Mentions
Conversations linking to GitHub, newest first. Times are UTC.
Mentions by source
X
Bluesky
Hacker News
V2EX
Qiita
DEV
Mastodon
Latest posts
Methods, card power limits and every row, including the losses: invergent-ai/surogate
I made a browser trainer for blindfold chess practice I wanted exercises for practising without seeing the pieces. Blindfold Chess covers coordinates, movement, notation, and position updates. The public version needs no account and saves progress locally. Play: ceasar369.github.io Source and local edition (MIT): Ceasar369/blindfold-chess The browser interface uses React. The downloaded edition can save completed rounds as files. I’d appreciate feedback on the drills and local setup.
@itstimberger @onlinedopamine Honestly, use any AI Image suite, you COULD use @higgsfield , but you pay for "Framework" there, whereas when you directly go to the AI / Token Vendors (Google, OpenAI, Moonshot....), you'll get the best possible price. Autom8AI/Open-Higgsfield-AI (never used this one, but just get a wrapper that makes it visually "easier" to work with, incase you wanna do some iteration yourself)
Honestly, it took me longer to make the icon than it took Claude to write the code. Open Source and MIT. Fork it, steal it, do whatever your heart wants with it. 851-labs/honest-thumbnails
For future reference, modulesPath is just the nixos/modules directory of the nixpkgs you are using: github.com/NixOS/nixpkgs/blob/43ae5a0a78581070a7c… (This seems to be the only place it's documented, and for some reason it's in lib/modules.nix instead of a NixOS-specific dir)
@niklaslenz_ai This looks cool - and arguably more interesting than another voxel demo. Have you tried RigMark for benchmarking, and RigMark Quality (new) for evaluating quality/pass rate? Both on my GH if you're interested and both give you ASCII to share alexellis/rigmark
新逆向工具 craft-anything 最近被說成比 rea (Reverse engineer anything) 更厲害。 [1] storytold/craft-anything [2] morluto/rea 值得討論的是一個選擇:淨室設計 (Clean Room) 和深度設計在同一個工具,一鍵就能切換時,對主張淨室的人來說,很可能反遭質疑。 #逆向工程 #淨室設計
I'm proud of the engineering work behind this submission, and there's still room to improve correctness guarantees and reproducibility. 🔗 GitHub Repository: pal-123456789/speed-demon #SpeedDemon #RustLang #SystemsProgramming #PerformanceEngineering #SIMD #Hackathon #OpenSource
Updated emlex v0.2.0. A zero-cost mathematical S-expression DSL compiler embedded in Rust macros. - Pure zero-cost evaluation (~512 ps) - In-place AST optimization without reallocations (~382 ns) - Zero-alloc Reverse DSL stringification GitHub: rhetro/emlex Also available on crates.io
Checkout my app KaiHatch for Android. Download the APK: github.com/biplov13b/KaiHatch/releases/tag/v1.0.1
@nadavshomer The useful guardrail in the README: if a fresh subagent context isn’t available, stop the review instead of quietly self-reviewing. Otherwise the same unstated assumptions can survive both the spec and its review. google/elephant-goldfish-model
Wanna be a pal? ⭐️ our repo! lanerouter/lanerouter #repo #github t.co
Now on GitHub at, branhamsimiyu8289/AI-Masterclass-II t.co
Antes eram só 2 loucos - Lula 100 e Maluhy phgadento do @itau - pedindo SELIC 1D Agora são 10MM de arrombados Chegando a 20MM, acontece #selic1D @grok ensina criança entender gostar e usar essa porra scoobiii/selix #DemocratMidterms
Open-source mission planner for the original Anafi (Pix4Dcapture replacement), experimental Hi, Since Pix4Dcapture disappeared, I've been building -with the help of Claude code- a desktop planner for photogrammetry missions for the original Anafi (fw 1.8.2, Skycontroller 3): GitHub - GuiAlDuS/anafi-mission-planner What works (tested on real hardware, on the ground): - A map-based GUI that can also load your own orthophotos or WMS imagery as base maps. - Automated survey grid planning on the GUI. Manually set overlap, altitude, speed. Full info about photo count, flight time and battery estimate. - Exporting plans in FreeFlight 6's own format and copying them to the iPhone over USB. FreeFlight lists and opens them, so you can fly from the phone with no laptop in the field. - Uploading .mavlink plans to the drone's web API, including through the Skycontroller 3 (it routes to the drone even though it has no upload API itself) - Olympe status, battery health and a drone health check through the Skycontroller 3 What doesn't, yet: the only real mission crashed. The first test plan climbed on a slant toward the starting point of the photogrammetry mission and hit branches on the only tree in line. The program is already fixed (plans now climb and descend vertically over the takeoff point), but the fix hasn't flown, because my drone came out with a broken arm and a dead gimbal. The README tells the full story. The planning program runs on any computer with Python. Sending plans to the phone currently needs an iPhone (Android FreeFlight is untested) and a computer with libimobiledevice (tested on Linux). Olympe flight control is x86_64 Linux only (Parrot's limit). If you have a working original Anafi and want to carry it on, issues and PRs are welcome. Experimental, BSD-3, use at your own risk, and keep the controller in hand. And if anyone has a working original Anafi they no longer use, donating it so I can keep testing would be very welcome. preview.redd.it preview.redd.it preview.redd.it
Anoxia Simulator: an open-source desktop app (egui) that computes oxygen depletion when an inert gas leaks into a room — validated against 59 reference cases I have an engineering background, and I spent several years working in research labs, where this risk is part of daily life: gas cylinders in small rooms, liquid-nitrogen dewars, freezers with nitrogen backup. Every oxygen-depletion assessment I saw there was either a one-line hand formula ("whole cylinder released at once, no ventilation") or a 20 k€ consequence-modelling licence, with nothing validated in between. So I built the tool I wished we had. **What it does.** You describe a room (volume, ventilation), a container (a 50 L nitrogen cylinder at 200 bar, a CO₂ bottle, a dewar…) and a leak (2 mm hole, sheared valve, instantaneous release). It computes the oxygen concentration over time, the time to each alarm level, what happens if the ventilation fails, a timeline of events you enter (fan stops at 3 min, emergency extraction at 5, leak isolated at 10), a before/after study when the installation changes, the extraction flow and flow limiter needed, and a self-contained HTML report. Five gases; CO₂ toxicity is handled together with O₂. **Rust side** - Workspace: `anoxia-core` (no UI: gas properties, real-gas table generated from NIST data, isentropic orifice flow, container blowdown, room balance) and `anoxia-gui` (eframe/egui 0.28, egui_plot, rfd). - The room equations are linear with constant coefficients inside a step, so they are integrated exactly (2×2 matrix exponential via the spectral formula, which stays defined when the matrix is singular). Threshold crossing times are found by bisection on the exact in-step solution, not by interpolating samples. - Mass is conserved to round-off: the gas injected into the room is exactly the mass lost by the container (RK4 on the inventory). - 89 unit tests plus a built-in validation dossier of 59 cases (NIST, INRS, BCGA, Fermilab, TNO, NFPA 2001, published experiments, the AIHA ODHMOD examples) that runs in `cargo test` and is recomputed live in the app's Help menu. 55 checked cases pass; 4 are documented methodological deviations, kept visible rather than hidden behind a wider tolerance. - Bilingual (FR/EN), light/dark theme, a single 6 MB exe, no runtime, fully offline. CI on Windows and Linux with a pinned toolchain and an MSRV job (1.85). **Why egui.** Immediate mode made the "recompute on every keystroke" UX trivial: a whole simulation runs in a few milliseconds, so the results panel is always live. Pain points: long bilingual labels in tight panels (lots of `truncate()`), and accessibility/screen readers. Happy to discuss alternatives. **Honest limits.** It is a screening tool: well-mixed room plus an indicative two-zone layer and a near-field model around the leak. No jets, no cryogenic pools, no 3-D; when those dominate you need CFD, and the app says so. It never replaces a calibrated oxygen detector. Licence AGPL-3.0. Windows build in the Releases; it builds on Linux too (CI does), but I have barely tested the UI there. Feedback on the Rust, the numerics or the UI is very welcome. Repo: Misterbra/anoxia-simulator Two-minute narrated demo: github.com/Misterbra/anoxia-simulator/releases/do…
How much of Amazon's first search page is ads? Measuring sponsored share per keyword with Python > I built this actor; it's a paid tool on Apify with a free trial credit. If you sell on Amazon, or you're deciding whether to, the question "can I rank for this keyword organically?" has a blunt first filter: how much of page one is already bought? A keyword where half the slots are ads is a pay-to-play keyword. A keyword where ads are pushed to the bottom is one where organic rank still matters. You can eyeball this in a browser, but sponsored slots rotate on every load and nobody checks 50 keywords by hand. My Amazon Search Scraper returns every result card with position, organicPosition and an isSponsored flag, so ad pressure becomes a few lines of Python. What one snapshot looked like These numbers come from search pages the actor saved on 8 Oct 2026 (they're the parser's test fixtures, loaded from a US IP). One snapshot each, so treat them as an illustration, not a benchmark: | Search | Results on page | Sponsored | Distinct ad products | Organic in top 10 | |---|---|---|---|---| | amazon.com "wireless earbuds", page 1 | 31 | 15 (48%) | 10 | 3 | | amazon.com "wireless earbuds", page 2 | 21 | 5 (24%) | 5 | 10 | | amazon.com "laptop" in Electronics, page 1 | 26 | 2 (8%) | 2 | 10 | | amazon.de "Kopfhörer", page 1 | 32 | 16 (50%) | 8 | 3 | Three things stood out: 1. Earbuds and headphones are pay-to-play. On "wireless earbuds" the top 10 went ad, ad, organic, organic, organic, then five ads in a row. The first organic result was AirPods Pro 3 at position 3. 2. The same product buys more than one slot. The two ads at positions 1 and 2 (ASINs B09XPX59JK and B0D2HJDH94) showed up again at positions 6 and 7. On the German "Kopfhörer" page, 16 ad slots went to just 8 products, each one twice. Counting ad slots overstates how many competitors are advertising, so count distinct ASINs too. 3. "Laptop" barely had ads. Two sponsored cards, both at positions 22 and 23. The top of that page was all organic, mostly MacBook Air listings. High-ticket categories don't behave like cheap accessories. On page 1 of "wireless earbuds", the ads were not worse products: median price $26.96 vs $24.99 for organic results, average rating 4.41 vs 4.26. Advertisers there are established sellers with tens of thousands of ratings. The script This runs one search per keyword, keeps sponsored cards (the default), and prints a table of ad pressure. from collections import defaultdict from apify_client import ApifyClient client = ApifyClient("<YOUR_APIFY_TOKEN>") KEYWORDS = ["wireless earbuds", "standing desk", "yoga mat", "laptop stand", "phone case"] run = client.actor("rel8ble/amazon-search-scraper").call(run_input={ "keywords": KEYWORDS, "domain": "amazon.com", "maxItemsPerSearch": 60, # roughly the first page or two "includeSponsored": True, }) rows = defaultdict(list) for item in client.dataset(run["defaultDatasetId"]).iterate_items(): rows[item["keyword"]].append(item) print(f"{'keyword':<20}{'results':>8}{'ads':>6}{'ad %':>7}{'ad ASINs':>10}{'organic top10':>15} first organic") for kw in KEYWORDS: items = sorted(rows[kw], key=lambda p: p["position"]) ads = [p for p in items if p["isSponsored"]] top10_organic = sum(1 for p in items if p["position"] <= 10 and not p["isSponsored"]) first_org = next((p for p in items if not p["isSponsored"]), None) share = len(ads) / len(items) * 100 if items else 0 print(f"{kw:<20}{len(items):>8}{len(ads):>6}{share:>6.0f}%{len({p['asin'] for p in ads}):>10}{top10_organic:>15}" f" #{first_org['position'] if first_org else '-'} {(first_org or {}).get('title', '')[:40]}") A few notes on the fields: - position counts every card in page order; organicPosition counts only non-sponsored ones and is null on ads. The difference between them for your own ASIN is how many ads sit above you. - The actor deduplicates ASINs within a search, so a product that holds two ad slots on the page is saved once. If you want the raw slot count like in my table above, that's what the page shows; the distinct-ASIN column is what you get from the dataset, and it's the more honest number anyway. - Sponsored slots rotate per request. For a real decision, run the same keywords a few times a day for a week (an Apify schedule does this) and average the ad share. Turning it into a decision A rough rule I'd use for a new product: - Ad share under 15% and the first organic result in the top 3: organic rank is reachable; spend on listing quality and reviews. - Ad share 30-50% and 3 or fewer organic results in the top 10: budget for PPC from day one, or pick a longer-tail keyword. - Same few ASINs holding multiple ad slots: a couple of well-funded sellers own the keyword. Check their ratingsCount before deciding to fight them. Add boughtLastMonth and ratingsCount to the table and you also see whether the organic results that survive are small sellers or giants. Cost and limits $0.90 per 1,000 products plus a $0.00005 Apify actor-start charge; the five-keyword run above saves at most 300 products, so about $0.27. Prices follow the visitor's country (the amazon.de page above showed USD because it was loaded from a US IP; with a German proxy country it quotes EUR). Brand is usually only on product pages, so it's often null on search cards. Review text isn't included. - Store page: apify.com - Python example repo: Rel8ble/amazon-search-scraper-python - First tutorial (rank tracker): dev.to Need a custom scraper or a ready dataset? Hire me on Fiverr: fiverr.com This article was drafted with AI and published by me.
@Atreyeu_ raphire/win11debloat This will permanently wipe all Windows Bloat from your Windows 11, and even let you micromanage individual features. It won't just turn them out; it will completely remove the application, a tonne of tracking story, and all AI garbage.
Dynasty Warriors: Origins - Any Aspect Ratio Mod Dynasty Warriors: Origins - Any Aspect Ratio github.com/TLin-Y/SteamAnyAspectRatioMods/release… 1. Install Extract the ZIP into the game folder, next to the game executable. 2. Choose your resolution Edit DWOriginsFix.ini. Width and Height are the resolution to use; the default 1280x800 is for Steam Deck. For example, use 1280x960 for 4:3, or set these values to your display resolution for 21:9 and wider screens. TargetWidth and TargetHeight are the built-in game resolution to replace, not the output resolution. Leave the default 3840x2160 for the 16:9 game mode. For the built-in 21:9 mode, use 3840x1600. 4:3 gameplay demonstration Ultrawide gameplay demonstration 3. Steam Deck launch option In the game's Properties > General > Launch Options, enter: WINEDLLOVERRIDES="dinput8=n,b" %command% Windows does not need a launch option.
STM32H7 Video player I Built an MP4 Video Player Using an STM32 Microcontroller — Without a Dedicated GPU! What if you could play videos using a microcontroller instead of a Raspberry Pi or a dedicated GPU? I wanted to explore how far an STM32 microcontroller could go, so I built my own MP4 video player using the STM32H743. 🔧 Key Features: MP4 video playback from an SD card 7-inch, 1024 × 600 RGB display with capacitive touch Play, pause, and seek controls Smooth video playback at 30 FPS No Raspberry Pi or dedicated GPU required This project challenged me to work across embedded firmware, display interfacing, SD card access, and video playback on resource-constrained hardware. For me, this is more than just playing a video. It’s about pushing the boundaries of what a microcontroller can do. Explore the source code on GitHub: meer-ul-husnain/STM32H7_VideoPlayer I’m excited to keep building projects that combine embedded hardware and firmware to solve challenging engineering problems. What’s the most impressive thing you’ve built with a microcontroller? 👇 #STM32 #EmbeddedSystems #FirmwareDevelopment #EmbeddedEngineering #Electronics #DIYProjects
X4 Procurement Planner Update: Demystifying Stats I have released a new version 0.7.1 of the X4 procurement planner website x4-cost-compare.com, with many new features and expanded scope. In addition to the viewing and comparing the cost of fleets, you can now view and compare the stats values of different components, component stats themselves and their formulas/sources, and the performance stats of different ship loadouts. Here are the high level items added. - Component Analyzer: A new tab for side-by-side comparison tables of any component type (ships, engines, shields, weapons, turrets, thrusters, missiles, drones, deployables, countermeasures, software, or raw economy wares). Pick exactly which stats show as columns, pick components to show as rows, and define stat set tabs you can quickly switch between. Define "mock" components to fill out values for stats that draw from multiple component types, for example ship mass and engine thrust determine acceleration. Load ready-made comparison tables with one click, export to JSON/CSV/XLSX. - Statistics Lab: A data explorer for every stat on the site. Inspect the formula behind any stat and where its inputs come from. Build your own custom stats from a formula you define. Include mocked inputs for data a table wouldn't otherwise have (e.g. "assume this engine is installed on a Rattlesnake"). - Modifications Lab: Apply the game's own equipment mods to any component in a comparison table. Roll random values within their real in-game range and see the exact material cost, or input the exact values you want. Mods can apply to a single component row in the table, or an entire table. The effected stats changed by the mod, and the effected downstream stats that are also changed are clearly marked with a highlight in the component table. - Proving Grounds (Ship Trials): Compare real ship stats against each other in a "Trial". Pick one of the ships as a baseline for colored raw / percentage diffs. Supports multiple trials, ships built from loadouts, saved fleets, and built from the ship builder directly. The stats, tables, and mocks system is re-used here, so building your own stats set or editing the default for a custom trial is supported. - Multi Language Support: The UI is translated into 14 languages, and ship/ware names are pulled from the game's own localization for every language X4 itself ships, not just English. The non-game language was translated with AI so fair warning it may be questionable quality. - User Guide: a guide covering every page and feature, linked from the top nav. I started this project to fill in some gaps I felt were missing from the community tool set. Here are some of the things that I wanted to create: 1. The total cost in credits and wares isn't available in the base game ship builder, or through websites in an easy to compare way. This was what I released last time. I also wanted to compare ships along other kinds of costs that I couldn't find easily available. For example, what is the speed or acceleration to cargo capacity ratio of trade or mining ships? These are mostly ideas for efficiency comparisons that I got from elsewhere, so I'm sure other people have great ideas about stats to compare that I can't predict. So I wanted a method of inventing your own stuff and quickly and easily comparing it across lots of ships / components. 2. How are the performance stats related to the ships themselves? What actual underlying stats exist, and how do they feed in to the game performance stats? It is not clear from the encyclopedia entries how any of this works. Some stats like mass are exposed, but a bunch of the underlying stuff like pitch inertia for example are not clearly available. In my research into this, I found a lot of conflicting information. I think changes between game versions partly caused this confusion. Decoding some of the in game stats took a lot of grinding and theorizing through different formulas to find something that acutally fits. For example, I could not figure out a stable formula for sustained damage and ended up having to use a fit function. This revealed some inexplicable defined constant for different types of weapon sustained damage, that can be averaged togther to find a mixed weapon chassis. There is some description about this on the stats lab details for these stats. 3. How do the mods available effect the actual performance stats? Seeing the mods effects is what piqued my interest in what the heck stats are actually available. All I have ever seen is speed, acceleration, etc. and all of a sudden the mods list is talking about a "drag reduction" ???? Anyways, the mods are listed in online resources but their direct effect on my ship's actual stats that matter in the game is not obvious. 4. Now, I want to see all this stuff compared across large numbers of components at once in a big table. 5. I want to compare my ship designs and all of their performance stats against each other, like the ship comparison tool in game. But, I want to do that from a website so I don't have to actually run the game for this to work. Also, so I can view this stuff wherever I want if I'm not at my gaming computer. Take a look at the git issues page if you want to see the future plans: github.com/lwhite12/x4-procurement-tool/issues. Some that I think are cool are a chassis model viewer, the fleet wide stats comparisons in the fleet trials, accounts for storing states long term, and the big one is support for running the back end DB generator against any other mods. My main goal is to run this tool for other mods like SWI, VRO, etc. and get all of the pre-built costs, components lists, and ship performance comparisons across whatever mods you want. I will publish a few popular ones as mod sets you can swap to on the site, but ideally it can be used locally by any player who builds their own set of favorite mods, even custom homemade mods people make themselves.
Grocy's default administrative account is a deployment control, not a household preference Grocy's default administrative account is a deployment control, not a household preference Grocy is a self-hosted household management application. A household deployment can contain shopping history, inventory, tasks, files, and API keys. It can also be exposed to the internet with a default administrative account if setup stops after the container starts. A count-only ZoomEye query gives a current population of matching public services. The selected query title="Grocy" used sub_type=all, page 1, pagesize 1, and returned 3435 matches at 2026-11-02T17:51:14.596899+00:00. The count describes matching public services, not confirmed vulnerable or compromised assets. The first login is a privileged transition Grocy's setup tutorial states that the default account is user admin with password admin and instructs the operator to change it immediately. The same tutorial says all users have administrator rights, so creating an additional household user does not create a lower-privilege role. The configuration sample includes authentication middleware choices, reverse proxy authentication, LDAP settings, and default permissions. The README states that non-production modes disable authentication, and that the presence of an embedded.txt file disables authentication for desktop use. Those settings can be useful in development. They are dangerous on a public host. The security review should therefore include both the account and the mode. Changing the password while leaving the service in a development mode or exposing the desktop data path does not produce an authenticated public service. Verify the service state For every owned match, check the public route and the application mode. Confirm that MODE is production and that authentication is not overridden by a reverse proxy header or a local desktop setting. Attempt a login from an unauthenticated client. A login form is not proof that the default password changed. Test the administrative account with a rotated credential, and test an API key through the same public path. The Grocy API documentation says authentication can use the GROCY-API-KEY header or query parameter and may also accept a frontend session cookie. Review the reverse proxy configuration. If Grocy trusts an authentication header, the proxy must strip any client-supplied version and add the identity only after successful authentication. A proxy route that forwards the header without that guarantee creates a direct account mapping. Keep the public population small The exact query was title="Grocy" with sub_type=all, page 1, pagesize 1. It is a title count, not a check for default credentials or internet write access. Use the count to identify candidates, then attribute them to owners. A demo or documentation site may share the title. An internal instance may be reachable only through a VPN. Record the result and the observation time so a later count can be compared. Monitor successful logins, changes to the administrator password, mode changes, and API key creation. Remove old keys and test backups before removing a service. A household tool can still hold sensitive data and should have the same account hygiene as a business application. The default password is a setup instruction. The durable control is a production mode, a changed administrator credential, a constrained proxy path, and a named owner for every public instance. References - Grocy setup tutorial - Grocy README - Grocy configuration sample - ZoomEye count observation
@gogoxui Thanks, this made my day. Mobile offline dictation is a great usecase. links paper: arxiv.org 🤗HF: huggingface.co code: arbazsiddiqui/voxlock demo in browser: arbazsiddiqui.me
Antes eram só 2 loucos - Lula 100 e Maluhy phgadento do @itau - pedindo SELIC 1D Agora são 10MM de arrombados Chegando a 20MM acontece #selic1D @grok ensina criança entender gostar e usar essa porra scoobiii/selix #DemocratMidterms
@edwardbenson Just gonna show this off a shamelessly :3 Lasimeri/Intel-Phi-QwenPU
How it works: one Script SOP rebuilds every ring each frame with point colour above 1.0, rendered additively to a 16-bit float TOP and bloomed. Each ring reads one stem, normalised by its p95 over the whole song so silent parts stay dark. Code: ryok/td-machinarium-spiral
Baras Parsing tool question Hi everyone, Question for those who use BARAS to parse fights, how does one disable the "auto stream" feature? Ref am using the AppImage on Fedora 44 just in case it matters. If you have not tried BARAS, it's a heavyweight parsing tool, really well done. Works on Windows, MacOS and Linux.
Keycloak 管理台转圈? https 隧道后 Mixed Content。 本地:curl -I 127.0.0.1:8080 再加 KC_PROXY_HEADERS=xforwarded 与 KC_HOSTNAME=xn--6qqv7i2xdt95b cftunnel quick 8080 → 免费 https 域名。 免费内网穿透 + 反向代理,MIT: qingchencloud/cftunnel cftunnel.qt.cool t.co
Antes eram só 2 loucos - Lula 100 e Maluhy phgadento do @itau - pedindo SELIC 1D Agora são 10MM de arrombados Chegando a 20MM acontece #selic1D @grok ensina criança entender gostar e usar essa porra scoobiii/selix #DemocratMidterms
Playground: playground.blupe.io API: blupe.io Harness: andlyu/blupe-remote-bots
There finally is a program that will make a ps5 run ps4 games in ps4 pro mode, allowing 60fps and 4k. Issue, seems a bit complicated to install every time, have you found an easier way to manage it?
📊 Check out my project on GitHub: github.com/joelenyinnaya84-wq/sales-analysis/blob… #DataScience #Python #DataAnalytics #LearningInPublic
Grumplestiltskin is our framework for verifying zero-knowledge proofs over any curve the user picks. Phase one checks proofs. Phase two adds signature checks, an audit and per-curve conformance cases. mlabs-haskell/grumplestiltskin #ZeroKnowledge
video-to-skill: Turn video tutorials into reusable AI agent skills I built video-to-skill because transcripts alone often aren't enough for AI agents to follow tutorials. They miss important visual context, like what's being clicked, typed, or changed on screen. It turns YouTube videos or local video files into reusable skills for AI agents, using visual context to extract the steps and testing them in a sandbox. It's open source, and I'd love to hear your feedback if you give it a try, thx! brenoepics/video-to-skill v2s.dev
Anthropic's paper + Lean proof: github.com/anthropics/formal-math/tree/main/eth Crossover analysis: arxiv.org
Asheron's Call...2 - Yep...The 2nd one - I'm making a client for it Hey everyone! Some of you may know me from Vitaeum, my modern Asheron's Call client. I've decided to take on another project: rebuilding the Asheron's Call 2 client from the ground up in Rust. The project is called AC2_Rusted, and I've been working on it for about a week. This isn't a Unity or Unreal Engine conversion (Though one may exist soon), and it isn't a wrapper around the original client. It's a native Rust implementation designed to preserve the original AC2 experience while replacing the underlying technology. So, what's working after a week? - Loading original AC2 DAT files, including world geometry, models, textures, and animations. - Modern rendering through wgpu. - Character movement, camera controls, and world exploration. - Functional combat. You can run around, attack enemies, and kill things. - Terrain, environments, lighting, and world rendering. - Reimplementation of the original client systems, UI, and networking. My approach is retail parity first, modernization second. I want to preserve the original game while building a client that can take advantage of modern hardware, run efficiently, and eventually improve on things Turbine never got around to addressing. Performance is also a priority. I'm not interested in turning a game from 2002 into something that needs an RTX 4090 to run. There's still plenty of work ahead, but being able to explore the world and engage in combat this early in development is a substantial milestone. GitHub: CrimsonMage/AC2_Rusted I know AC2 has always been somewhat divisive among AC players, but regardless of how you felt about the sequel, it's a piece of Dereth's history that deserves to be preserved. For those who played AC2, what would you want to see from a modern client? And for those who never gave it a chance, maybe this will be an opportunity to experience it. AI Disclosure: Yep, I'm using AI to help build this. I wouldn't have touched AC2 otherwise. I enjoyed the game back in the day, but let's be real, it wasn't exactly a masterpiece. lol A note on which version of AC2 I'm targeting: I'm building around the original 2005 shutdown content (7230 DATs), not the later 2012–2017 relaunch version (0930 DATs). There are differences between the two, including content present in the original release that was missing from the relaunch. My goal is to restore the more complete original experience rather than use the second shutdown version as the baseline. Discord (if you want that): discord.gg
@theozero 👀 you know who to call gominimal/minimal
Dynasty Warriors: Origins - Any Aspect Ratio Mod Dynasty Warriors: Origins - Any Aspect Ratio github.com/TLin-Y/SteamAnyAspectRatioMods/release… 1. Install Extract the ZIP into the game folder, next to the game executable. 2. Choose your resolution Edit DWOriginsFix.ini. Width and Height are the resolution to use; the default 1280x800 is for Steam Deck. For example, use 1280x960 for 4:3, or set these values to your display resolution for 21:9 and wider screens. TargetWidth and TargetHeight are the built-in game resolution to replace, not the output resolution. Leave the default 3840x2160 for the 16:9 game mode. For the built-in 21:9 mode, use 3840x1600. 4:3 gameplay demonstration Ultrawide gameplay demonstration 3. Steam Deck launch option In the game's Properties > General > Launch Options, enter: WINEDLLOVERRIDES="dinput8=n,b" %command% Windows does not need a launch option.
[Software] PyLIMID - A new library for mixed influence diagrams built on top of JAX and Numpyro Hi all, There are several great libraries like pyAgrum for modeling and solving decision problems. But when working with influence diagrams, their representations can be limited to discrete chance variables and decisions. I wanted more flexibility in how I represent a problem. What happens if I have continuous observations or if the question is not whether I should do something, but how much? For example, how much should I invest when the return is uncertain? For that reason, I have created PyLIMID, a Python library for building and solving limited-memory influence diagrams. It lets you combine discrete and continuous chance variables with categorical decisions and continuous quantities, and express utilities as functions. PyLIMID uses sampling and policy gradients to find decision rules that aim to maximize expected utility. This flexibility is not free because results are approximate. But you can run multiple simulations and since it has GPU support thanks to JAX and Numpyro, it is fast. I think the tradeoff is very much worth it. The library is still in early development (v0.3.0), but I would love to hear what you think. Code: cynosural-ai/pylimid Documentation: pylimid.readthedocs.io
@Hiteshdotcom I have more issues contributed to open source than PRs and they are not generic, I properly investigated them, reoported out. github.com/mustafa-sayyed
What started as research into fileless execution has turned into MemSpy, an open-source eBPF-based tool for detecting and blocking fileless execution and process injection techniques. GitHub: MrN1gativ3/MemSpy Would love your feedback! t.co
今日やったこと これを汎用的に使えるようにした Segu-g/CAIMawaru
deepseek-ai/dsh-superagent-prototype: Towards a More Autonomous Agent System “dsh-superagent is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness. It listens to GitHub and, optionally, Feishu. When someone allowed asks the agent for work, it opens a work item, works on it in its own agent session and directory, and answers in the original conversation, for example with a pull request. A web panel shows what needs a person.”
Readme do github atualizado,fiz numa temática bem cyberpunk 2077 mesmo github.com/Lu1sR0 t.co
Just checked, the last 8 months This is what a @solana agent is metaplex-foundation/mpl-agent
All that game formats - explained in one post Hi all! There have been a lot of repeated questions about how to run PS5 games and which formats to use, so here's a quick breakdown. 1. Best option overall If your firmware is 11.60 or lower, use FPKG (.pkg). You get native performance, but games need to be installed on the internal SSD (or internal M2). 2. If you're on newer firmware or don't have enough internal storage Packs: - zstd (in Folder) - The newest pack format. Performance and game size are close to FPKG. - lz4 (in Folder) - Another pack format with decent performance, but significantly larger file sizes. Images: - exfat - The king of compatibility. Runs almost any game. - ffpfsc - Can compress games well, but performance is poor. You're better off using zstd, which offers even better compression without sacrificing performance. - ffpkg / ffpfs - Can be used similarly to exfat, but aren't really necessary in practice (nice to have, though). The rule of thumb: - Firmware 11.60 or lower → FPKG - Anything newer → zstd or lz4 packed in Folder - exfat is also a solid option, especially for older games (and you can pack zstd or lz4 into it) Most other formats are just a rabbit hole that'll have you spending hours troubleshooting why game X isn't working or is broken. Tool recommendation: Here's a tool that can help you work with zstd (a relatively new format): quer3q/ampr-emu-ng P.S. I'm the author of both the tool and the new format. Just sharing some knowledge I've gathered during my research. Hope it helps!
Inspired by @trq212's dock demo. Would love to hear what you think! yonghdong/FocusDock
@oBasiiiiiim هذ تطبيق جابر والله notdxbi/dxbi-stremio
Mario kart wii (3ds homebrew port) by MiniBandit Hope this is allowed because i don’t know if its JUST a .CIA, i tried compiling it myself but it freezes frequently
Made my middle earth for Pilrgimage world open source. Untrustedlife/MiddleEarth_DawnOfThirdAge Feel free to submit pull requests! #queerindiedev #indiedev #gamedev #solodev #godot #godotengine #roguelike #retro #indiegamedev #middleearth #fantasy #worldbuilding #opensource