Survival of the fittest small AI

By Iskander Smit in weeknotes — May 6, 2025

Next to all the news in human-AI and beyond, some thoughts triggered thru vibe coding small AI and reflecting on survival strategies of a big tech player.

Midjourney: a swarm of small creature like things connected through invisible wires, working together for a shared goal

Hi all!

Weeknotes 338 - Thanks for reading my weekly newsletter, where I sharpen and share my findings and thinking. If you are new here, find a complete bio on targetisnew.com. This newsletter is my personal weekly reflection on the news of the past week, driven by my curiosity to understand the unpredictable futures of human-ai co-performances. Don’t hesitate to reach out if you want to dive deeper into the specific questions you ran into.

What did happen last week?

The final visual design touches of the Civic Protocol Economies report have been completed! Find it here. 🔥

Had some good conversations with people as expected, and dived even deeper into the analogies, abstractions and reasoning.

The AIxDesign festival was really nice. To be honest, based on the program, I found it hard to know what to expect. Was it a conference, an exposition, or a gathering? Of course, all of the above. The community operates between artistic interpretations and experimentation with AI and critical design. The crowd is great, and a good mix of works. The two workshops I did made me reflect on different elements (see one below), and the talks were good. And good to be reconnected with the people from MARAI from Barcelona.

It was a pity I could attend only one day.

What did I notice last week?

Check the news from last week, which I captured further down below. In short:

Meta introduces a chatbot and is rolling out its AI strategy in the best Meta traditions (you as product, etc.)
Advertising and commerce via AI chats have a lot of attention overall
Apple teaming up with Anthropic
ChatGPT rolling back uncanny models
Flattening humans through AI or new deep reading buddies?
Multimodal logistic robotics
The weekly relations of geopolitics and tech
And some futuring galore. Back to the futuring.

What triggered my thoughts?

Last week, I attended the AIxDesign festival (though I could only attend for one day), and the first workshop was on smallAI. This concept aligns well with the ideas of the conference, which is themed slowAI. The conference encourages being more thoughtful about using AI without dismissing it outright. The festival offered a nice mix of artists and critical design.

In the workshop, we downloaded a small model, a GPT2 model, to our computers to get hands-on experience. It was refreshing to dive deeper into the terminal for an hour, though too short to explore more thoroughly.

Meta embarked last week on a PR tour for their new AI companion integrated into existing social media platforms, with Mark Zuckerberg conducting some promotional interviews. One notable point he emphasized was their deliberate choice not to try to create the smartest AI but the fittest. When evaluating LLM models, traditional benchmarks focus on human intelligence, measuring typical human-designed tests for math, legal knowledge, etc. This is, of course, a valid strategy if we are working towards an AI that becomes AGI or at least an AHI - artificial human. But what if the ideal fit for AI is being quick and focused on responding in an understanding way, helping you find or connect information, not necessarily to solve the hardest problems.

This appears to be what Meta is aiming for. It seems not like a misguided strategy. Last week we discussed the combination of multiple AIs, working together in multidisciplinary teams, as a strategy to deal with problems that could be too wicked for even the smartest AGI. These small, specialized AIs might fit well as team members. The strategy seems to be: Diversify, and ultimately, AI might follow an evolutionary path in a Darwinian sense; survival of the fittest, or more precisely, the most suitable for specific contexts.

It feels weird that a critical festival full of engaged people follows a similar strategy to Meta, but we should not forget the differing intentions behind each approach. As Meta mentioned in its press release, their end goal is to offer an AI that finds the best-fitting media plan for companies via their channels. This means developing even more seductive advertising strategies to retain you as a product through applying AI packaged as friendly companions… Worth noting.

Our goal is to make it so that any business can basically tell us what objective they're trying to achieve -- like selling something or getting a new customer -- and how much they're willing to pay for each result, and then we just do the rest.

(…) more engaging experiences. This will come in two forms: better recommendations for existing content types, and better, new types of content.

Our focus for this year is deepening the experience and making Meta AI the leading personal AI (…)

(…) we're all going to have an AI that we talk to throughout the day -- while we're browsing content on our phones, and eventually as we're going through our days with glasses (…)

They enable you to let an AI see what you see, hear what you hear, and talk to you throughout the day.

These excerpts from the earnings call reveal Meta's vision. Mark his words…

To end on a more optimistic note, I am hopeful we can take greater control more and more, building our own AIs, ”vibe coding” our way toward technological autonomy, an approach that will become increasingly accessible. Though progress will likely follow two steps forward and one step back.

Dramatic things for enacting smart home social structures. A paper from Chi last week by some former researcher colleagues. Dramatic Things: Investigating Value Conflicts in Smart Home through Enactment and Co-speculation

The findings challenge the seamless and harmonious vision of smart homes conceived by technologists, proposing shifts in the common narrative: from value alignment to value transparency, from service provision to mutual care, and from autonomy to responsiveness.

*Nazli Cila, Maria Luce Lupetti, Luciano Cavalcante Siebert, and Janna van Grunsven. 2025. Dramatic Things: Investigating Value Conflicts in Smart Home through Enactment and Co-speculation. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25). Association for Computing Machinery, New York, NY, USA, Article 939, 1–17. https://doi.org/10.1145/3706598.3713138*

What are the plans for the coming week?

You can expect a new monthly newsletter at the Cities of Things substack. I am reading the draft entries for RIOT 2025 (and need to complete my own…). Looking forward to a new deep dive Human Values for Smarter Cities. Also attending a session with ESC network, and a potential presence of Wijkbot at Makerfaire Delft.

Looking at the calendar of events, the Amsterdam UX has an important topic of design for accessibility. And next week, an evening on AI & Human Creativity in Amsterdam, and at v2 Rotterdam; “the other AI: automated intimacies”.

Have a great week!

Notions from the news

Human-AI partnerships

Meta introduced a new chatbot, which you will probably have seen on your WhatsApp or other Meta premises, you could not avoid… Mark Zuckerberg did a promotional tour and interviews. Is private processing of AI chats a real thing or a trust-marketing strategy?

In the meantime, ChatGPT was updated with a new personality, or better said, a new attitude. And rolled back. Parallel Sam Altman is rolling out his ‘real human verifier’

And there was the Geoguessr meme with o3. Can it find a random place, and watch the reasoning as a new form of entertainment? And benchmark.

GeoBench

GeoBench is an LLM/LVLM benchmark for GeoGuessr.

A new Chinese model on the block. Qwen.

Kinda make sense: Apple teaming up with Anthropic to build the xCode-co-pilot. First step in more Anthropic Intelligence? Not a conscious AI for now.

Wikipedia uses AI for… optimizing and stimulating the work of human editors.

Are we flattening communication?

How long will prompt tools stay relevant?

Can we stop AI making humans obsolete?

Will a seamless experience of NotebookLM change you relation with long-form readings? Imagine all these people in the train having conversations with the things they are reading.

Synthetic media can also mean reviving legends from the past.

Robotic performances

This box-mover looks like an insect, which makes sense.

A social network for supporting self-driving capabilities.

Checking robots in TV shows. Inspiration for new formats?

New form of multimodal…

The robot logistics revolution is silently happening inside. Waiting to go outside on the streets?

Ok, one humanoid post per week should be ok.

Immersive connectedness

Thoughts on Time by Tina. Subjectivity and relativity, how to express these.

Does this feel like a promise finally met? One device to fit all…

Project Kuiper aka Amazon’s Starlink.

Tech societies

Everyone is wondering how jobs are infected. Duolingo is another company that is embracing AI to create more efficient workstreams with fewer people.

Are we living in the future that was projected in the past?

LLMs are coming for a piece of the advertising cake. New players, OpenAI experimenting with shopping, Perplexity earlier with a hyper-personalized browser, and existing tech; Meta focusing on offering an AI media planner. Google is testing ads in chatbots.