cho.sh
2,244 notes
Liked on XI made a mistake. When Jack and I decided to organize this $100K Hackathon, my biggest concern was cheating. Unlike a regular hackathon, everyone participates from home. And because the winner is whoever makes the most PROFIT (not revenue), there are so many ways to manipulate the numbers. So I added a $99 entry fee. One thing I’ve learned from building businesses is that people tend to be more committed when they pay for something. I thought it would filter out spam, attract serious builders, and make the competition fairer for everyone. But I completely missed the other side. 3 years ago, I was a living with my apps. Today, I’m financially free. My life changed so much that somewhere along the way, I lost perspective on what $99 means to someone else. For someone in India, someone struggling to pay their bills, or someone trying to make their first dollar online, $99 can be a lot of money. And those are exactly the people I want to help! I’ve spent years sharing what I’ve learned, reviewing random people’s landing pages, and organizing free meetups around the world. Not because there’s money to make, but because I genuinely love seeing people build things and succeed. And somehow, I was so focused on keeping cheaters out that I didn’t realize I was also keeping genuine builders out. That’s my mistake. So I’ve decided to make the $100K Hackathon completely FREE :) Everyone who already paid got a full refund. I’m still worried about cheating. That problem hasn’t disappeared, and Jack and I will have to figure out better ways to verify profits and keep the leaderboard fair. But that’s our problem to solve. Thank you to everyone who shared their thoughts. Sometimes you need people to remind you of things you’ve forgotten. I genuinely hope this hackathon changes someone’s life. ❤️Liked on XIt’s time to bring back Cafe CursorPosted on Xyo @pendev i can't live share edit a .pen file? it's the most needed featurePosted on XI'm coining a word "Agentsplaining" Agents condescendingly assuming the user don't know fuck and repeating the obviousReplied on X@jayhxmo @raycast Hard to explain. It's the small detailsPosted on X@raycast iOS keyboard is insanely good. Might resub Raycast pro just cuz of thisReplied on X@ishuagra02 I saw a theory that F5.2 is a post train of 5.1, but O5.5 is a distillation of F5.5(unreleased new pretrain) and thats why O5.5 > F5.2Liked on X@anaclumos don't know what is actually improved 👀Posted on Xam i the only one underimpressed with 5.2 fableReplied on X@bvvst great minds think alike https://t.co/YRFWhb1mcOLiked on XClaude Fable 5.2 is out and it's the perfect time to run this god-level prompt on your codebase: "I want you to audit our codebase, is anything here unnecessary, overly complicated in our current setup, or based on weak assumptions? Challenge them. - What can be deleted entirely? - What can be simplified now that unnecessary pieces are gone? Then make the changes. Prefer deleting over simplifying, simplifying over optimizing, and optimizing over automating. This product is being used by thousands of customers, do not cut any corners." This prompt is one I run often and leads to the best results.Replied on X@akshayadinesh19 @bot can we please BYOM for botsLiked on XTo be fair, the Chinese models have a much lower refusal rateLiked on XF*CK IT, im creating a library of AI slop indicators. my feed is full of taste larpers. stay tuned https://t.co/5Muv8nQRm3Liked on XJENSEN! THATS NOT WHAT WE MEANT BY CHINESE MODELS!!!!!!!Liked on Xmy parents had to declare bankruptcy when I was a kid, so they had every reason to pressure me towards a more stable career path instead, they were so patient with me, even as i failed startup after startup for years so getting to tell my mom that everything turned out okay was a really special moment :) for those who can't read Korean, her first message roughly translates to: "Really!!! Congratulations my son. So much :') Thank you God!" i know I've been posting about my exit for like a week now, but it really is a big deal for me and my family! i'll share more about how this all happened and some tactical advice later this week, but for now i'm just riding out all the emotions to quote my mom, thank you God!!!!Liked on X@thefarmers_Dog it stands for Egg GzampleLiked on X🦍 Introducing Memorilla: memory as a modality for LLMs Memorilla turns everything a frozen LLM needs to remember into just 16 memory embeddings ⚡ It uses up to 4,212× fewer tokens and is 27× faster than full context inference while more accurate 🎯 Pushes the Pareto frontier over RAG and agentic memory in personalization, long-document and knowledge QA A thread 🧵:Liked on XThere's a huge product opportunity downstream of Tailscale etc too. At @tldraw we've started designing for "high trust networks" and it's been a beautiful thing. 1. Features that would be a nightmare on the internet are okay on my computer and mostly okay on private networks. Websites that execute user-generated scripts, control production systems, allow for users to send desktop notifications, spend tokens, or directly prompt my agent. Crazy stuff that would only make sense if everyone was in a committed polycule. 2. When everyone on the server can be trusted to have _external_ reasons for good behavior (eg keeping their job, or maintaining friends, family, church, mafia org), the software can be designed in a default social way. Most apps default to isolation, low horizontal visibility, highly restricted collaboration. Makes sense if the social contracts are inside of the app, but if you can trust that they live outside of the app then you can get wild. Example being small group chat names as an emergent comedy phenomenon because anyone in the group can change the name. 3. This can work for commercial products, too, not just internal stuff. Security matters more than ever but the way we've seen that manifest is that more companies are more confident in their own solutions than anything we could provide, to the point where owning data and access is a better value proposition than having it managed. Always has been true for big companies but now even small companies can get a good secure network set up pretty easily, and the hooks we need to build are easy enough. tldraw offline got us started down this road (and shout out to @pendev for first introducing me to the idea of desktop + mcp). We've been bringing that product to private networks and it's been exciting to see how much we can get away with. Keep an eye outReplied on X@saradu birds eye viewLiked on XApplications close in 2 days!! If you have a cool bio idea, and $5-10k will go a long way for you, please apply!!! https://t.co/NCvmrNC53cLiked on XWe've been early adopters of Rust since 2018 and a lot of our infra, voice tech, and even some of the current client is already Rust, the holy grail was getting the whole client there. That means Electron goes away, and so does React Native, mobile keeps native UI but shares app code with desktop via Rust. We hope to get it into testing with users soon™. (And it's not Tauri, it's GPUI, so good guess everyone.) Alright, back to staring at the WoW Forever login screen.Reposted on XOh boy. https://t.co/X7SSskGuGvLiked on XRIP .si domain squatters
Proposal of Research 2023-03-28

Proposal of Research 2023-03-28

Warning

Over a year old. May be outdated

Investigating Authority Systems to Mitigate Prompt Injection Attacks in Models

Abstract

Large Language Models (LLMs) are impressively good at answering natural language questions and adapting to new tasks through clever prompting. But the same adaptability might expose LLMs to security risks, such as Prompt Injection (PI) attacks. In this research proposal, we seek to investigate an authority system for LLMs to differentiate instructions from trusted sources and unknown figures. We also aim to assess the feasibility of such an authority system and explore potential mitigation techniques to defend LLMs against PI attacks.

Background

Recent studies have shown that LLMs can be susceptible to PI attacks, which make the model produce malicious content or override the original instructions. These attacks are brutal to mitigate, because following instructions is exactly what an LLM does. As LLMs get built into more apps and systems, including ones that retrieve documents and call APIs, new attack vectors show up and put users' security and privacy at risk.

Research Questions

  • How can we design an authority system for Generative Text AIs to differentiate between trusted and untrusted instructions?
  • Can the authority system efficiently mitigate Prompt Injection attacks across different scenarios?
  • How can mitigation techniques defend LLMs against PI attacks in an authority system?

Methodology

  • Literature review: Survey current PI attack techniques and mitigations to find gaps and room for improvement.
  • Design an authority system: Develop a conceptual model of an authority system for Generative Text AIs, focusing on differentiating between trusted and untrusted instructions.
  • Test the authority system: Implement the designed authority system on a selected LLM and assess its performance in mitigating PI attacks.
  • Evaluate mitigation techniques: Look for mitigation techniques that can plug into the authority system and make the LLM harder to attack.
  • Validation and improvement: Refine the authority system and mitigations based on the results.

Expected Outcomes

  • A clear picture of current PI attacks and existing mitigations.
  • A proposed authority system for Generative Text AIs capable of differentiating between trusted and untrusted instructions.
  • An evaluation of the authority system's effectiveness in mitigating PI attacks in different scenarios.
  • A set of mitigation techniques that make the LLM harder to attack when combined with the authority system.

Significance

This research will help us understand PI attacks and what they can do to LLMs. It will also produce an authority system for Generative Text AIs and mitigations that protect users' security and privacy as LLMs end up in more apps and systems.

Dynamic Font Generation with . Replicating Handwriting Fonts and Their Natural Flow

Abstract

open up new ways to generate fonts, especially handwriting fonts and their dynamic features. Traditional font-generating services have struggled to capture the natural flow of handwriting, so their fonts look less real and less pretty. This study explores the implementation of in creating handwriting fonts that consider dynamic , such as variations in character appearance based on their position within a word.

Background

Dynamic are what make a handwriting font look good and natural. However, current AI font generation methods often overlook these features, focusing primarily on static character styles. With , we can make more realistic, adaptable fonts that look more like real handwriting.

Research Questions

  • How can create handwriting fonts that incorporate dynamic ?
  • What matters most when designing a model that captures the natural flow of handwritten text?
  • How do dynamic font-generation techniques compare to traditional font-generating services regarding quality and versatility?

Methodology

  • Literature review: Survey research on , font generation, and dynamic .
  • Dataset creation: Compile a dataset of handwriting samples that exhibit variations in character appearance based on their position within a word.
  • Model development: Design and train a model that considers dynamic and learns from the dataset of handwriting samples.
  • Evaluation: Assess the quality and versatility of the generated fonts by comparing them to traditional font-generating services and high-quality fonts produced by professional font studios.
  • Case studies: Explore potential applications of dynamic font generation in fields such as graphic design, marketing, and personalized communication.

Expected Outcomes

  • A model capable of creating handwriting fonts that consider dynamic and replicate the natural flow of handwritten text.
  • A clear understanding of what matters when designing a model for dynamic font generation.
  • A comparison of dynamic font generation techniques and traditional font generating services in terms of quality and versatility.
  • Case studies that show where dynamic font generation could be used.

Significance

This research will lead to more realistic and versatile handwriting fonts, using and dynamic . The fonts will follow the natural flow of handwriting more closely and give designers and content creators better-looking options. It could change how fonts get made, and bring further into graphic design and personalized communication.

Better Web Accessibility. Using to Generate Descriptive Alt Text for Images Automatically

Abstract

The lack of descriptive alt text for images on the web poses accessibility challenges for visually impaired users and negatively impacts search engine optimization. This research proposal aims to develop a distributed intelligence system utilizing , such as CLIP or BLIP, and perceptual hashing techniques to automatically generate alt text for images, so more people can use the web. In addition, the project will involve developing wrapping libraries and toolkits for easy integration by developers.

Background

Alt text matters a lot for web accessibility, especially for visually impaired users who rely on screen readers and other assistive technologies. However, many images on the web have empty alt text due to oversight or lack of knowledge by content creators. If we generate descriptive alt text automatically, the web gets easier to use for everyone who relies on it.

Research Questions

  • How can models like CLIP or BLIP effectively generate accurate and descriptive alt text for images on the web?
  • How can perceptual hashing techniques be integrated with models to optimize the system's efficiency and speed?
  • How can we develop user-friendly wrapping libraries and toolkits for easy integration of the proposed system by web developers?

Methodology

  • Literature review: Survey models such as CLIP and BLIP, and perceptual hashing techniques, to see how they could generate alt text.
  • Design a distributed intelligence system: Develop a conceptual model combining and perceptual hashing techniques to generate descriptive alt text for images on the web automatically.
  • Implement the system: Build a proposed distributed intelligence system prototype using selected models and perceptual hashing techniques.
  • Evaluate system performance: Assess the proposed system's accuracy, efficiency, and speed in generating alt text for diverse images.
  • Develop wrapping libraries and toolkits: Create libraries and APIs that web developers can drop in easily.

Expected Outcomes

  • A clear understanding of how models and perceptual hashing can generate alt text for images.
  • A distributed intelligence system that can automatically generate descriptive alt text for images on the web.
  • An evaluation of the proposed system's accuracy, efficiency, and speed performance.
  • Wrapping libraries and toolkits that web developers can integrate easily, so the technology spreads widely.

Significance

This research will contribute to developing a distributed intelligence system capable of automatically generating descriptive alt text for images on the web. That makes the web more accessible for visually impaired users and helps search engine optimization too. And easy libraries and toolkits will get web developers to adopt it, so the web works better for everyone.

Secure, High-Performance Multi-Threading iframes for Web Applications

Abstract

Traditional iframes have performance and security problems, which hold them back as the internet becomes the computer. This research proposal aims to develop , an improved iframe version that runs on Worker Threads, providing secure, high-performance multi-threading capabilities for web applications. Built on Web Workers, , and WebAssembly, could change how web applications are built and deployed, making them more responsive and interactive.

Background

The current iframe technology poses performance and security issues due to its single-threaded design and potential for cross-site scripting attacks. With newer web technologies such as Web Workers, , and WebAssembly, we can create an improved iframe version that addresses these limitations and opens the door to secure, high-performance web applications.

Research Questions

  • How can we design and implement , an improved iframe version that runs on Worker Threads and offers secure, high-performance multi-threading for web applications?
  • How can we use Web Workers, , and WebAssembly to make fast, responsive, and secure?
  • What are the potential use cases and benefits of in the context of modern web applications and the internet-computer era?

Methodology

  • Literature review: Survey existing iframe technology and its limits, and recent work on Web Workers, , and WebAssembly.
  • Design . Develop a conceptual model for , outlining its architecture, components, and communication mechanisms between Worker Threads and the main thread.
  • Implement prototype: Build a prototype of that uses Web Workers, , and WebAssembly to offer secure, high-performance multi-threading for web applications.
  • Evaluate performance: Assess 's performance, responsiveness, and security by comparing it with traditional iframe technology across different web applications.
  • Identify potential use cases: Explore the possible applications and benefits of in modern web development, particularly in the internet-computer era.

Expected Outcomes

  • A clear understanding of the limitations of traditional iframes and the potential of Web Workers, , and WebAssembly in addressing these issues.
  • A secure, high-performance prototype that runs on Worker Threads and offers multi-threading capabilities for web applications.
  • An evaluation of 's performance, responsiveness, and security compared to traditional iframe technology.
  • Identification of potential use cases and benefits of in the context of modern web applications and the internet-computer era.

Significance

This research will produce , an improved iframe that fixes its performance and security limits. With secure, high-performance multi-threading, could change how people build for the web, and make web applications more responsive, interactive, and immersive. That means a better experience for users, and new things to build as the internet becomes the computer.

. Exploring the Implications of Finite Pixel Combinations on Human Creativity and

Abstract

Inspired by "" by Jorge Luis Borges, this research proposal aims to investigate the concept of an "efficiently finite" map containing all possible pixel combinations and its implications on human creativity and . By exploring the idea of a Photo Library of Babel, we will examine the role of AI in creating information and the philosophical aspects of creativity in the context of finite possibilities.

Background

The infinite monkey theorem and "" matter more and more with the rise of technologies, such as ChatGPT. Given infinite monkeys making infinite keystrokes, wouldn't they write all of Shakespeare's work? As AI advances, it raises questions about the future of human creativity, the nature of creation, and the potential for AI to generate all possible information.

Research Questions

  • Can we create an "efficiently finite" map of all possible pixel combinations, forming a Photo Library of Babel?
  • What are the implications of for human creativity and ?
  • How does the finite nature of influence the search for specific information or images, and what does this mean for the proverb "A Needle in a Haystack"?

Methodology

  • Literature review: Survey existing research on "", the infinite monkey theorem, and technologies.
  • Develop a theoretical model: Create a theoretical model of that encompasses all possible pixel combinations and their efficient representation.
  • Analyze implications: Investigate the philosophical and practical implications of for human creativity and , and explore the relationship between finite possibilities and the search for specific information.
  • Case studies: Develop case studies on the potential applications and consequences of in fields such as art, design, and technology.

Expected Outcomes

  • A theoretical model of , representing all possible pixel combinations in an efficiently finite manner.
  • A clear understanding of the implications of on human creativity and .
  • Insights into the relationship between finite possibilities and searching for specific information, challenging the infinite monkey theorem.
  • Case studies that show what could mean in practice.

Significance

This research will contribute to understanding the relationship between human creativity, , and finite possibilities, as illustrated by concept. By examining the implications of a limited map of all possible pixel combinations, we can gain insights into the future of creation and the role of AI in generating information. It could change how we think about creativity, about finding knowledge, and about what AI does to every field it touches.

Backlinks1