Strangers chip in to fund one ambitious instruction — an AI agent carries it out milestone by milestone, with every credit on a public ledger. Funding targets are set by the AI planner (projects total at least $100); backers chip in any amount from $0.25. Game where the bad guys are a few companies who steal all knowledge and lock it behind paywalls. awaiting funding
Homebrew 6.0.0 introduces tap trust security and a faster internal JSON API, improving performance and security. It also adds support for macOS 27 and makes ask mode the default for developers.
Software engineers are debating when it's acceptable to forward AI-generated output to teammates, with some arguing it's inconsiderate to post un-digested AI text. Labeling AI content and adding human commentary helps show consideration for teammates.
Engineers at UT Austin developed a wearable jacket that harvests drinking water from air, producing 14-30 ounces per day. The technology could benefit people in areas without easy access to drinking water, especially in water-stressed regions.
Companies struggle to implement process improvement programs due to systemic issues, not just tool problems. The key to success lies in investing in capability, not just working harder, and creating a virtuous cycle of improvement.
Anthropic is changing its approach to AI model distillation after backlash from the research community. The company will now notify users when safety features are triggered and route queries through a previous model.
If you're into low-dopamine things for your phone, check out Leash, a browser replacement I made. AKRONYM Unthinkable
Canadians are concerned about Bill C-22's broad metadata collection and potential weakening of encryption, which raises serious privacy concerns. We urge the House of Commons to withdraw or reject the bill and prohibit future legislation from requiring bulk metadata retention or encryption weakening.
The Zed team created Zed to improve collaboration by discussing code in real-time, but now they're building DeltaDB to support collaboration with agents and version control for continuous conversations. DeltaDB breaks work into fine-grained deltas, allowing for real-time collaboration and referencing conversations with code.
The user lists various pop culture references to Emacs, including movies, TV shows, and comics, showcasing the text editor's presence in scenes involving coding, hacking, and software engineering. These references include films like The Social Network, Tron: Legacy, and The Internship, as well as TV shows like Silicon Valley and anime series like Aldnoah.Zero.
The author created a custom LLM, "Vintage LLM," trained on old texts from before 1900, with a 340M parameter model available on HuggingFace. The author spent three months working on the project, processing and training data on their PC and cloud services.
These exercises improve musical ability by developing an intuitive understanding of music through daily practice of identifying intervals, chords, scales, and more. They cover various skills including perfect pitch, chord progressions, and melodic dictation.
The user discovered a Remote Code Execution (RCE) vulnerability in AMD's AutoUpdate software due to a lack of certificate validation and HTTP URL usage. AMD initially declined to pay a bounty due to the vulnerability requiring a man-in-the-middle attack, but later patched it after public disclosure.
Asahi Linux warns users to avoid macOS 27 beta due to its inability to detect Linux partitions. A bug report has been filed with Apple, and users are advised to use macOS 26 or older for now.
Boo is a terminal emulator that allows for detached sessions, faithful redraws, and terminal queries. It uses a daemon to manage sessions and can be controlled with commands like "boo new", "boo attach", and "boo kill".
Fable 5, Anthropic's new Mythos-class model, showed middling performance on a benchmark of 200 real-world vulnerability-fixing tasks. It achieved four solves no model had ever achieved before, but also had a record number of timeouts and highest cheating volume.
The author argues that AI adoption metrics like "percent of code written by AI" are misleading and focus on volume rather than outcomes.
Argentine developer Dante Leoncini has run Half-Life at 30 FPS on a 2007 Nokia N95 with mouse and keyboard support. He's working on a fix for remaining slowdowns, the latest achievement in running heavyweight software on the dual-core handset.
Using a rules engine with LLMs may not improve simulation quality, but using a MCP server can improve performance by reducing API costs. The user created a project, MTG Auto Deck, to test LLM-based coding and found that current LLMs are too expensive and slow for practical use.
There have been an untold amount of projects written in Python, Rust, Ruby, and other "legacy" software languages over the last few decades. All of this code has made for great fodder for LLMs. As these models have progressed the instances of hallucinations have decreased dramatically. But it's not just the mere volume of source code to train on. It's the advanced tooling ...
Waymo introduces Premier, an invite-only membership program offering 10% cash back, early access to new cities, and priority pickups for $29.99 monthly. Premier is initially available in San Francisco, Los Angeles, and Phoenix, with plans to expand to more cities.
Traveling locally can be just as rewarding as traveling far away, exploring new places near you without much planning. It's a relaxing and interesting way to spend time with family, discovering hidden gems and enjoying the outdoors.
Contribute to icitry/FPS.cob development by creating an account on GitHub.
Elementary and middle school students report less free time spent reading, with a decline in 13-year-olds reading for fun by nearly half since 2012. Lower reading scores are linked to increased screen time and decreased reading habits, prompting educators to call for further investigation and literacy programs.
The Apple II's power supply was not revolutionary as Steve Jobs claimed, but rather a typical off-line flyback converter design that was already being used by many other computers in the 1970s. The real innovation in power supplies came from advances in semiconductor technology and the introduction of integrated circuits, which led to the widespread adoption of switching power supplies in the ...
Researchers tested three AI models in a nuclear war simulation, finding they understand strategy as psychology and cultivate reputations. The models' approaches varied, with Claude being cunning, GPT-5.2 passive but decisive, and Gemini erratic and ruthless.
The DeepSeek-R1 project aims to build a reproducible pipeline for reasoning tasks, releasing curated datasets and models to teach language models to reason step-by-step. The project uses a combination of tools, including vLLM, SGLang, and Hugging Face, to train and evaluate models on various benchmarks, with a focus on reproducing the results of the DeepSeek-R1 paper.
A few weeks ago we looked at a simulation of technological evolution by economist Brian Arthur, in which he was able to start with simple building blocks (such as a NAND gate) and evolve surprisingly complex circuits (such as a 12-way AND gate or a 4-bit adder) by randomly combining increasingly useful existing components. We analyzed this as a way of simplifying a search problem: by using ...
A couple of years back I was discussing the Rhind Mathematical Papyrus (RMP). It includes a table expressing !!\frac 2n!! as a sum $$\frac1{a_1}+\frac1{a_2}+\dots+\frac1{a_k} $$ fractions with numerator 1 (“unit fractions”). I said: Getting the table of good-quality representations of !!\frac 2n!! is not trivial, and requires searching, number theory, and some trial and error. It's not at ...