A list of things I found interesting on the web in January.
If there exist AI systems that can perform 8 of the 10 tasks below by the end of 2027, as determined by our panel of judges, Gary will donate $2,000 to a charity of Miles’ choice; if AI can do fewer than 8, Miles will donate $20,000 to a charity of Gary’s choice
The bet here is for 8/10, but 1/10 would be hugely disruptive already, and getting half the performance on many of these tasks would equally be hugely disruptive. And this bet is for the next two years.
found on 2025–01–01
A lot has happened in the world of Large Language Models over the course of 2024. Here’s a review of things we figured out about the field in the past twelve months
A must read for a review of what happened with LLMs in 2024.
found on 2025–01–01
open-webui/open-webui: User-friendly AI Interface (Supports Ollama, OpenAI API, …) by Open Webui team
the best open-source AI user interface.
It’s interesting to see the evoluiton of interfaces for interacting with LLMs, this is a local running app that can work offline, or interact with models for you. We are still in ham radio territory here, but a lot is going on.
found on 2025–01–01
Shockinly fast code generation. I don’t know how good this is, but this is faster than the speed of thought. This is too fast.
found on 2025–01–01
Enter Croissant, an emerging community standard, that provides machine-readable metadata for ML datasets, enhancing their accessibility, discoverability, and reproducibility while complementing their data cards or other documentation formats. Supported by major platforms like those mentioned above, and championed by the ODI, Croissant is set to play a pivotal role in ensuring robust data management and accountability in the evolving landscape of data-centric AI
Useful if you are working with LLM training data.
found on 2025–01–04
‘It is impossible to outrun them’: how drones transformed war in Ukraine | Drones (military) | The Guardian by Dan Sabbagh
This is the dystopian future - here already.
found on 2025–01–04
‘What have we here?’: how asking yourself a simple question can transform the way you think about your life | Life and style | The Guardian by Oliver Burkemann
Here are the specific questions and example questions mentioned in the article:
General Questions: 1. “What have we here?” • Encourages engagement with the reality in front of you rather than fantasizing about ideal scenarios. • Example: Instead of fantasizing about a perfect fitness routine, ask what small, realistic action you could take today, like a brisk walk. 2. “Is the path I’m on, or the path I’m proposing to take, one that enlarges or diminishes me?” • Helps evaluate whether a choice promotes personal growth or avoidance. 3. “What would you do if money were no object?” • A thought experiment to explore passions and priorities without financial constraints. 4. “What would you do if you weren’t afraid?” • Encourages setting aside fear temporarily to identify deeper desires or aspirations.
Practical Applications: 5. “How much time am I truly willing to give this each day?” • Example: When struggling with anxiety or procrastination, find a realistic starting point, even as little as 15 minutes.
Philosophical Questions: 6. Rainer Maria Rilke’s quote: “Be patient toward all that is unsolved in your heart and try to love the questions themselves like locked rooms and like books that are written in a very foreign tongue.” • Encourages living with unanswered questions and letting life unfold gradually.
These questions emphasize self-reflection, practicality, and embracing uncertainty, providing a framework for personal growth based on your unique context.
Questions for a better life.
found on 2025–01–04
The goal of this system is to help make reviews more constructive and actionable for authors. The review feedback agent will provide suggestions on three potential categories of issues in reviews.
An important initiative. I am very interested in seeing how this pans out.
found on 2025–01–04
Eight years ago, not even $1 million went into startups focused exclusively on humanoids. Today, the total amount of private capital for the industry has climbed over $1.4 billion.
Will this be the robotaxi or the loom - the former an advance we wait for a long time - the latter something that massively displaces labour? I don’t think humanoid robots are going to make a difference in wet lab work as the heterogeneity of task is too high and the main of time doing repetitive tasks is fairly low, but if we seem them making headway in intricate tasks such as embroidery then we will know that a big change is coming.
found on 2025–01–05
you can see who parked outside the intelligence agencies, where they live, and when they visited the brothel.
Just an example of a huge data breach from Volkswagen.
found on 2025–01–07
Computation is a kind of universal solvent. In studying computation, you are not simply thinking about the nature of for-loops, or data structures, or databases. By taking the nature of computation seriously, you are able to interrogate the nature of language and how humans think, as well as the limits of mathematics.
Amazing resources around the history and culture of code.
found on 2025–01–08
“Twelfth Night Till Candlemas” – the story of a forty-year book-quest and of its remarkable ending – The Law and Policy Blog by David Allen Green
“Twelfth Night Till Candlemas” – the story of a forty-year book-quest and of its remarkable ending – The Law and Policy Blog by David Allen Green
we should be wary of the mischievous demons of our own age – that is ChatGPT and Artificial Intelligence – and to renew our trust in the Noble Professions who hold the old knowledge and skills: librarians, archivists, cataloguers, and indexers.
A delightful account of faulty memory, going down rabbit holes, and coming out the other side, whilst avoiding the perils of fabricated recall.
found on 2025–01–10
Tacit knowledge is extremely valuable. Unfortunately, developing tacit knowledge is usually bottlenecked by apprentice-master relationships. Tacit Knowledge Videos could widen this bottleneck
This is an amazing resource, and if you stop for a moment to reflect on the scale of information that is on YouTube alone, it is beyond lifetimes of information. I’d love to look at any of the videos on this list but I simply don’t have that time right now. LLMs trained on this will be very helpful when I am at the point where I may need to apply any of this knowledge.
found on 2025–01–11
If Meta and X were confident in heading off the regulatory impositions of the EU, Brazil and elsewhere, they would not need to swing behind Trump and the new administration. The fact that they are doing so openly and unapologetically — indeed shamelessly — means they know they have a challenge, and one which they may not meet
A good take on the strategic pivot of Facebook, and why ultimately the tech giants are weak in the face of regulation.
found on 2025–01–12
This is a great overview of the recent rise of decentralised platforms, how users have to balance flight vs raising their voices, a history of working the refs, and a good perspective on the trade offs between central and decentralised platforms
found on 2025–01–12
The essential truth of every social network is that the product is content moderation, and everyone hates the people who decide how content moderation works.
This is a great read on the nature of social media platforms, but two years on from when it was written we now have a much clearer view of the motivations of Elon musk and his massive disregard for any of the concerns laid out here.
found on 2025–01–12
Croissant 🥐 is a high-level format for machine learning datasets that combines metadata, resource file descriptions, data structure, and default ML semantics into a single file; it works with existing datasets to make them easier to find, use, and support with tools.
Useful for working in the open with ML data, but I’m not aware if this being mandated anywhere nor adopted yet by places such as huggingface.
found on 2025–01–16
Tooling to help integrate ML automation workflows into schematron workflows.
found on 2025–01–16
Science of Science: Understanding the Foundations and Limits of Science from an Interdisciplinary Perspective | Oxford Academic by Alexander Krauss
How do we drive new knowledge and science? What are their present boundaries? And how can we improve science? We still do not understand these essential questions about science well, even though science is at the foundation of modern society. The field of science of science can provide answers to these foundational questions.
An open access work on the science of science. Some reading for me for the year!
found on 2025–01–16
Critical Ignoring as a Core Competence for Digital Citizens - Anastasia Kozyreva, Sam Wineburg, Stephan Lewandowsky, Ralph Hertwig, 2023 by Kozyreva eat al, via kottke
I definitely need this.
found on 2025–01–23
Claude can now provide detailed references to the exact sentences and passages it uses to generate responses,
More innovation in how we interact with models!
found on 2025–01–23
So this is all pretty depressing, then? Actually, no. I think that DeepSeek has provided a massive gift to nearly everyone.
I’m not sure about the overall conclusions, but what this should highlight is the importance of thinking about product strategy in light of readily available cheap flawed models.
found on 2025–01–28