Weeknotes 232; the AGI gap

Hi! The 4th continues.

Still, GPT-4 is the talk of the town, and people start playing with it, finding out what its capacities are. I did use it for the first time as a coder; I did not before with 3, but it worked quite well. As a prototype tool and a way to tinker and visualise thoughts. Designing the most complex interface of the platform we are building with Structural helps to demo some interfaces. With a paragraph description and 15 min thinking and iterating, I had something that fulfilled that need. Nothing will ever be used, but as a think piece, it works very well.

There are numerous podcasts with Sam Altman of OpenAI on GPT-4 and beyond, On by Karen Swisher was rather good and compact and now I am still listening to the always lengthy podcast with Lex Fridman.

The one of Benedict Evans I still need to here, but reading this week's column of Benedict Evans (in the paid newsletter) on the hype cycle of GenML triggered some thoughts:

That is how Moore’s Law has applied to computing - just hit it with more compute! - and it’s one way that machine learning has evolved- just hit it with more data and more bigger models.

The question is if this will lead to AGI (Artificial General Intelligence) and if this intelligence is the same as human intelligence after all. Undoubtedly, we -as humans- will be outperformed on certain tasks. Still, without choosing a different strategy for increasing the capabilities of AI, it is more likely that human and machine intelligence will widen the gap than become similar. This is not a bad idea; let us stand on each other's shoulders rather than replace one with another.

Below much more articles and opinions shared last week. And a view of other topics too.

Events

I attended an evening meetup last week organised by Amsterdam UX, at argodesign. Guus Baggermans -who did a presentation at the summer edition of ThingsCon last year too- did a solid introduction to the role of (visual) AI for the work of UX and brand designers. And after the talk, we were invited to play with some tools to build collaborative pieces in Invoke.ai. The idea of creating a hidden zoomed-out version of images.

I also attended a part of the demo day of Amsterdam Smart Cities, the break-out session on Mobility as a Common. The presenters of the municipality made very clear how they work on shaping opportunities for new forms of mobility initiated by citizens and commons. However, on the other hand, they still are struggling with loosening control and having real trust in the intentions of citizens beyond their interests. It relates to the work on building trust in agreements through the promises architecture of Structural language, as the goals we have with the MUC AMS Cities of Things fieldlab Collect|Connect Community Hub. Useful to attend.

Events for the coming week

• Responsible AI meetup, 28 March, Utrecht
• Interaction 23 Redux, 29 March, Online
• Future Image Making, 5 April, Amsterdam
• ThingsCon Salon Listening Things, 14 April, Eindhoven

Found news from last week

As promised last week, I will try to be more selective to limit the number of articles. News on Twitter source code leaks are all over the internet already, and the opinion of Bill Gates on AI too.

Let’s start with a round of new AI upgrades like Adobe (Firefly), Levi’s (to increase diversity), Wolfram Alpha (the other way around, as a plug in in ChatGPT), the internet (or vv).

And some tips on prompt-writing for GPT-4.

Google opens up Bard to the world. The Verge is not that impressed.

Google’s Bard chatbot doesn’t love me — but it’s still pretty weird

Bard is hard to break and also hard to get useful info from.

The VergeDavid Pierce

In a direct comparison of ChatGPT, Bing and Bard later in the week, that does not change much. “And if you are shorting Google’s stock and want to reassure yourself you’ve made the right choice, try Bard.” In the meantime, ChatGPT is opening to the web.

ChatGPT becomes even more valuable with the new plugins. And connecting to the internet. Becoming even more 'app store-like.

ChatGPT Gets a Computer

It’s possible that large language models are more like the human brain than we thought, given that it is about prediction; that is why ChatGPT needs its own computer in the form of plug-ins.

Stratechery by Ben ThompsonBen Thompson

The Two minutes papers channel is -of course- also keeps track of the latest.

How does GenAI fit in the IoT? What we will discuss at ThingsCon Salon probably. Are we entering the third level (The Creat0r) of products as agents (see this paper)?

Where does generative AI fit in the IoT? - Stacey on IoT | Internet of Things news and analysis

Major publications are really worried about Alexa, Siri, and Google’s digital assistant these days. That’s because, so far, these products aren’t using generative AI, a set of machine learning models that are trained to guesstimate the next set of words or the right image to create based on a prompt…

Stacey on IoT | Internet of Things news and analysisStacey Higginbotham

Is AI art protected?

Some AI Artworks Now Eligible for Copyright

New guidance from the US Copyright Office sets some policies around AI-generated images.

HyperallergicRhea Nayyar

And should you be able to be free from automation?

The Right To Be Free From Automation | NOEMA

A qualified right to freedom from automation can designate what is worth protecting in a technologically dependent human world.

NOEMAZiyaad Bhorat

The critique of Gary Marcus. Especially on the lack of openness of OpenAI with this release.

The Sparks of AGI? Or the End of Science?

Marching into the future with an obstructed view

The Road to AI We Can TrustThe Road to AI We Can Trust

Cybersecurity and data poisoning.

Protecting AI Models from “Data Poisoning”

New ways to thwart backdoor control of deep learning systems

IEEE SpectrumPayal Dhar

And will Web3 be more about AI than community-driven governance?

AI is the real Web 3

Watch now (8 min) | After Web 2.0 will come a Web underpinned by AI.

Exponential View by Azeem AzharAzeem Azhar

Triggering conflicting emotions in tech.

Awe, Anxiety, and AI

Navigating the conflicting emotions of this moment in tech

Chain of ThoughtDan Shipper

Can Microsoft ruin it all?

ChatGPT Opened a New Era in Search. Microsoft Could Ruin It

Startups say Microsoft and its Bing chatbot—not just Google—are stifling competition when it comes to creating better search engines.

WIREDCondé Nast

In other news, regarding autonomous driving, China is taking steps:

Baidu launches first capital-city driverless cab service in Beijing

Baidu has won a permit to provide a driverless ride-hailing service in Beijing, the first of its kind in any capital city anywhere in the world.

RCR Wireless NewsJames Blackman

Also interesting how they found a way to speed up testing: create simulated terrible drivers.

Simulated terrible drivers cut the time and cost of AV testing by a factor of one thousand

From left, Sean Shen, a research engineer at UMTRI, Xintao Yan and Haowei Sun, both civil engineering PhD students, prepare to take out one of the autonomous vehicles for a test run at Mcity on North Campus of the University of Michigan in Ann Arbor on Wednesday, January 18, 2023. Image credit: Bren

University of Michigan NewsJim Lynch, College of Engineering

NVIDIA is catering for robots to make them easier to use.

NVIDIA is making AI easier to use - The Robot Report

NVIDIA announces new features and capabilities that support the acceleration and growth of AI-based solutions for robotics and simulation.

The Robot ReportMike Oitzman

When the Cities of Things research started at TU Delft back in 2017 we were looking at possibilities to clean the polluted air via moving objects. That is now becoming more realistic.

Low-cost device can measure air pollution anywhere

Flatburn is an open-source, mobile pollution detector from the MIT Senseable City Lab intended to let people measure air quality cheaply.

MIT News | Massachusetts Institute of TechnologyPeter Dizikes | MIT News Office

The rumours go that the mixed reality goggle by Apple had an internal release.

Apple shows MR headset to executives ahead of rumored June unveiling | AppleInsider

Apple’s executives have been given a demonstration of the inbound mixed reality headset at Apple Park, a report claims, months before its expected public unveiling.

AppleInsiderMalcolm Owen

How risky is the rapid dependability on FinTech?

Fintech is expanding – and so are the risks

Fintech – the use of technology to deliver financial services – is expanding globally. Millions of people now use financial technology applications and tools on a regular basis and access services. Yet, as SOMO’s new report, Fintech’s red flags, demonstrates,…

SOMOjenny

And don't forget the Climate Crisis…

Listening To The Big Ice Of Antarctica

In Antarctica, there are more than 5 million cubic yards of ice per person on Earth. In it, there are deep questions about us, the planet and the future.

NOEMABruno Giussani

Paper of the week

I was unsure if I should also have a paper here on GPT-4, but ok...

Sparks of Artificial General Intelligence: Early experiments with GPT-4

“In this paper, we report on our investigation of an early version of GPT-4, when it was still in active development by OpenAI. We contend that (this early version of) GPT-4 is part of a new cohort of LLMs (along with ChatGPT and Google's PaLM for example) that exhibit more general intelligence than previous AI models. We discuss the rising capabilities and implications of these models.”

Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., ... & Zhang, Y. (2023). Sparks of Artificial General Intelligence: Early experiments with GPT-4. arXiv preprint arXiv:2303.12712. Chicago

https://doi.org/10.48550/arXiv.2303.12712

See you next week!