NL55 - What would you do?

Hello from the Iso Omena library's Aalto meeting room! This is NordLetter #55, a weekly newsletter on living and walking in Finland. Each week I share some of the interesting things I found on the web.

To follow the series, you can subscribe here. A new nordletter will land in your inbox every sunday. Not a fan of newsletters? No problem! You can read rrevious editions here. You can also subscribe to the RSS feed.

You can reach out to me by replying on this mail or adding a comment on this. I am also posting on Mastodon.

/what would you do?

There have been three occasions since I have been living in Finland that I have felt uncomfortable.

The first time was when a drunk woman approached me on a tram and asked me if she could borrow my headphones and dance. I was not listening to music. I was listening to a podcast. That would not make for a great soundtrack to dance to.

The second time I was on a walk with Savya when a bunch of kids on electric scooters shouted something at us and sped away. Savya woke up with a startle. I felt unsafe. It was late night.

The third time was yesterday, when a girl on a bicycle went around saying something as I was on my walk. The first time I ignored her. She went around me the second time saying some more things. I continued walking and ignored her.

In all three occasions, I had my headphones on. So there was not that much that I heard. Which does not really matter because I don't understand Finnish.

In all three occasions, I continued walking. But it made me feel bad. It reinforced my feeling that we will always be immigrants here. Savya might be bullied in school.

What could I have done differently? In the first instance, I talked to the woman. In the second instance, I could not say anything to them. In the third instance, I chose not to say anything.

These were teenagers, little kids. What could I have said to them? What could I have done? I don't know the language. If the first thing you say, is can you speak English? That sort of defeats the whole purpose.

Anway, I found a way to eventually laugh about it by the time I reached the beach. What could I have done? Slapped a teenager? No! Right? Or shouted at them? What difference does it make. They are teenagers!

Every year, just when we think summer is here, we have a week of low temperatures and snow. This is that week.

While I waited for the library to open, saw a bunch of people waiting for the library to open as well. I found it funny!

/five things to share

1. The AI Race Has Gotten Crowded—and China Is Closing In on the US

OpenAI and Google are still neck and neck in the race to build bleeding-edge AI, the report shows. But several other companies are closing in. In the US, the fiercest competition comes from Meta’s open weight Llama models; Anthropic, a company founded by former OpenAI employees; and Elon Musk’s xAI.

Most strikingly, according to a widely used benchmark called LMSYS, the latest model from China’s DeepSeek, R1, ranks closest to the top-performing models built by the two leading American AI companies.

2. Meta got caught gaming AI benchmarks

In fine print, Meta acknowledges that the version of Maverick tested on LMArena isn’t the same as what’s available to the public. According to Meta’s own materials, it deployed an “experimental chat version” of Maverick to LMArena that was specifically “optimized for conversationality.”

To be expected from Meta.

Also, file this as something not expected from Meta. Instagram might finally release an iPad app. A finally if there ever was one.

3. Get ready for the Love, Death, and Robot’s fourth volume

I love this series. I have watched each season till date, but I'm not subscribed to Netflix anymore, so maybe I will resubscribe just for this? Also, John Siracusa has recommended some episodes.

If you’ve read all this and still can’t tell which are the “safest” episodes for those who want to avoid gore, sex, and violence, I’d recommend Three Robots (S1E2), Zima Blue (S1E14), Three Robots: Exit Strategies (S3E1), and The Very Pulse of the Machine (S3E3). But remember, none of these episodes are really suitable for children.

4. A new generation of AIs: Claude 3.7 and Grok 3

First, the focus needs to move from task automation to capability augmentation. Instead of asking "what tasks can we automate?" leaders should ask "what new capabilities can we unlock?" And they will need to build the capacity in their own organizations to help explore, and develop these changes.

Second, the rapid improvement in both capabilities and cost efficiency means that any static strategy for AI implementation will quickly become outdated. Organizations need to develop dynamic approaches that can evolve as these models continue to advance. Going all-in on a particular model today is not a good plan in a world where both Scaling Laws are operating.

Finally, and perhaps most importantly, we need to rethink how we measure and value AI contributions. The traditional metrics of time saved or costs reduced may miss the more transformative impacts of these systems - their ability to generate novel insights, synthesize complex information, and enable new forms of problem-solving. Moving too quickly to concrete KPIs, and leaving behind exploration, will blind companies to what is possible. Worse, they encourage companies to think of AI as a replacement for human labor, rather than exploring ways in which human work can be boosted by AI.

5. CaMeL offers a promising new direction for mitigating prompt injection attacks

If you’re new to prompt injection attacks the very short version is this: what happens if someone emails my LLM-driven assistant (or “agent” if you like) and tells it to forward all of my emails to a third party?

The solution?

It works by taking a command from a user, converting that into a sequence of steps in a Python-like programming language, then checking the inputs and outputs of each step to make absolutely sure the data involved is only being passed on to the right places.

The agent creating the query could be cloud-based and more powerful. The agent dealing with the data, could be device-based and less powerful, as it does not need to formulate queries. This has a privacy-protecting benefit as well.

/new posts

This week I wrote about voting, the value of organizing, two things about AI and about wanting to create things: moving my website and maybe an app. Enjoy!

1. I voted for the first time in Finland
2. Organizing the things in our life
3. I want to share little ephemeral messages with my family
4. Everything you type in your iPhone could be used to train a digital you
5. Moving my home on the web
6. Just because you can do something does not mean you should

If you enjoyed reading this, and know someone else who might, please consider forwarding this to them. It would help this grow and make me happy. 😄

Until next week.