Weeknotes 2024.03

Another tough week mentally. Which is incidentally why I’m a bit late.

Technology had to take a back seat due to very disturbing revelations here in Germany. Apparently far-right politicians, right-wing extremists (aka Nazis) and a few businesses people met to discuss the displacement and deportation of people “not German enough”, whether they have a German passport or not. If this sounds very Wannsee conspiracy to you, I had the same thought. It was even taking place so close to the Wannsee villa in a similar location, you could argue it was as close as it can be without German authorities being immediately up their asses for crimes against the constitution. My take is: if each and everyone participating will not spend 20-25 years behind bars, our laws to protect us against Neo-fascist movements are not strong enough and need serious amendment.

To refocus my mind, I was seeking refuge in a few relaxing talks about creative coding, language and compiler design. And some Star Trek: Strange New Worlds, which is excellent in so many ways and balm for the soul. I also reviewed a few interesting branding projects.

I’m also thinking about restructuring how I post links and move to singular links with commentary. This week will be business as usual, but let me know what you think via socials.

Changelog

• Started canary testing Eleventy 3.0 previews. It gets better not bigger. Which is a unique trait these days.
• Website: fixed minor bugs.
• New project: remixing a few existing ideas about technology scouting & assessment. Estimated public release: 4-6 weeks from now. Need to relearn a technology.

Linklog

AI

• This could be in security as well. Text Embeddings Reveal (Almost) As Much As Text. TL;DR: Treat embeddings in your fancy vector database like normal text, especially if they contain personal information. It also means in order to be GDPR/CPRA compliant, you better know which embeddings contain personal information from a specific user! (via)
• Is GenAIs Impact on Productivity Overblown? A valid question to ask, since you need to double check the results and what not. Nobody disputes that there is a (positive) impact. Whether it reaches the heights of some pundits is questionable. I’m also in the group of sceptics on the longterm impact in organisations and believe there might be a hangover event coming for organisations which go too far too soon.
• AI firms’ pledges to defend customers from IP issues have real limits. Fine-tuning can easily push you outside of indemnities. Also depending on your jurisdiction, a legal claim against the user can’t easily be offloaded to the vendor. The solution once again: know your sources.
• The legal battle over copyrighted material used to train models has begun. OpenAI admits it can’t train useful models without copyrighted material, stretches the definition of fair use quite a bit, which the news outlets obviously fought in their Senate hearings. Then OpenAI accused The New York Times of prompt engineering their verbatim material, which I think is a red herring, when it comes to copyright claims. It’s gonna get messy 🍿
• The Mistral of Experts paper is out and I agree with Simon, it’s “Unsurprising but disappointing: there’s nothing in the paper at all about what it was trained on”.
• WikiChat: Stopping the Hallucination of Large Language Model Chatbots by Few-Shot Grounding on Wikipedia. You can’t see it and you don’t know why, but when I discovered this paper, I was smiling on the inside and the outside. 😎 (via)
• More than an OpenAI Wrapper. It makes Perplexity even more desirable. The answers are way more useful than ChatGPT even when using OpenAI’s models. With its sources footnotes it may become researcher’s (and publisher’s) best friend. Highly recommended. ⭐

Crypto

• First it was a hoax via a hack, then it turned out to be true: SEC approved the first Bitcoin ETF. Bitcoin basically went “if you can’t beat them, join them”.

Programming

• Is Multi-tenancy still the right approach to scale? Much of what I’m pondering these days asks the same question. It’s good to see others thinking that as well. The tech-sector is so enamoured with scaling that most don’t even stop to ask themselves whether that is a good idea. I’m betting three things:
  • We reached the point where running an instance for each user or team is feasible and a valid complexity trade-off, esp. in B2B cases it’s already pretty common.
  • Most scaling for distinct groups could be met with bigger (virtual) iron.
  • Even in situations where there is no easy way to splice the load into distinct groups, having many instances interact via explicit federation might be not only a valid scaling complexity trade-off, but also be desirable in terms of security and privacy.
• Cold-blooded software. An interesting analogy to think about software and a much better take than most. How is it different to the Enterprise policies I fought all those years ago? It makes little assumptions about the environment. The Enterprise software that I disliked so much, tried to artificially freeze the environment they are in, so their hot-blooded software could survive, until they could not anymore. Then they paid the price for a massive delta. (via)
• Delayed Open Source Publication is inspired by patents: the creators grant themselves an exclusivity period and only release dated (maybe even obsolete) code as open source software. Everyone loses.
• Domain Driven Design is 20 years old. Enterprise Integration Patterns, 20 years as well now, is one of the books that equally gained and lost due to its “Enterprise” focus. Both are good tools to have in the arsenal, but as with all good things, some have overdone it (which is where the backlash usually comes from).

Security

• Supply chain security #1: If you ever wondered how malicious code gets into libraries in the first place, read this report on how it was done by exploiting wrong Github settings. (via)
• Supply chain security #2: How Golang is managing it.
• And your usual bunch of zero days, exploits and other nastiness. 😞

CES highlights

Spatial Computing

The new term for AR/VR/MR, basically any kind of computing that requires you to wear an audio-visual headset.

• Sony and others presented a slew of Apple Vision Pro competitors and Qualcomm the required chip to compete with Apple’s R1. The heat is on.

We’re taking bets on how long it takes for companies which have been Blockchain experts in 2021-2023 and are AI experts since 2023, to announce they are the go-to Spatial Computing experts in 2024 and forward. 😂

Other reports to note

• The reporting on the various “adult entertainment” gadgets is hilarious. Some journalists found pleasure in convoluted phrasing, pun fully intended 🤣
• Transparent OLED TVs are a thing. I don’t know whether it’s useful, but I’m astonished nonetheless.
• I have to agree, Apple is in a unique position to create a desktop OS / tablet OS hybrid without putting two computers into one device. Maybe if those Frankenstein combos catch on, Apple will budge.
• E Ink is an exciting technology. I don’t know about you, but I prefer the “built in customisable case” to the toilet.

This was a long one. Again let me know whether you would like to see individual link posts via the socials.