iiinova — where your digital self keeps evolving

Nicolay Gerold @nicolaygerold.com · 19/05/2025

So I am not directly responsible for it. Which is just off. At the moment, it's just a massive waste of time. Agents should live where I do the work unless they can automate me with 99% accuracy.

020

Nicolay Gerold @nicolaygerold.com · 19/05/2025

make some small adjustments and push again (or rather just discard it). It makes sense that it is working locally, where the dev is working. I think it also creates a sense of detachment from the code. Usually when I pull something, it's the work of someone else....

120

Nicolay Gerold @nicolaygerold.com · 19/05/2025

Not sure whether Codex or Devin are saving me time. The workflow is off. They are made to push to the remote, which completely breaks the workflow (working directory -> staging -> remote). So you push or make a pr, pull it down, check whether it's working or ...

120

Nicolay Gerold @nicolaygerold.com · 19/05/2025

Garmin (yesterday) : you did a hard workout, minimum recovery 48 hours Garmin (today) : time to run 24 minutes at 5k threshold pace

010

Nicolay Gerold @nicolaygerold.com · 17/05/2025

Iteration cycles are really long. Create a task, send it off, get it to local. See how it looks (frontend) or run your test suite. Then continue.

000

Nicolay Gerold @nicolaygerold.com · 16/05/2025

OpenCode is really stepping it up. I love the sidebar stuff. Really plays well with nvim.

000

Nicolay Gerold @nicolaygerold.com · 15/05/2025

i dream of a world with a single consistent markdown format

010

Nicolay Gerold @nicolaygerold.com · 13/05/2025

@transistor.fm When I put and episode in "Draft" and then schedule it a few days out, the start date in the analytics should still be the day I post the episode not the day I created the draft

110

Nicolay Gerold @nicolaygerold.com · 13/05/2025

@howaiisbuilt.fm is back. most AI conversations obsess over capabilities. This one focuses on constraints - the right ones that make AI actually useful rather than just impressive demos. open.spotify.com/episode/1MbZ...

100

Nicolay Gerold @nicolaygerold.com · 30/04/2025

I think there UX is too wish for, bad defaults, bad keymaps, also that I can edit anywhere in the chat is really weird :D Otherwise agree

000

Nicolay Gerold @nicolaygerold.com · 30/04/2025

Why does only Zed have autocomplete in the chat

120

Nicolay Gerold @nicolaygerold.com · 14/03/2025

trying to stay backwards compatible :D

000

Reposted by Nicolay Gerold

Daniel Davis @trustspooky.bsky.social · 17/02/2025

Thanks to @nicolay.fyi for giving me the opportunity to talk about @trustgraph.bsky.social on How AI Is Built! Labels such as "facts", "observations", and "assertions" take on new meanings when we begin to consider time. Click 👇 to watch the full episode! 🎙️ youtu.be/VpFVAE3L1nk?

youtu.be

Temporal RAG: Embracing Time for Smarter, Reliable Knowledge Graphs

YouTube video by How AI Is Built

063

Nicolay Gerold @nicolaygerold.com · 13/02/2025

Does anyone have a clue about what data structure is used for Google Docs? Suggestions, comments, collaborative editing,...?

020

Nicolay Gerold @nicolaygerold.com · 13/02/2025

That's a shame for the technology that is actually developed here. We have Mistral AI, Black Forest Labs working on the state of the art. We should support and celebrate their breakthroughs, not when a foreign company decides to open an office to sell to EU companies.

120

Nicolay Gerold @nicolaygerold.com · 13/02/2025

We also need to celebrate the breakthroughs we make here in the EU. In the last few days politicians started to celebrate OpenAI moving to Munich as a breakthrough for the EU and technology in Germany.

120

Nicolay Gerold @nicolaygerold.com · 13/02/2025

- Limited liability protection for innovative technology testing - Fast-track certification processes for new products - 2-week fast-track work visas for technical roles - Remote work permits across EU without additional paperwork - Reduced mandatory healthcare contributions for first 10 employees

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

- Safe harbor provisions for data experimentation within defined boundaries - R&D tax credits from day one - Reduced social security contributions for the first X employees - Fast-track banking setup and simplified financial reporting

110

Nicolay Gerold @nicolaygerold.com · 13/02/2025

Otherwise we force our founder to focus on taxes and accounting and not building something that sets them apart. We need a "Startup Innovation Pass": - Temporary exemptions from regulations with basic safety and ethics standards - Tax exemptions for startups under certain revenue/user thresholds

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

Otherwise we set them up for failure from the get-go. We need easy ways to open up a new company. Otherwise we don't incentive people to fund, because it's just easier to take high-paying jobs from US startups or big-tech. We need tax exemptions for new startups for the first year.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

Our startups are left behind. By the time the EU startup achieves compliance, the US company has gone through multiple product iterations. We need sandboxes that allow startups to innovate, while breaking some of the rules.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

We are creating fights on unequal footing. US startups can compete without the same regulations. Incumbents can compete with a bunch of lawyers figuring out ways around regulations.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

Or they can't offer features the US competitors can. I am working on a startup right now, where we cannot offer a feature because of GDPR. Even though all the information is publicly accessible. It just biases our customers to not go with us.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

The regulation EU startups are facing (GDPR, AI Act, DSA,...) are great in spirit, but when startups have to think regulation first, they are forced to move slower than the non-EU competitors.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

EU mobilizes 200 billion Euros for AI. Unless we have a massive political change that money will just go to waste. For innovation to happen, we don't need money first, but deregulation. We cannot work on breakthrough technologies, when we are at constant fear of being sued.

120

Nicolay Gerold @nicolaygerold.com · 13/02/2025

arxiv.org/pdf/2502.05167 Sorry :D

arxiv.org

010

Nicolay Gerold @nicolaygerold.com · 13/02/2025

Long-context isn't solved. We just had bad tests. What this means: - RAG isn't dead - it's more important than ever - Breaking text into smaller chunks still helps - Models need to get better at reasoning, not just matching at long contexts

220

Nicolay Gerold @nicolaygerold.com · 13/02/2025

So no more matching, but actual reasoning. Even top AI models struggle as texts get longer. - o3-mini degrades to 18.9 at 32K context window - Deepseek R1 to 20.7 at 32K - Most models even struggle past 4k For most tasks good retrieval + smaller contexts will improve your performance.

110

Nicolay Gerold @nicolaygerold.com · 13/02/2025

For example: Question: "Which character has been to Dresden?" Hidden fact: "Yuki lives next to the Semper Opera House" (opera house in Dresden) The AI must know that the Semper Opera House is in Dresden to make this connection.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

NoLima is a benchmark to test if AI models can truly understand long texts. - Insert one key fact (called a "needle") into a long text of irrelevant information - Ask the AI to find and use this fact to answer a question - The twist: The question and the key fact use different but related words

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

BUT! The tests often let AI find answers by matching words that appear in both the question and text. - Question: "Who was the first person to land on Mars?" - Text: "...and after years of preparation, Sarah Chen became the first person to land on Mars..."

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

People say long-context is solved, because we perform well on "needle in the haystack". We place an answer somewhere in the context window and ask a question. When the model can answer correctly we can say, the model works at the given context length.

100

Nicolay Gerold @nicolaygerold.com · 13/02/2025

RAG is dead. Long live RAG. LLMs suck at long context. This paper shows what I have seen in most deployments. With longer contexts, performance degrades.

161

Nicolay Gerold @nicolaygerold.com · 04/02/2025

foqos.app "Physically block distracting apps" Grab a NFC tag, lock your app, throw NFC tag off balcony, start coding.

foqos.app

Foqos - Focus with NFC

010

Nicolay Gerold @nicolaygerold.com · 01/02/2025

Anything that's a formula is ignored and not stored, since it can be recomputed in the sheet. Multiple sheets are probably not possible unless one sheet holds the data and the rest are computations or visualizations.

010

Nicolay Gerold @nicolaygerold.com · 01/02/2025

I am dreaming up an Excel backend. Excel as the frontend, Iceberg as the backend. But how do you handle formulas, pivot tables, ...? How do you malformatted sheets? You probably have to put restrictions on what you can do with Excel so it's valid to be stored.

121

Nicolay Gerold @nicolaygerold.com · 01/02/2025

human clipboards is the best description I have heard for people who just copy and paste code and errors from and to ChatGPT

010

Nicolay Gerold @nicolaygerold.com · 01/02/2025

The more I learn about MLOps, the more it's ops first, ML second. Start with: - Build serverless apps - Write infrastructure code - Set up CI/CD - Run VMs & databases - Deploy Kubernetes ML comes after you can build things that work.

011

Nicolay Gerold @nicolaygerold.com · 31/01/2025

TIL: In Postgres, you can dollar quote strings with a lot of ' so you don't have to double each '. $$Dianne's horse$$ $SomeTag$Dianne's horse$SomeTag$

000

Nicolay Gerold @nicolaygerold.com · 31/01/2025

Youtube: youtu.be/6xQL4w4KtPg Spotify: open.spotify.com/episode/2Pc6... Apple: podcasts.apple.com/us/podcast/i...

youtu.be

Inside Vector Database Quantization: Product, Binary, and Scalar | S2 E23

YouTube video by How AI Is Built

000

Nicolay Gerold @nicolaygerold.com · 31/01/2025

Dropping some new episodes on @howaiisbuilt.fm . Links below.

111

Nicolay Gerold @nicolaygerold.com · 25/01/2025

New podcast with Alex Garcia on search in sqlite

120

Nicolay Gerold @nicolaygerold.com · 18/01/2025

TIL: Ionos has 1€ email for your custom domain.

020

Nicolay Gerold @nicolaygerold.com · 18/01/2025

Coming back and back again to the same question: How can I make it easy to experiment? With genAI it will be easier to generate alternatives and options. So for almost anything, I want to generate a lot of options, vote for the best one, test it in a sandbox, A/B test, deploy it... (rollback)

010

Nicolay Gerold @nicolaygerold.com · 18/01/2025

Really interesting paper. Machine Learning: The High-Interest Credit Card of Technical Debt static.googleusercontent.com/media/resear...

static.googleusercontent.com

060

Nicolay Gerold @nicolaygerold.com · 18/01/2025

That's surprisingly on brand.

000