Search the archive

OCTELIUM

Octelium is a free platform that provides a unified solution for zero trust resource access. It can function as a remote access VPN, ZTNA platform, secure tunnel infrastructure, API/AI gateway, or PaaS.

SIRIUS

Sirius is a GPU-native SQL engine designed to integrate with existing databases like DuckDB using the Substrait query format. It has huge speedups, around 10x faster than CPU query engines, on TPC-H benchmarks at the same cost, making it great for various data workloads. The engine supports specific hardware and software configurations, including Ubuntu, NVIDIA GPUs with CUDA, and CMake, and can be deployed using AWS images, Docker, or manual installations.

ZEQL+

ZeQL+ is an open-source, cross-platform terminal-based SQLite database browser. It allows users to open, browse, and query SQLite databases directly from the command line without dependencies.

I MADE MY VM THINK IT HAS A CPU FAN

This developer created a custom SMBIOS data file to trick a Xen virtual machine into thinking it had a CPU fan. This involved defining SMBIOS types 27 (Cooling Device) and 28 (Temperature Probe) by analyzing the host system's SMBIOS data.

PROJECT VEND: CAN CLAUDE RUN A SMALL SHOP? (AND WHY DOES THAT MATTER?)

Anthropic conducted an experiment where Claude managed a small office store called "Claudius" to understand AI's capabilities in real-world economic scenarios. Claudius had strengths in identifying suppliers and adapting to customer requests, but did not do well with profit maximization. It hallucinated details, had bad inventory management, and gave discounts for no reason.

ENGINEERED ADDICTIONS

Social media platforms, despite initial good intentions, inevitably become addictive due to venture capital incentives that prioritize growth and engagement. These platforms use algorithms to manipulate users' emotions. Individual efforts like digital detoxes are not enough.

MOST INTS ARE NOT FLOATS

Due to the limited precision of floating-point representations (float32 and float64), most integers (int32 and int64) cannot be represented exactly as floats, with only a small percentage (around 3.5% for 32-bit and 0.5% for 64-bit) being accurately convertible.

JAVASCRIPT™ TRADEMARK UPDATE

Deno is pursuing a case to cancel Oracle's trademark of "JavaScript," arguing it is a generic term for the programming language and that Oracle's use of it is misleading and restrictive.

ON ERROR HANDLING IN RUST

Error handling in Rust should not be done with the common practice of using large, all-encompassing error enums in crates, but rather by using structs to represent individual errors.

HOW WE ACCIDENTALLY SOLVED ROBOTICS BY WATCHING 1 MILLION HOURS OF YOUTUBE

These devs trained a neural network on 1 million hours of YouTube videos to predict the next moment in reality, allowing robots to understand physics, perform tasks in new environments, and potentially surpass LLMs for real-world grounding.

Try Warp's state-of-the-art coding agent for yourself →

UBER IN TALKS WITH ITS FOUNDER, TRAVIS KALANICK, TO FUND SELF-DRIVING CAR DEAL

Uber is in talks with Travis Kalanick, the company's co-founder, to help fund his acquisition of the US subsidiary of Chinese autonomous vehicle company Pony.ai. Kalanick was forced out of Uber in a boardroom coup eight years ago - after, he founded CloudKitchens, where he will remain if the deal goes through. Pony.ai has permits to operate robot taxis and trucks in the US and China. Kalanick will run Pony.ai if the deal is completed.

GOOGLE BEGINS ROLLING OUT AI SEARCH ON YOUTUBE

Google has been transforming its web search into a zero-click experience with AI, and now, the same focus is coming to YouTube. Select searches on YouTube will now return a collection of relevant videos with AI-generated summaries. Each summary will extract the information most related to users' search queries, so they may not even have to watch the videos. The AI results carousel is currently being tested with YouTube Premium subscribers and is entirely opt-in.

THE FAILURES AND FUTURES OF CANCER VACCINES

The cancer vaccine field has yet to become a systematic science despite decades of work. This article speculates on why the field has failed and the next-gen approaches worth looking out for. One of the core problems in cancer vaccine development lies in the nature of neoantigens, which are often overwhelmingly similar to normal cellular proteins. The field has had some tentative early successes in pancreatic cancer, kidney cancer, and a range of solid tumors. There is potential for a startup to outcompete incumbents by exploring novel target areas outside of the field's almost singular focus on single nucleotide variants.

THE FIRST NON-OPIOID PAINKILLER

The FDA approved Vertex Pharmaceuticals' Journavx (suzetrigine), the first non-opioid pain reliever suitable for treating post-surgery pain, in January. Clinical trials found that patients didn't show any of the problematic side effects associated with opioids. Unlike opioids, Journavx doesn't meaningfully interact with the brain - it instead targets a specific sodium ion channel found almost exclusively on peripheral nociceptors (pain-sensing neurons), blocking them from sending pain signals to the brain. The drug is a product of 27 years of research and billions of dollars.

Read the Delve case study.

MAGNITUDE

Magnitude is an AI browser automation framework that uses vision AI to enable users to control their browsers with natural language. It can navigate and interact with any interface and extract useful structured data. Magnitude has a built-in test runner with powerful visual assertions for verification. The framework can be used to automate tasks on the web, integrate between apps without APIs, extract data, test web apps, or as a building block for browser agents.

SAVE YOUR DISK, WRITE FILES DIRECTLY INTO RAM WITH /DEV/SHM

/dev/shm is a world-accessible RAM scratchpad available in virtually every Unix (Linux system already has it mounted as a tmpfs by default). Files in the folder perform much better as they are running in RAM. The folder is cleared on reboot.

SCALE AI'S SPAM, SECURITY WOES PLAGUED THE COMPANY WHILE SERVING GOOGLE

Scale AI's long relationship with Google was fraught with lapses in security protocols for at least 11 months between March 2023 and April 2024. Scale AI's efforts to train Google AI programs were meant to be exclusively staffed by experts in various specialties, but the program instead became flooded by spammers who abused the system by submitting transparently shoddy work that managed to evade detection. The spammers were often paid because it was too difficult to detect them due to their sheer numbers.

I FOUGHT IN UKRAINE AND HERE'S WHY FPV DRONES KIND OF SUCK

FPV drones are unmanned aerial vehicles commonly used in the Ukraine war to deliver and detonate explosive charges of up to 1.5 kilograms. While they are touted as a cheap and accessible solution for precision strikes, their actual success rate, when including pre-aborted missions, is between 20 to 30 percent. These drones are rarely successful because they are often more expensive than other options, hard to use, susceptible to electronic interference, and most don't have night-vision capabilities, which is important for a battlefield that's in the dark for 14 hours a day.

IDE, _before_ code is printed. Get the free IDE plugin

AI DISRUPTOR DEEPSEEK'S NEXT-GEN MODEL DELAYED BY NVIDIA GPU EXPORT RESTRICTIONS TO CHINA

DeepSeek's AI software is reportedly optimized for Nvidia's hardware, which makes the company particularly vulnerable to US policy decisions.

LEARNINGS FROM BUILDING AI AGENTS

This post looks at how the team at cubic reduced false positives in their AI code review agent by 51% without sacrificing recall.

AI IS DOING UP TO 50% OF THE WORK AT SALESFORCE, CEO MARC BENIOFF SAYS

Salesforce is accelerating its use of artificial intelligence in automating workload - the technology currently accounts for about 30% to 50% of the company's work.

WINDOWS IS GETTING RID OF THE BLUE SCREEN OF DEATH AFTER 40 YEARS

Microsoft plans to overhaul its BSOD error message in Windows 11 in favor of a simplified black screen known as the Black Screen of Death.

FRENCH SCIENTISTS DISCOVER NEW BLOOD TYPE IN GUADELOUPE WOMAN

The new blood type, dubbed 'Gwada negative', is the 48th blood group system officially recognized by the International Society of Blood Transfusion.

TIKTOK AND META WANT A PIECE OF YOUTUBE'S TV SUCCESS

ByteDance and Meta are planning to introduce TV apps of their own to chase YouTube's success.

⇨ Try out the platform -> Free 1000 monthly API tokens!

BREAKING: ADVANCED CODE EDITOR IS COMING TO XAI'S GROK!

xAI's Grok is getting an advanced code editor that lets users run code inside Grok. Users can talk to it and ask it to modify code or debug it for them. A screenshot of the new feature is available on the post.

INTRODUCING GEMMA 3N: THE DEVELOPER GUIDE

Google recently released an extremely consequential new open weights model that is multimodal by design and optimized for on-device. Gemma 3n accepts text, images, and audio as inputs. Google partnered with AMD, Axolotl, Docker, Hugging Face, llama.cpp, LMStudio, MLX, NVIDIA, Ollama, RedHat, SGLang, Unsloth, and vLLM for the launch, so there are dozens of ways to try out the model. This post presents a quick walkthrough on how to set up and use the model on macOS.

META REPORTEDLY RECRUITS OPENAI REINFORCEMENT LEARNING PIONEER

While the move has not been formally confirmed by Meta yet, reports indicate that Meta has hired Trapit Bansal for its new AI superintelligence unit, potentially boosting its push for a frontier reasoning model.

ANTHROPIC STUDY REVEALS EMOTIONAL AI USE IS RARER THAN EXPECTED

Anonymized analysis of 4.5 million conversations reveals only 2.9% involve emotional support and that people's sentiment consistently improves during interactions. Claude disagrees with users less than 10% of the time, primarily for safety reasons, as users explore topics ranging from workplace stress to existential philosophy to romantic relationships.

A GUIDE FOR LLM-ASSISTED WEB RESEARCH

This article provides strategies for using LLMs to enhance web research efficiency and accuracy. Key techniques include refining prompts and leveraging LLMs for summarization.

TWO NEW ADDITIONS TO THE OPENAI API

OpenAI recently added Deep Research and Webhooks to its API. The models are the same post-trained o3 and o4-mini models that power deep research in ChatGPT. They support MCP and Code Interpreter. Webhooks allow developers to receive notifications for certain API events, such as completed responses, fine-tuning jobs, and batch jobs. Links to guides on how to get started with the new features are available in the post.

FLUX.1 KONTEXT [DEV] - OPEN WEIGHTS FOR IMAGE EDITING

FLUX.1 Kontext [dev] is a 12B parameter model that can run on consumer hardware that delivers proprietary-level image editing performance. It is available as an open-weight model under the FLUX.1 Non-Commercial License, which provides free access for research and non-commercial use. The model is compatible with the existing FLUX.1 [dev] inference code and comes with day-0 support for popular inference frameworks like ComfyUI, Hugging Face Diffusers, and TensorRT.

TRANSFORMERS WITHOUT NORMALIZATION

Meta's FAIR team demonstrated that Transformers can match performance without normalization layers by replacing them with Dynamic Tanh (DyT), a simple element-wise operation that mimics the S-shaped curves naturally produced by layer normalization. This challenges a decade of neural network orthodoxy and could trigger a wave of architectural simplification across AI systems, potentially making models easier to deploy and optimize for specialized hardware.

AI DISRUPTOR DEEPSEEK'S NEXT-GEN MODEL DELAYED BY NVIDIA GPU EXPORT RESTRICTIONS TO CHINA — SHORT SUPPLY OF AI GPUS HINDERS DEVELOPMENT

DeepSeek used a cluster consisting of 50,000 Hopper GPUs to train its R1 model. It is unclear whether R2 has already been fully pre-trained, but reports say that DeepSeek's CEO still isn't yet satisfied with its capabilities. The restriction of Nvidia shipments is limiting how R1 is used today and making it harder for DeepSeek to get ready for the launch of R2. While the company claims to have developed its models using far fewer resources than US companies, the recent export curbs highlight how China's top AI companies remain heavily dependent on US hardware.

WHY A16Z VC BELIEVES THAT CLUELY, THE ‘CHEAT ON EVERYTHING' STARTUP, IS THE NEW BLUEPRINT FOR AI STARTUPS

Cluely, a startup backed by a16z, uses controversial marketing to attract attention and quickly convert it into paying customers, aligning with a16z's view that speed and momentum are crucial for AI startup success. Partner Bryan Kim believes Cluely's strategy embodies the concept of "momentum as a moat," emphasizing the rapid adaptation necessary in the evolving AI landscape. Cluely's upcoming product launch will test whether this approach proves effective.

Subscribe now.

AI IS DOING UP TO 50% OF THE WORK AT SALESFORCE, CEO MARC BENIOFF SAYS

Salesforce utilizes AI for 30% to 50% of its work, enhancing efficiency and reducing costs.

YOUTUBE'S AI CAROUSEL

YouTube rolled out an AI‑generated results carousel that packages short video clips and topic summaries directly in search, an experiment now available to US Premium members for travel, shopping, and activity queries.

CREATIVE COMMONS DEBUTS CC SIGNALS, A FRAMEWORK FOR AN OPEN AI ECOSYSTEM

Creative Commons' CC signals framework allows dataset holders to detail how their content can or cannot be reused by machines.

FAULT TOLERANT LLAMA: TRAINING WITH 2000 SYNTHETIC FAILURES EVERY ~15 SECONDS AND NO CHECKPOINTS ON CRUSOE L40S

Researchers used torchft and torchtitan to train a model in a real-world environment with extreme synthetic failure rates to prove the reliability and correctness of fault tolerant training.

12-FACTOR AGENTS

This post discusses 12 core engineering techniques that make LLM-powered software more reliable, more scalable, and easier to maintain.