Ideas Worth Exploring: 2025-03-12
- Charles Ray
- Mar 13
- 4 min read
GitHub Repo: FilePizza
FilePizza is Peer-to-peer file transfers in your browser.
Using WebRTC, FilePizza eliminates the initial upload step required by other web-based file sharing services. Because data is never stored in an intermediary server, the transfer is fast, private, and secure.
A hosted instance of FilePizza is available at file.pizza.
Some of the newest features:
A new UI with dark mode support, now built on modern browser technologies.
Works on most mobile browsers, including Mobile Safari.
Transfers are now directly from the uploader to the downloader's browser (WebRTC without WebTorrent) with faster handshakes.
Uploaders can monitor the progress of the transfer and stop it if they want.
Better security and safety measures with password protection and reporting.
Support for uploading multiple files at once, which downloaders receive as a zip file.
Streaming downloads with a Service Worker.
Out-of-process storage of server state using Redis.
LESSWRONG: The Most Forbidden Technique
The Most Forbidden Technique is training an AI using interpretability techniques.
An AI produces a final output [X] via some method [M]. You can analyze [M] using technique [T], to learn what the AI is up to. You could train on that. Never do that.
You train on [X]. Only [X]. Never [M], never [T].
Why? Because [T] is how you figure out when the model is misbehaving.
If you train on [T], you are training the AI to obfuscate its thinking, and defeat [T]. You will rapidly lose your ability to know what is going on, in exactly the ways you most need to know what is going on.
Those bits of optimization pressure from [T] are precious. Use them wisely.
Ideas: The Cultural Divide between Mathematics and AI
Ideas: The Cultural Divide between Mathematics and AI is an excellent deep dive into the differences in cultures, values, and approaches between traditional mathematics researchers and those working on AI in industry. These divisions were observed at the 2025 Joint Mathematics Meeting, where AI-related topics comprised approximately 15% of the presentations, a significant increase from just five years ago.
Mathematicians place value on understanding, elegance, openness, transparency, and human collaboration, while industry researchers must prioritize delivering valuable products or capabilities for their organizations. Concerns were expressed about various aspects of AI development, including potential military applications, lack of transparency in research, enormous energy consumption, and the growing elitism as research becomes concentrated in well-funded private labs.
The tension between openness and secrecy was particularly evident in discussions about collaboration with industry. Mathematicians emphasize that mathematics is a communal effort, yet there's increasing secrecy in AI research labs, which clashes with mathematical traditions.
Understanding what mathematics truly is helps illuminate the cultural divide. Mathematics isn't primarily about finding proofs; it's about building understanding. This quest for deep understanding explains why black-box proofs are rarely considered satisfying and why mathematicians often prioritize elegance over efficiency.
The human element in mathematics is crucial, with mentorship playing an essential role in mathematical development. The process of discovering mathematical truths is deeply human, relying on intuition, creativity, and collaboration. Attribution in mathematics highlights this human dimension by naming results after people rather than describing them functionally.
Mathematics requires extreme patience, with problem-solving often involving long periods of unconscious processing punctuated by moments of insight. The precision and density of mathematical language can come as a shock to those from other disciplines, requiring careful unpacking and strategic thinking.
AI's potential contributions to mathematics were discussed at the meeting, with interest primarily in processing and organizing existing knowledge rather than creating new mathematics. While AI might be valuable for accelerating textbook writing or automating routine calculations, many mathematicians are skeptical about AI creating significant open problems solutions without providing deeper understanding.
The debate between formal methods and natural language approaches reflects another tension in this field. Bridging the divide between mathematics and AI requires mutual respect and understanding, with potential collaboration focusing on literature management, theorem verification, refactoring proofs, teaching, accessibility, and counterexample generation. Maintaining mathematics' human, open, and understanding-centered culture while embracing technological advances is essential for a fruitful collaboration between mathematics and AI. To bridge this divide, it is important to establish frameworks for collaboration that respect mathematical values such as open research sharing, focus on explanation rather than just results, and maintaining human oversight and interpretation.
In conclusion, the cultural divide between mathematics and AI presents an opportunity for mutual learning and growth. Both communities bring valuable perspectives, with mathematics offering deep traditions of rigor, patience, and beauty, while AI research brings energy, resources, and new computational approaches. Bridging this divide requires mutual respect and understanding.
Mathematician Ralph Furman, who attended the 2025 Joint Mathematics Meeting, emphasized that AI won't "solve" mathematics but may help mathematicians explore mathematical landscapes more efficiently, connect disparate areas of knowledge, and perhaps even suggest new directions for human creativity and insight. As technology continues to evolve, it is essential that mathematics remains at the forefront of innovation while preserving its rich intellectual tradition.
GitHub Repo: FIR (Fast Incident Response)
FIR (Fast Incident Response) is an cybersecurity incident management platform designed with agility and speed in mind. It allows for easy creation, tracking, and reporting of cybersecurity incidents.
FIR is for anyone needing to track cybersecurity incidents (CSIRTs, CERTs, SOCs, etc.). It was tailored to suit our needs and our team's habits, but we put a great deal of effort into making it as generic as possible before releasing it so that other teams around the world may also use it and customize it as they see fit.
Comments