.
1
80
PSA: btop is amazingly nice. Among the first things I install on any linux box I use
1
3
849
Andreas Eriksen retweeted
Hypencoder is now ready for production using Vespa! Props to @andreer for making it happen! Really great stuff!
Just added a sample app for how to search with hypencoder models on Vespa. A large meta-model that generates a small query-specific model that scores your docs - it feels like pure science fiction, but of course we can do it: github.com/vespa-engine/samp…
1
2
1
13
1,402
Andreas Eriksen retweeted
A metamodel that generates weights for a NN for each query on the fly and using the same net for ranking documents against the same query - cool idea by @Julian_a42f9a But, can you do it completely within @vespaengine , at scale, with reasonable latency (a few hundred ms)? @andreer: hold my beer... (red bull) 🤯
Just added a sample app for how to search with hypencoder models on Vespa. A large meta-model that generates a small query-specific model that scores your docs - it feels like pure science fiction, but of course we can do it: github.com/vespa-engine/samp…
1
1
6
234