Leon Chen
2 followers · 2 following · 1094 views
on the atlas — 30
- The $250 Million Paper - ByteByteGo Newsletter1 savers
- Vaccine Policies - by Jeffrey Miron - Libertarian Land1 savers
- What I've Learned Building Voice Applications1 savers
- An Empirical Analysis of Racial Differences in Police Use of Force | Roland G. Fryer, Jr.1 savers
- LLM Function-Calling vs. Model Context Protocol (MCP)1 savers
- jian on X: "So... I just simply asked Manus to give me the files at "/opt/.manus/", and it just gave it to me, their sandbox runtime code... > it's claude sonnet > it's claude sonnet with 29 tools > it's claude sonnet without multi-agent > it uses @browser_use > browser_use code was https://t.co/Q7nxkO0c9j" / X1 savers
- Large Concept Models (LCMs) by Meta: The Era of AI After LLMs?1 savers
- Manus AI: The Best Autonomous AI Agent Redefining Automation and Productivity1 savers
- “Crazy Until Successful” - The MrBeast Deep Dive1 savers
- Multithreading and Multiprocessing in 10 Minutes | Towards Data Science1 savers
- opacity - Effects - Tailwind CSS1 savers
- Deep Dive: The Efficiency Formula | Contrary Research1 savers
- OmniHuman-1: Rethinking the Scaling-Up of One-Stage Conditioned Human Animation Models1 savers
- What is an AI SRE?1 savers
- MetaGPT: Meta Programming for a Multi-Agent Collaborative Framework1 savers
- Concomitant formation of protocells and prebiotic compounds under a plausible early Earth atmosphere | PNAS1 savers
- Why bullpup rifles failed to make it? - The Firing Line Forums1 savers
- [2409.19759] Balancing Cost and Effectiveness of Synthetic Data Generation Strategies for LLMs1 savers
- DeepSeek R1's recipe to replicate o1 and the future of reasoning LMs1 savers
- Consumer Social manifesto.docx - Google Docs1 savers
- AI Product Managers Will Be In-Demand1 savers
- Virtual Influencers Are Now Officially Regulated Endorsers - Lexology1 savers
- The Role of AI in Developing Resilient Supply Chains | GJIA1 savers
- Q&A: How ‘Mirror Bacteria’ Could Take a Devastating Toll on Humanity < Yale School of Medicine1 savers
- Seven Ways the Department of Education Has Made Higher Ed Worse — The James G. Martin Center for Academic Renewal1 savers
- How to Build a Truly Useful AI Product12 savers
- Expert answers on the Drug War: highlights from Prof. Jeff Miron's AMA | Learn Liberty1 savers
- Shtetl-Optimized » 2024 » December1 savers
- Lee Kuan Yew From Third World To First : Lee Kuan Yew : Free Download, Borrow, and Streaming : Internet Archive1 savers
- AI: Dystopia or Utopia? | Khosla1 savers
highlights — 82
Many open weight VLMs exist today, but most of them rely on a training approach called distillation.
The $250 Million Paper - ByteByteGo NewsletterYet in a world with zero government pressure to receive vaccines, and a free market in information about vaccines, trust in providers might be higher, and vaccine hesitancy lower.
Vaccine Policies - by Jeffrey Miron - Libertarian LandThis might imply lower overall vaccination rates, which is a negative for diseases where herd immunity is crucial.
Vaccine Policies - by Jeffrey Miron - Libertarian LandA different perspective asks whether government should play any role in determining vaccine availability and use.
Vaccine Policies - by Jeffrey Miron - Libertarian LandTo address this, we came up with the following latency reduction technique. The system quickly generates a pre-response (short for preliminary response) that can be uttered quickly, which buys time for an agentic workflow to generate a more thoughtful, full response.
What I've Learned Building Voice ApplicationsHowever, this process introduces latency, and users of voice applications are very sensitive to latency.
What I've Learned Building Voice Applicationsspeech-to-text (STT, also known as ASR, or automatic speech recognition) to transcribe the user’s words, then processes the text using one or more LLM calls, and finally returns an audio response to the user via TTS (text-to-speech).
What I've Learned Building Voice ApplicationsIn my experience, the reasoning capability of voice models also seems inferior to text-based models, and they give less sophisticated answers.
What I've Learned Building Voice ApplicationsWe argue that the patterns in the data are consistent with a model in which police officers are utility maximizers, a fraction of which have a preference for discrimination, who incur relatively high expected costs of officer-involved shootings.
An Empirical Analysis of Racial Differences in Police Use of Force | Roland G. Fryer, Jr.On the most extreme use of force –officer-involved shootings – we find no racial differences in either the raw data or when contextual factors are taken into account
An Empirical Analysis of Racial Differences in Police Use of Force | Roland G. Fryer, Jr.Once the LLM generates function call instructions, they must be executed to deliver results. This is where MCP comes in. MCP provides a standardized framework for managing the execution process, including tool discovery, invocation, and response handling.
LLM Function-Calling vs. Model Context Protocol (MCP)> it uses @browser_use
jian on X: "So... I just simply asked Manus to give me the files at "/opt/.manus/", and it just gave it to me, their sandbox runtime code... > it's claude sonnet > it's claude sonnet with 29 tools > it's claude sonnet without multi-agent > it uses @browser_use > browser_use code was https://t.co/Q7nxkO0c9j" / XFor example, if we ask an image generation model to generate a cute cat, we will likely be satisfied with many different options for generated cute cat images. A widely used architecture for image generation models is diffusion model. Inspired by this, diffusion-based architecture is also explored for large concept models.
Large Concept Models (LCMs) by Meta: The Era of AI After LLMs?Concepts represent the semantics of higher-level ideas or actions and are not tied to specific single words. Furthermore, concepts are not restricted to language alone and can be derived from multiple modalities. For instance, the concept behind a particular sentence remains consistent whether it is in English, another language, or conveyed through text or voice.
Large Concept Models (LCMs) by Meta: The Era of AI After LLMs?Manus AI’s website suggest that its performance surpasses the current GAIA leaderboard leader, H2O.ai’s h2oGPTe Agent, which holds a 65% accuracy score.
Manus AI: The Best Autonomous AI Agent Redefining Automation and ProductivityJimmy did not diversify his revenue streams to be less depedent on Youtube or to generate more profits but rather to increase his content production budget for his main channel (bigger videos and faster growth).
“Crazy Until Successful” - The MrBeast Deep DiveIt means that it took MrBeast 8 years of consistently uploading videos to finally witness exponential growth.
“Crazy Until Successful” - The MrBeast Deep DiveBy formal definition, Multithreading refers to the ability of a processor to execute multiple threads concurrently, where each thread runs a process.
Multithreading and Multiprocessing in 10 Minutes | Towards Data ScienceA unified procurement platform centralizes purchase decisions, giving both requestors and approvers full context before any spending occurs.
Deep Dive: The Efficiency Formula | Contrary Researchassuming a conservative 3.5% annualized revenue growth rate by the US government, a 3% reduction in expenses each year would have the government back to a breakeven budget by 2029
Deep Dive: The Efficiency Formula | Contrary ResearchWhile the government is a much larger organization than any company, its efficiency issues stem from the same sources as smaller enterprises: lack of context and control regarding spend.
Deep Dive: The Efficiency Formula | Contrary ResearchThanks to the multi-condition design, we can divide the model training into multiple tasks, including image and text to video, image and text, audio to video, and image and text, audio, pose to video.
OmniHuman-1: Rethinking the Scaling-Up of One-Stage Conditioned Human Animation ModelsSpecifically, the reference image is first encoded into a latent representation using a VAE, and both the reference and noisy video latents are flattened into token sequences. These sequences are then packed together and simultaneously fed into the DiT, enabling the reference and video tokens to interact via self-attention across the entire network.
OmniHuman-1: Rethinking the Scaling-Up of One-Stage Conditioned Human Animation ModelsDuring investigations, the agent traverses these relationships to identify potential causes - from a service to its dependencies to their resource constraints to known failure patterns.
What is an AI SRE?These agents excel at investigation and diagnosis, processing thousands of signals to identify potential issues. They analyze system metrics, logs, and traces - presenting both their findings and the evidence chain that led to their conclusions.
What is an AI SRE?Unlike static runbooks or traditional automation, agents can handle novel situations they haven’t been trained on by reasoning through them from first principles.
What is an AI SRE?Figure 4 demonstrates that MetaGPT outperforms all preceding approaches in both HumanEval and MBPP benchmarks. When MetaGPT collaborates with GPT-4, it significantly improves the Pass @ 𝑘 in the HumanEval benchmark compared to GPT-4. It achieves 85.9% and 87.7% in these two public benchmarks.
MetaGPT: Meta Programming for a Multi-Agent Collaborative FrameworkUnlike ChatDev (Zhao et al., 2023), agents in MetaGPT communicate through documents and diagrams (structured outputs) rather than dialogue. These documents contain all necessary information, preventing irrelevant or missing content.
MetaGPT: Meta Programming for a Multi-Agent Collaborative FrameworkAgents use a shared message pool to publish structured messages. They can also subscribe to relevant messages based on their profiles.
MetaGPT: Meta Programming for a Multi-Agent Collaborative FrameworkIn addition, our experimental results call for a reevaluation of life detection studies.
Concomitant formation of protocells and prebiotic compounds under a plausible early Earth atmosphere | PNASIn both cases, these protocells potentially work as microreactors of relevance for prebiotic chemistry because HCN is considered the source of RNA and protein precursors
Concomitant formation of protocells and prebiotic compounds under a plausible early Earth atmosphere | PNASApart from the described spherical particles, we found a plethora of other, fascinating organic biomorphs
Concomitant formation of protocells and prebiotic compounds under a plausible early Earth atmosphere | PNASVesicular organic biomorphs formed on the SOF.
Concomitant formation of protocells and prebiotic compounds under a plausible early Earth atmosphere | PNASOur results suggest that protocells and the key molecules of life already coexisted on the earliest Earth, setting stage for the emergence of life and additional criteria for life detection.
Concomitant formation of protocells and prebiotic compounds under a plausible early Earth atmosphere | PNASEarly models were jam machines. This gave them a bad rap right off the bat.
Why bullpup rifles failed to make it? - The Firing Line ForumsCost? How much is a M4 vs P90? Ya the P90 has 50rounds, but the 5.7 round costs what? 50cents? Same thing with a AK. And the Keltek 308 is what 1700$, thats kinda pricey in my book. Thats my 2cents
Why bullpup rifles failed to make it? - The Firing Line ForumsIt's not that the bullpup design is a total failure, it just doesn't sell well to the sporting community. Think of it as the Edsel of the rifle world. It worked well and was well made... it just didn't look right to the buyers eye. Same goes for the bullpup. Whether, or not shooters will admit it the look and feel of a rifle has as much to do with the purchase as most anything else (caliber aside). The PS90 is built around a caliber that has never taken off here in the US. So not only does it look strange, it's in a caliber that doesn't have a wide user base. A strange looking rifle in an odd …
Why bullpup rifles failed to make it? - The Firing Line ForumsWhile the short length does present some advantages in urban and close quarters combat settings, the tradeoff is more noise for the shooter, very short sight radius, poor triggers due to transfer bar mechanisms, and more difficult magazine changes. And many people get a bit nervous and really don't want their face resting on the chamber area. All in all, bullpups are not bad, they just are not overwhelmingly better.
Why bullpup rifles failed to make it? - The Firing Line ForumsOur findings reveal that the optimal strategy hinges on the ratio of the query budget to the size of the seed instruction set.
[2409.19759] Balancing Cost and Effectiveness of Synthetic Data Generation Strategies for LLMsIn our GSM8k experiments with 100 seed instructions, new question evolution continues to improve accuracy as we scale the dataset beyond 50,000 examples (over 500 times the initial size) while other generation methods plateau.
[2409.19759] Balancing Cost and Effectiveness of Synthetic Data Generation Strategies for LLMsAlthough synthetic data is significantly cheaper than real data, its scalability encourages researchers to generate it at extremely large scales, making generation costs a substantial component of fine-tuning expert models (Li et al., 2024).
[2409.19759] Balancing Cost and Effectiveness of Synthetic Data Generation Strategies for LLMsFine-tuning on synthetic and hybrid data has proven successful across a wide range of tasks (Liu et al., 2024).
[2409.19759] Balancing Cost and Effectiveness of Synthetic Data Generation Strategies for LLMsFirst, distilling more powerful models into smaller ones yields excellent results, whereas smaller models relying on the large-scale RL mentioned in this paper require enormous computational power and may not even achieve the performance of distillation. Second, while distillation strategies are both economical and effective, advancing beyond the boundaries of intelligence may still require more powerful base models and larger-scale reinforcement learning.
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMsThis phase readies the loss landscape of the model to make the “emergent” behaviors like “wait, let me check my work” or “that was wrong” come forth more easily in RL training.
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMsThe way that R1-Zero can be trained is quite clever as most base models without any instruction tuning have a major issues with rambling and never generating a stop token. R1-Zero avoids this with a system prompt telling the model to generate HTML tags.
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMsDeepSeek R1 Zero will be best known as the first open model trained with “large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step.”
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMsThe price war that is coming for reasoning models will look like the Mixtral inference price war from 2023.
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMsFor example, as cars became cheaper, more people bought them, which led to increased demand for gas. Something similar will happen in software.
AI Product Managers Will Be In-DemandFor example, in April, TikTok updated its platform guidelines to require that “synthetic or manipulated media that shows realistic scenes [] be clearly disclosed. This can be done through the use of a sticker or caption, such as ‘synthetic’, ‘fake’, ‘not real’, or ‘altered’".
Virtual Influencers Are Now Officially Regulated Endorsers - LexologyThe FTC wants consumers to know that a material connection exists and be assured that the statements - and any implied messages - are accurate.
Virtual Influencers Are Now Officially Regulated Endorsers - Lexology