Category: Software Development

Posted on

Thoughts on “AI 2027”

  1. Home page: https://ai-2027.com/

  2. Authors addressing some critique, explaining “Why America Wins” (Ed.: Would that even matter much in an end-game between AI and humans? I doubt it. Rather just a bias.)

  3. A YouTube video summary: AI 2027: A Realistic Scenario of AI Takeover

My take:

  • It’s good and important to think about AI, the use/misuse/abuse of AI, scenarios why, how and when AI could take over power (and obviously: How to prevent this).
  • I’m astonished that one can spot quite a few biases (e.g. the good old USA vs. China power struggle, good government vs. bad private companies, perceived technical or intellectual superiority) and seemingly simplistic assumptions (i.e. that the further AI development is merely a matter of compute, ruling out fundamental theoretical and practical limitations on the path to AGI, or even some obvious potential geopolitical events and decisions; some of which have already manifested -> e.g. student visas, wars). Sure, the scenarios have to be largely based on known knowns, but they should account more for known unknowns and the possibility of unknown unknowns.
  • Is AGI achievable in the next 5 years? I doubt it, mainly due to – still valid – fundamental theoretical and practical constraints of the prevalent approach to achieving AGI using LLMs, like the symbol grounding problem. LLMs depend on symbols (tokens) – digitally sharply defined. But the analogue world is exactly the opposite: There are no symbols (except those introduced by living beings, i.e. humans, animals).
    Then again, I doubt this matters much in the described scenarios, as the states described in these scenarios are, because of the widespread digitisation, probably achievable with mere control over the digital world. To achieve this in a digital world, true (universal) AGI isn’t required; it’s sufficient to achieve AI that is close enough to “digital AGI” to convince humans that digital AGI has been achieved (as we’re so dependent on the digital world already).
    This assumption is both frightening and reassuring: On one hand, it makes far less sensational, but equally dangerous scenarios of AI (partial) dominance more likely; on the other hand, it opens up a possible ‘escape route’ for humanity: we must “simply” ensure that the analogue world always retains the upper hand and that we could exploit this in our favour in a final battle against AI if necessary.
  • That said, we should urgently look to make our analogue supply and communication networks more resilient, and ironically not via digitalisation as we have done so far. Rather, it is precisely through the deliberate, explicit renunciation of digitalisation.
    In my worst-case scenario, we would thus have to assume that an AI – even one lacking true/universal AGI – could take over the digital world (and therefore also many highly powerful and dangerous interfaces to the analogue world -> weapon control systems, bio-labs, food chain, water supply systems, power systems, hardware/robot/clone factories, air control, traffic lights, navigation systems, satellites, emergency networks, life-saving medical devices, subtle but decisive influence over political decisions and communication, etc.), but not a purely analogue world (thanks to the assumed lack of universal AGI).
  • Once true, universal AGI has been achieved, it’s probably “game over”, unless humanity could somehow convince the AGI of a lasting, sustainable win-win deal (but in the end, the fight for resources and power will be real).

 

 

Posted on

NVIDIA developer workshop: “s1: Simple test-time scaling” paper

At the NVIDIA developer workshop days I attended last week, the following paper was highly recommended:

“s1: Simple test-time scaling” by Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, Tatsunori Hashimoto
(PDF)

Github project: https://github.com/simplescaling/s1

Not directly related, but an interesting application of the NVIDIA RAPIDS Accelerator that was also presented: https://aws.amazon.com/blogs/industries/accelerating-fraud-detection-in-financial-services-with-rapids-accelerator-for-apache-spark-on-aws/

Posted on

An open-source AI hedge fund team by @virattt

“The AI Hedge Fund is an educational model using AI agents to simulate trading decisions, emphasizing various investment strategies and analyses.”

Looks very interesting and inspiring, and @virattt seems to be a cool guy.

https://github.com/virattt/ai-hedge-fund/tree/main

Caveat/Disclaimer: “Can LLM-based Financial Investing Strategies Outperform the Market in Long Run?” by Weixian Waylon Li, Hyeonjun Kim, Mihai Cucuringu, Tiejun Ma – 11 May 2025 (PDF)

 

Posted on

Nomad Network – Communicate Freely

Off-grid, resilient mesh communication with strong encryption, forward secrecy and extreme privacy.

Nomad Network allows you to build private and resilient communications platforms that are in complete control and ownership of the people that use them. No signups, no agreements, no handover of any data, no permissions and gatekeepers.

https://github.com/markqvist/NomadNet

Information is the basis of society in the digital world.
Freedom of expression/speech and free access to information are indispensable prerequisites for democracy.

Sadly, there are misleading and dubious organisations, lobbyists, parties, secret services, legislative and executive bodies that work against these liberal, enlightened and humanitarian values and demand or advocate censorship, online and offline.
This is a big threat for civil society – you and me, all of us.

And this is what makes projects like NomadNet important and valuable.

Stand up for freedom of expression/free speech and fight censorship!

(And as always, please note the disclaimer: All software and hardware can have security bugs).

Posted on

New Apple Magic Keyboard with Numeric Keypad is surprisingly good, even excellent!

Recently, I’ve managed to render my beloved old Apple Keyboard (full-size, with numeric keypad) useless – accidentally pouring half a glass of tap water over it was sufficient, unfortunately (due to the mineral ions in the tap water; distilled water wouldn’t have conducted electricity and thus wouldn’t have shorted circuits; on the other hand, drinking distilled water would probably shorten your life, so please don’t consider doing this).

Luckily, I could temporarily use a similarly old, compact Apple Bluetooth keyboard instead. As I really wouldn’t recommend that keyboard for everyday work though (poor, bubbly typing experience, odd placement of keys requiring weird function key combinations, no numeric keypad), I had to order a full-size keyboard as a replacement again, so I ordered one of the new Apple Magic Keyboards with a Numeric Keypad (in Space Gray and I really like that, but the colour doesn’t matter in regard to the typing experience):

https://www.apple.com/ch-de/shop/product/MRMH2SM/A/magic-keyboard-mit-ziffernblock-schweiz-space-grau

(above is the Swiss German version, US version: https://www.apple.com/shop/product/MRMH2LL/A/magic-keyboard-with-numeric-keypad-us-english-space-gray)

I first thought that the even smaller lift of the keys (luckily with scissor and not butterfly switches) of this new keyboard would be very disturbing and that I’d have a hard time getting accustomed to it.

To my big surprise however I got accustomed to this new typing experience within a couple of hours already and now, after about 3 weeks of using it, I can confidently say: I love this new Apple Magic Keyboard with Numeric Keypad even much more than my previous, old Apple Keyboard with Numeric Keypad and wouldn’t want to switch back anymore.

Typing with it feels so immediate, so quick and so “raw” and “crunchy”, it’s literally almost addictive. I can type considerably faster with it than with the previous keyboard, let alone any regular IBM-type keyboards (although I like those too, for their build-quality, for the interesting history and stories behind them, for their customizability and standardization, for the bustling keyboard enthusiast scene around it). Further, typing for a prolonged time feels much less tiring for the fingers, hands and forearms.

It feels as if you had to work with a wobbly tool for quite some time, then all of a sudden, get a very precise and exact instrument, like e.g. skiing with racing skis vs. with allround skis. It’s pure joy!

The difference is difficult to describe, so I would recommend you rather go experience it yourself and judge for yourself. For me, it’s my most favourite keyboard so far.

I’m even thinking of getting one for the Windows workstation at work too, it’s that good.

Posted on

GitLab 10.5 and later: Solution for error “Validation failed for domain” with Let’s Encrypt

GitLab 10.5 introduced built-in support for Let’s Encrypt.

Unfortunately, if you follow the official GitLab instructions how to enable Let’s Encrypt support, you may encounter the following error when rebuilding GitLab:

Running handlers:
There was an error running gitlab-ctl reconfigure:

letsencrypt_certificate[yourhost.yourdomain.com] (letsencrypt::http_authorization line 3) had an error: RuntimeError: acme_certificate[staging] (/opt/gitlab/embedded/cookbooks/cache/cookbooks/letsencrypt/resources/certificate.rb line 20) had an error: RuntimeError: [yourhost.yourdomain.com] Validation failed for domain yourhost.yourdomain.com

Running handlers complete
Chef Client failed. 11 resources updated in 11 seconds

Warnings:

Let’s Encrypt is enabled, but external_url is using http

The last line is rather misleading, as the domain validation can apparently also fail if one sets  external_url = “https://yourhost.yourdomain.com”

As a workaround, add the following two additional lines to /etc/gitlab/gitlab.rb (hat tip to Kai Mindermann and Thomas Jost for the hints):

nginx['redirect_http_to_https_port'] = 80
nginx['redirect_http_to_https'] = true

 

So, all in all, you need to set in /etc/gitlab/gitlab.rb:

external_url 'https://yourhost.yourdomain.com'

and add the following lines (adjust the notification e-mail address):

letsencrypt['enable'] = true
letsencrypt['contact_emails'] = ['gitlab-notifications@yourdomain.com'] # optional
nginx['redirect_http_to_https_port'] = 80
nginx['redirect_http_to_https'] = true

Make sure that your firewall doesn’t block access to ports 22 (SSH), 80 (HTTP), 443 (HTTPS).

After that, reconfigure GitLab (in a shell):

# gitlab-ctl reconfigure

That’s it! You can now register/login at https://yourhost.yourdomain.com.