Hans-Peter Zorn

Head of AI @inovexgmbh #nlproc

[Blaise Reymondin @reymondin]

@ylecun @eladgil Born in Zurich 🇨🇭:

Vision Transformer (ViT) Google
SigLIP, BiT, MLP-Mixer (Google Brain)
Google Lens
Google Maps
pLSA (foundation of modern NLP)
Microsoft HoloLens vision/SLAM
Multimodal core of Gemini (DeepMind ZH)

[Azeem Azhar @azeem]

Love a bit of exponential

[Azeem Azhar @azeem]

Weird conclusion

[Andrej Karpathy @karpathy]

- Drafted a blog post
- Used an LLM to meticulously improve the argument over 4 hours.
- Wow, feeling great, it’s so convincing!
- Fun idea let’s ask it to argue the opposite.
- LLM demolishes the entire argument and convinces me that the opposite is in fact true.
- lol

The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.

[François Chollet @fchollet]

ARC-AGI-3 is out now! We've designed the benchmark to evaluate agentic intelligence via interactive reasoning environments. Beating ARC-AGI-3 will be achieved when an AI system matches or exceeds human-level action efficiency on all environments, upon seeing them for the first time.

We've done extensive human testing that shows 100% of these environments are solvable by humans, upon first contact, with no prior training and no instructions.

Meanwhile, all frontier AI reasoning models do under 1% at this time.

[Hans-Peter Zorn @data_hpz]

@yoavgo I have some MCP that can create quite nice images and charts for those typst-presentations: https://t.co/OcRX46jWG0

[Hans-Peter Zorn @data_hpz]

@yoavgo Nobody wants this, especially if you can have Claude generate Lean typst code (or LaTeX Beamer, if you must) for presentations that you can check into git.

[Simon Willison @simonw]

Thankfully the LiteLLM package has now been marked as "quarantined" on PyPI so attempting to install the compromised update via pip et al shouldn't work

[Hans-Peter Zorn @data_hpz]

The harness -- not the prompt -- is the product.

What's your approach: prompt, context, or harness?

6/6

[Hans-Peter Zorn @data_hpz]

The evolution of AI engineering in 3 years -- from hoping the LLM listens to ensuring the system enforces.

1/6

[Hans-Peter Zorn @data_hpz]

The shift is about WHO is in control.

Most teams are still in prompt engineering mode, trying to get the LLM to "behave." The real shift is architectural: build systems that constrain what the LLM can do.

5/6

[Hans-Peter Zorn @data_hpz]

How do you handle context loss when your AI agent works with large specifications?

Tulla is my experimental open-source implementation of Semantic SDD.

It is research, beware. But I am open for feedback.

no warranties.

https://t.co/Gi0VONXW36

6/6

[Hans-Peter Zorn @data_hpz]

Spec-Driven Development is the current trend: write a detailed spec, let the AI code.

But there the fundamental flaw.

1/6

[Hans-Peter Zorn @data_hpz]

SDD: hand the construction crew a bunch of textfiles and hope they remember page 42.

Semantic SDD: feed them verified instructions for the specific brick they're holding.

5/6