Sign in Subscribe

Weeknotes 248; the fundamental nature of AI

By Iskander Smit in weeknotes —

Exploring the AI Barbenheimer moment, new personal software paradigms, xenobots and much more, but not too much on x.

A closed box placed on a busy city square that is shaped through the fundamental laws of AI creating a shape that represents its surroundings - Midjourney

Hi y’all! Happy Barbenheimer…

I am not yet part of this meme, but I might become one, as the movies are both on the watchlist. The Oppenheimer movie is linked to the state of AI a lot by the way. You can wonder if there is a similar Manhattan program now aiming. Still, the goal to gain a competitive advantage on nation level and the potential deconstructive impact might make the comparison not strange.

I have to say that I still believe that AI will not rule or threaten our life directly very soon, but the impact of AI systems and AI tools on how we as humans will play out conflicts and life, in general, might create a shift in agency and dependency. And in that sense, might the control of the biggest AI in this 21st century be comparable to what control of the biggest bomb was back in the 20th century…

The future of AI was also a topic in last Sunday's conversation in Zomergasten on Dutch television with Thomas Hertog, a famous cosmologist that worked with Stephen Hawkins most of his life. The conversation dealt with the relationship of humans with fundamental research, and the concept of artificial physics was mentioned as a way how people will relate more to and learn about the fundamentals of physics. Will AI offer the next Einstein to invent fundamental science? I had to think about another angle that was closeby in the conversation: will there be a moment that the artificial intelligence will behave as if it follows fundamental nature laws. Fundamental AI laws. Do we need to develop a fundamental science to understand AI as we need fundamental science for understanding life and nature?
The whole of the 3-hour conversation was quite interesting, so if you are understanding Dutch, and did not watch yet, check it out...

Another tip for Dutch speaking, as a vacation read; the summer special of De Groene has a guest editor in Marleen Stikker discussing AI for all, “Other Intelligence”.  I did not read all yet, so I cannot judge if the close follower of AI developments will learn much, but it covers a nice range of topics.

I am writing this while we are collectively saying goodbye to a once-loved conversational companion. The blue bird of Twitter… I do not intend to add a lot here to all the memes and commentary on this move. I am not, in principle, against rethinking a service like Twitter that was changing its face for some time, even before Musk took over. But this is so clearly a move to steer away from the real-time conversational engine that made so many connections and shortened distances, levelled playing fields and at the same time derailed from the positive place into the ultimate amplifier of a polarised media space… We miss with X all kinds of references to ‘tweeting’ and derived concepts. X seems more like a foundation for services that need interactions between people. The message protocol will be used for sending functional data mainly probably. But ok, something to return to later, as we have seen more of the future implementation plans…

Last week there was the academic conference on conversational AI: CUI 2023, which has a lot of interesting papers, see the whole list here. I did not have the chance to go through them all, but following some people on X attending the conference like @mlucelupetti, gave some good pointers, like this short provocation on creative capabilities of AI that made me think that it might be more about the creative attitude, being bold while humble, daring with the notion of consequences; that might be helpful for successful creative reframing AI...

Summer events

I could only find a few events to visit in this summer slowness.

Notions from the news

AI safeguards are hot, and seven large AI companies subscribe to that notion through a commitment made to the US government. Is this a good step or just another way to ethics washing?

SAP, the classic ERP giant, is doing the Microsoft trick by investing in several generative AI startups.

SAP invests in generative AI startups Anthropic, Cohere and Aleph Alpha | TechCrunch
SAP, the consulting giant, is making direct investments in three major generative AI startups: Anthropic, Cohere and Aleph Alpha.
TechCrunchKyle Wiggers

Google is creating an AI tool that can write news articles. It feels like something that is already a thing, but alas.

Google is testing an AI tool that can write news articles
Google is testing a tool that uses AI to write news stories and has started pitching it to publications, according to a new report.
TechCrunchAisha Malik

Opening up the models? Meta announced their Llama 2 is open source, however it has limits, both in what is open and who can use it (700 mln users cap)

Llama 2: Open Foundation and Fine-Tuned Chat Models | Meta AI Research
In this work, we develop and release Llama 2, a collection of pretrained and fine-tuned large language models (LLMs) ranging in scale from 7 billion to…
Meta AI Research
Llama copyright drama: Meta stops disclosing what data it uses to train the company’s giant AI models
Meta launched a big new AI model called Llama 2 on Tuesday. The company didn’t disclose what data it used to train the model, unlike the first LLaMA.
InsiderAlistair Barr
Llama 2: why is Meta releasing open-source AI model and are there any risks?
Possible reasons for release of LLM include potential for diluting rivals’ competitive edge
The GuardianDan Milmo

OpenAI is adding new features at the same time: custom instructions to make ChatGPT more specific when used in other tools.

Custom instructions for ChatGPT
We’re rolling out custom instructions to give you more control over how ChatGPT responds. Set your preferences, and ChatGPT will keep them in mind for all future conversations.

No surprise Apple is testing ChatGPT-like tooling. It seems like it is already built for internal use to prevent employees from using the tools of competitors.

It might make sense to have a model that is based on all of the metadata shared in using your devices combined with large language models that would deliver a different take.

Apple is testing a ChatGPT-like AI chatbot | TechCrunch
Apple is developing artificial intelligence tools to challenge Open AI, Google and others, according to a new report.
TechCrunchAisha Malik

The physical part of AI: 3 questions on AI semiconductors. Is there more than Nvidia?

AI Semis – 3 Big Questions
The three big questions for the AI semis market: 1) Is AI additive to the TAM? 2) How will the inference market shape up? 3) Can Nvidia’s dominance be challenged?
Digits to DollarsD/D Advisors

Self-fulfilling generative AI, training the models on synthetic data, lowers costs but might lower the quality.

AI Developers Are Quietly Training AI Using AI-Generated Data
There will come a time when AI models are built on data that was made by other AI models — and for some companies, that time is now.
FuturismNoor Al-Sibai
Study claims ChatGPT is losing capability, but some experts aren’t convinced
Either way, experts think OpenAI should be less opaque about its AI model architecture.
Ars TechnicaBenj Edwards

Some new AI tooling to test, debug and monitor LLM applications

Announcing LangSmith, a unified platform for debugging, testing, evaluating, and monitoring your LLM applications
LangChain exists to make it as easy as possible to develop LLM-powered applications. We started with an open-source Python package when the main blocker for building LLM-powered applications was getting a simple prototype working. We remember seeing Nat Friedman tweet in late 2022 that there was “n…
LangChainLangChain

What is time? It is a perception that can be manipulated now.

Scientists Discover a Way To Manipulate the Brain’s Perception of Time
The Learning Lab at Champalimaud Research has discovered a way to manipulate the brain’s perception of time by controlling neural activity in rats. Their research, which has potential applications in treating diseases like Parkinson’s and Huntington’s, could also influence the fields of robotics and
SciTechDailyMike O’Neill

Another humanoid is being shaped, resembling the Tesla look; GR-1; with a weird choice of representing textual data on the head.

Humanoid Robot Design Iterations Continue: Here’s the GR-1 - Core77
To date, no human being has been intentionally killed by a humanoid robot. But companies around the world are working hard to achieve this inevitable future. Our main question is, what will these robots look like, what will the last thing we see be, as they squeeze the life from
Core77Rain Noe

A different type of robot is “Xenobots”, self-replication living robots. It is a research project in progress now.

Redesigning Evolution, One Robot at a Time
For millions, if not billions, of years, life found a way. Through evolution, natural selection, and cosmic random chance, as some theories purport, simple organic compounds morphed into single-celled organisms which morphed into multi-cellular life, and so on into the sprawling world in which…
The University of Vermont

The future of building;

a fleet of robots is building the largest community of 3D-printed homes outside austin
explore a completed model home within the 3D-printed community outside austin, built by ICON and lennar, and co-designed by BIG.
designboom | architecture & design magazinekat barandy I designboom

The role of feedback loops explored the importance of these reflective learning.

Effort is evidence of broken feedback loops
You’re trying too hard. Stop trying and build feedback loops instead.
SubconsciousGordon Brander

The AI summary of this: “Ethan Mollick argues that there is an opportunity to democratize access to education and use AI's transformative power, but this requires a new vision for education and work systems and a recognition of the genuine and widespread disruption brought by AI.”

On holding back the strange AI tide
There is no way to stop the disruption. We need to channel it instead
One Useful ThingEthan Mollick

I believe in the power of different AIs working as a team with different perspectives together in tackling problems. A paper has proven this now:

AI That Teaches Other AI - USC Viterbi | School of Engineering
A new USC study finds that by sharing knowledge with each other at the same time, AI agents can quickly learn a wider range of tasks, with applications in medicine and beyond.
USC Viterbi | School of EngineeringGreg Hardesty | July 18, 2023

Is there a new type of software beyond the SaaS era, a hyper-personalised app that is created as forks of a master and adjusted using AI coding support? Dan Shipper is exploring this concept.

What Comes After SaaS?
Bespoke apps for everyone—customized by AI
Chain of ThoughtDan Shipper

Should we move back to the time of simplistic Web 1.0 to break the influence of the algorithmic services of the big companies back to the Internet of people? It feels like there are different routes, like personal tools.

What would the internet of people look like now?
Something weird is happening online.
The Verge

Tesla is changing strategies by opening up their Supercharger network to other brands, creating a service proposition (and potential ‘EV dongle town’), and also is licensing their Full Self-Driving technology. And is investing a lot in its supercomputer that does the self-driving AI training.

Tesla announces Full Self-Driving third-party licensing plans and confirms one-time FSD subscription transfer
In a surprising move, Tesla has announced that it plans to license its Full Self-Driving hardware and software to other electric vehicle manufacturers, adding that the company is already in talks with one other OEM to include FSD technology. Tesla will also allow existing FSD subscriptions to be tra…
NotebookcheckJulian van der Merwe
Tesla spending over $1B on Dojo for self-driving AI training
More data means better neural net training, but it also means more cores
The RegisterDan Robinson

General Motors has developed a technology that allows autonomous vehicles to communicate with human drivers more traditionally and can signal their path and intention to other drivers through notifications sent to infotainment screens or smartphones.

Or should autonomous driving be based on nature-inspired AI?

GM Has Found a Way to Make Autonomous Vehicles More Human
Autonomous vehicles are the next big thing in the automotive industry, and it’s certainly only a matter of time until we’ll be surrounded by ca…
autoevolutionBogdan Popa
Bengaluru startup Minus Zero thinks it has cracked autonomous vehicles with its nature-inspired AI
Founded in 2020 by Gagandeep Reehal and Gursimran Kalra, Minus Zero has developed a proprietary artificial intelligence model to help autonomous vehicles navigate better.
YourStoryAparajita Saxena

Are the current heat waves worldwide a signpost of climate change? Robot technology is used to create crash test dummies for heatwaves.

Heat-struck Mediterranean is climate change ‘hot spot’
Struck by near-record temperatures and wildfires during this week’s heatwave, the Mediterranean region is ranked as a climate-change “hot spot” by scientists.
Phys.orgScience X
This Robot Is a Crash Test Dummy For The Deadly Heat of Tomorrow
What happens to the body when a human gets heatstroke? How can we protect ourselves in a warming planet? To answer these burning questions, Arizona researchers have deployed a robot that can breathe, shiver and sweat.
ScienceAlertAFP

Paper for this week

A designerly look to the AI discourse: Risk and Harm, unpacking ideologies in the AI discourse.

The study employs a humanistic HCI methodology, applying narrative theory to HCI-related topics and analyzing the political differences between FLI and DAIR, as they are brought to bear on research on LLMs. Two conceptual lenses, “existential risk” and “ongoing harm,” are applied to reveal differing perspectives on AI’s societal and cultural significance.

(…)

The immediate consequences of longtermism lead to neglecting marginalized communities, exacerbating social inequalities, and undermining efforts to address ongoing harm and human rights violations. We believe that HCI’s engagement with AI and its societal impact should focus on these harms and advocate for the rights and voices of the marginalized. Making design choices that contribute to this focus should become standard practice.

Ferri, G., & Gloerich, I. (2023, July). Risk and Harm: Unpacking Ideologies in the AI Discourse. In Proceedings of the 5th International Conference on Conversational User Interfaces (pp. 1-6).

https://dl.acm.org/doi/10.1145/3571884.3603751

See you all next week!

Some changes in work focus; more on that later. These ‘low vacation times’ are not quiet anyhow, preparing for new events, further Wijkbot iterations and more.

Have a great week!