Mentions
Conversations linking to GitHub, newest first. Times are UTC.
Mentions by source
X
Bluesky
Hacker News
V2EX
Qiita
DEV
Mastodon
Latest posts
Hajime no Ippo: Victorious Spirits (PSP) is now fully playable in English – fan translation v1.0 The 2007 PSP Ippo game never left Japan, so I made an English translation patch for it. You can now play through the whole story and understand every scene. What’s translated - Story Mode: all 35 chapters, from Ippo’s first days at Kamogawa Gym through the big fights (Sendo, Volg, Mashiba, Date and more) - Every menu and help line: Exhibition, Practice, Replay, Data Book, Options, save/load - Boxers and moves: names, character versions, special moves (Dempsey Roll, Jolt, Smash…) - The real-boxer roster: ~140 Japanese pros from 2006, with their gyms and fighting styles - Graphics: menu buttons, screen titles, and the arena scoreboards and banners (Dynamite Glove, Champion Carnival, Rookie King finals, the fans’ cheering banners) Names follow the official English manga and anime: Ippo Makunouchi, Mamoru Takamura, Volg Zangief, Jimmy Sisphar, and so on. Sendo’s Osaka attitude, Take’s laid-back Kyushu style and Volg’s polite, careful Japanese all come through in how they talk. How to play It’s a small patch (818 KB) that you apply to your own copy of the Japanese game (ULJS-00125) with Delta Patcher. Then play it in PPSSPP or on a PSP with custom firmware. Full instructions and checksums are on the page. No game files are included or provided. - Download: github.com/Vortex-Lu/ippo-victorious-spirits-engl… - Details: vortex-lu.github.io Found a problem? Post a screenshot in the comments or open an issue on GitHub. Say where you were (mode, chapter or menu), and I’ll fix it in the next update. The translation and hacking were done with AI assistance and checked by playing through the game.
PhysX on AMD GPU for Steam games Hi everyone, with the advent of ZLUDA I wondered if it was possible to get PhysX effects working with an AMD Radeon GPU. Well, yes it is! A good (although short) list of games showcasing GPU effects: Alice: Madness Returns, Batman: Arkham Asylum GOTY, Batman: Arkham City GOTY, Batman: Arkham Origins, Borderlands 2, Mafia II classic, Mirror's Edge I own five of these and can confirm it's working, albeit with occasional side effects (particularly Alice's hair in "Alice Madness Returns" as it occasionally has sparkles). Anyway, here is the guide (I'm using Debian Forky). I hope it helps!! ## INSTALLATION ######################################################## ## Prerequisites ####################################################### 1a) Grant access to GPU by adding the current user to the 'render' and 'video' groups: sudo usermod -a -G render,video $LOGNAME # then, logout and back in again # create UDEV rule to make sure user can access GPU hardware... sudo tee /etc/udev/rules.d/70-amdgpu.rules << EOF KERNEL=="kfd", GROUP="render", MODE="0666" SUBSYSTEM=="drm", KERNEL=="renderD*", GROUP="render", MODE="0666" EOF sudo udevadm control --reload-rules sudo udevadm trigger 1b) Install ROCm stuff: sudo apt install rocminfo # check for a least Radeon RX 6000 series (5000?) or 660M iGPU rocminfo | grep -E 'Agent|Marketing Name:' # below is probably not be needed, but I installed during testing phase sudo apt install libamdhip64-6 libhsa-runtime64-1 librocm-smi64-1 rocm-smi hipcc libgflags2.2 libatomic1 libquadmath0 ## Configure game to use Proton-ZLUDA ################################# 2a) Obtain Proton-ZLUDA from: UndeadBulwark/Proton-ZLUDA and extract to Steam's 'compatibilitytools.d' location, with named folder: ZludaProton tar --zstd -xf Proton-ZLUDA-*.tar.zst -C ~/.steam/debian-installation/compatibilitytools.d/ 2b) Modify Steam's Compatibility tool option: In Steam, right-click on desired game (ex. Alice Madness) then... Properties → Compatibility → Force the use of a specific Steam Play compatibility tool → ZludaProton 2c) Set Steam Launch option (so that bundled ROCm libraries can be found by linux system): LD_LIBRARY_PATH="$HOME/.steam/debian-installation/compatibilitytools.d/ZludaProton/zluda-libs:$LD_LIBRARY_PATH" %command% ## Install PhysX software ############################################# 3) Run/launch the game at least once, so that its 'compdata' folder is created. Exit game. Then, use protontricks to install PhysX software into game's compdata folder... protontricks 19680 physx # Alice: Madness Returns protontricks 50130 physx # Mafia II classic protontricks 200260 physx # Batman: Arkham City GOTY protontricks 35140 physx # Batman: Arkham Asylum GOTY protontricks 209000 physx # Batman: Arkham Origins protontricks 17410 physx # Mirror's Edge protontricks 49520 physx # Borderlands 2, however doesn't seem to be needed ## If needed, ProtonTricks can be installed via: sudo apt install python3-pip python3-setuptools python3-venv pipx pipx install protontricks ## Turn on PhysX in game ############################################## 4) After starting game, go to Options/Settings → Video/Graphics and set PhysX to High/On ## TROUBLESHOOTING ##################################################### ## Get Xinput Game Controller working ######################################## 5a) Load uinput kernel module (if not already loaded), via: sudo insmod uinput 5b) Enable Steam Input In Steam, right-click game then: Properties → Controller and select Enable Steam Input. 5c) Disable SDL for "winebus" service protontricks 19680 regedit In Registry Editor, navigate to: HKEY_LOCAL_MACHINE → System → CurrentControlSet → Services → winebus Create or edit a DWORD (32-bit) Value named: Enable SDL Set its value to: 0 5d) Add gamescope to Steam Launch option gamescope --fullscreen -w 1920 -h 1080 -W 1920 -H 1080 -- %command% # also, grab keyboard for Alice Madness Returns and for Borderlands2 gamescope --fullscreen --grab -w 1920 -h 1080 -W 1920 -H 1080 -- %command% ## Fix stutters caused by gamescope #################################### 6) after 30+ minutes of game play, gamescope causes stutters / frametime spikes (confirmed for Alice Madness) SteamOverlay seems to be the culprit, but one simply cannot disable it because that disables SteamInput gamescope is needed, too. So, the fix is to unset LD_PRELOAD before gamescope, and restore it for actual game. Like this... env -u LD_PRELOAD gamescope -- env LD_PRELOAD="$LD_PRELOAD" %command% Thus, complete launch option looks like (for Alice Madness Returns with the Madness-Patch for QoL improvements)... LD_LIBRARY_PATH="$HOME/.steam/debian-installation/compatibilitytools.d/ZludaProton/zluda-libs:$LD_LIBRARY_PATH" WINEDLLOVERRIDES="dinput8.dll=n,b" env -u LD_PRELOAD gamescope -r 60 -f --grab -w 1920 -h 1080 -W 1920 -H 1080 -- env LD_PRELOAD="$LD_PRELOAD" %command%
@hyperbolic_labs Would like to integrate hyperbolic cloud with InftyAI/Nebula someday.
para quem está largando o grok bot e indo para o Hermes: hermes-agent#147724
@esrtweet A while ago I replaced all my public repo CoC with this. You might find it familiar: aadler/lamW
Can you fix a trained model by editing its weights, with no training?After @rikiyatakehi @a1zhang @lateinteraction’s Weight-Space Programming post, I turned the method into an open agent skill: weight-surgery 🧵lorenzofamiglini/weight-surgery
Hierarchos Native update: faster streaming on Intel HD 520, RDNA3 qualification, and a GPT-2 parity fix I’ve pushed a substantial update to Hierarchos Native, my Rust + Vulkan backend for language-model training, fine-tuning, and inference. This past week focused on porting the recent RDNA3 work to Intel Gen9, improving parameter streaming, and taking the changes back to physical RDNA3 hardware for regression testing. The native runtime requires no Python, PyTorch, CUDA, ROCm, or oneAPI installation. PyTorch and Transformers are used as development references. What changed: - DeepSeek V4.1 Flash port and optimization work: fused selected-expert execution, batched expert uploads, packed-weight paths, checkpoint caching, and native LoRA/rsLoRA training with save/resume/export. - Broader Gen9 coverage: fixes for Qwen3-Next padding and DBRX expert layouts, plus streaming paths for T5/BART/Switch encoder-decoder projections. - Less streaming overhead: retained file handles, fewer temporary allocations, coalesced reads, and removal of an unused hidden-state download. - Gen9 kernel improvements: parallel sparse-attention selection and hyper-connection statistics, plus fused SiLU/up-projection multiplication. - Bounded upload batching: small streamed projections can share a GPU submission within a 1 MiB coalesced payload window. The biggest Gen9 result was closing the Llama-wide streaming slowdown on an i5-6200U / Intel HD Graphics 520: |Measurement|Previous pass|Final pass| |:-|:-|:-| |Median forward latency|11.016 ms|4.597 ms| |GPU submissions per forward|19|4| That’s 58.3% lower latency, with every measured logit word unchanged. Across eight scaled architecture controls, the combined optimization passes reduced streaming medians by 42.4–78.4% and resident medians by 3.5–32.1% against the documented baseline. Gen9 measurements and methodology. Back on the Ryzen Z1 Extreme / RDNA3 device, the regression sweep caught a padded GPT-2 numerical parity issue. The fix matches the pinned CPU reference’s GELU/tanh behavior and short-row attention arithmetic. The original CPU expectations, checkpoints, and strict tolerance stayed unchanged. The completed RDNA3 qualification now passes: - 166/166 forward runs, covering 43 scaled profiles across 26 architecture identifiers. - 72/72 generation requests, checking cache-requested and full-prefix execution. - Maximum absolute CPU logit error strictly below 2e-7, with rtol=0. - Bit-identical resident and streamed outputs throughout the measured matrix. - Native streaming/offload tests and GPT-2 LoRA gradient, optimizer, save/reload, and resume checks. The GPT-2 repair is selected specifically for RDNA3; all 585 pre-existing shader binaries remain unchanged. Final GPT-2 closure and evidence. The RDNA3 testing also recorded lower V4.1 latency versus the previous handoff: 24.4% for resident prefill, 50.2% for resident decode, 17.2% for streamed prefill, and 18.4% for streamed decode. These are measurements on retained scaled checkpoints, rather than production-model throughput figures. RDNA3 measurements. Parameters remain resident by default; supported frozen-weight streaming is opt-in through --stream-parameters or the GUI. The registry contains 146 text architectures, while this qualification specifically covers the 26 identifiers above. Multimodal towers and arbitrary production checkpoints need their own validation. If you’re experimenting with native Vulkan training or inference on older Intel graphics or AMD hardware, I’d be interested in your results—and which workloads you’d like tested next.
Looks like TS 7.1 breaks Effect in multiple ways: effect#10306 - cc @typescript
@ranjan_29X Hey, I’m interested in the UI/UX Designer role. I have 4 years of experience in UI/UX and Product Design, alongside 4 years of software development experience. Portfolio: getbndlabs.com GitHub: github.com/godadcore I’d love to connect and learn more about the opportunity!
Finally jumped into @grok @bot over the weekend. Wired up a @telegram connector with media, buttons, and voice. Woke up Monday to find I’d already burned >50% of my weekly limits gone. Haven’t done anything useful yet 🙈. asmagin/grokbot-telegram
[For Hire] Computer Engineering Graduate Looking for Entry-Level Remote Work in AI Automation / n8n reddit.comHi everyone! I'm a Computer Engineering graduate from the Philippines currently looking for a job, preferably remote, in AI automation, workflow automation, or a related junior tech role. I'm still entry-level and don't have professional experience in automation yet. I can build working workflows and explain how their main components work, although I still rely on documentation and AI assistance when implementing unfamiliar features. I'm willing to learn and improve on the job. I'm open to remote opportunities with international clients, as well as onsite or hybrid positions around Taguig, BGC, Ortigas, and Makati. GitHub: github.com/Azarath-max If you're hiring, know someone looking for a junior automation assistant, or have advice on where I can find opportunities, I'd really appreciate it. Thank you!
4. Claude Code: the agentic coding tool that lives in your terminal. anthropics/claude-code
GTA V HAS BEEN RECOMPILED FOR PC, MAC, iOS & ANDROID! GTA V (Xbox 360) recompiled: Windows, Linux, macOS, iPhone and Android andrewnakas/gtav-recomp
@rikiyatakehi @a1zhang @lateinteraction @jayden_teoh_ Great work. I turned the method into an open agent skill, with a gate taken from your Gemma control: is the ability in the network or not? Public test: news NER on Wikipedia, head edit (0.006% of weights) → F1 0.479 → 0.588 lorenzofamiglini/weight-surgery
@christophersaum AgentBull/ankole Ankole is an open-source workspace (Pod) and runtime for AI-native teams. In one browser workspace, people and AI Agents share collaborative Documents and Slides, relational Base tables, on-demand Applets with stateless cloud backends, and multiplayer multi-Agent Pod Chat, while Slack, Lark/Feishu, Microsoft Teams, DingTalk, WeCom, and email connect as external channels. Ankole organizes company work into human-in-the-loop operational loops. Agents monitor data, build internal tools, and run long tasks in dedicated Linux containers; people supply real-world context and approve high-impact actions on Choice Cards. Shared Company Brain memory, Skill Lessons, passwordless authentication, and four-tier permissions ship in the Apache-2.0 self-hosted core.
LightMIS: Ultra-lightweight medical image segmentation for edge devices Hi everyone, I am one of the authors of LightMIS, and I would like to share our work on 2D binary medical image segmentation. LightMIS has 131K parameters and aggregates multi-scale encoder features directly, without a learned stage-wise decoder. It achieves close observed modality-macro Dice and IoU averages to nnWNet, while using 98.14% fewer parameters and 94.58% fewer GFLOPs, with both models trained and evaluated under the same nnU-Net protocol. Paper: arxiv.org Code: AndreiiArhire/LightMIS We also tested LightMIS and the reference models on a real edge device with an Arm Mali-G52 MC2 GPU, where the full LightMIS model ran around 10.5× faster than nnWNet. All LightMIS variants achieved 100% GPU operator coverage, without CPU fallback. We started by training several popular medical image segmentation architectures, both lightweight and heavyweight, from scratch in the same controlled environment using nnU-Net v2.3.1. The experiments covered six datasets and five imaging modalities. Our goal was to develop a lightweight architecture that performs well across these modalities, which led us to the current version of LightMIS. The source code, checkpoints, and details needed to reproduce the results are publicly available. Feel free to experiment with it and try it on your own segmentation tasks. I would be happy to hear how it works for you! Any suggestions or feedback are welcome!
NooDS U: New NDS emulator heavily optimized for the Wii U Hey everyone, just wanted to share my little sideproject. NooDS was already running on Wii U, but very poorly. I made my own mission to optimize it to the point to run Pokémon Black full speed, which I achieved today. VC injection gets you so much, and I really was missing an alternative for games that can't be injected. More informations: michyprima/NooDSU Enjoy! Notices: This is alpha software tested on real hardware. I'm a professional developer. I use AI.
Hierarchos Native update: faster streaming on Intel HD 520, RDNA3 qualification, and a GPT-2 parity fix I’ve pushed a substantial update to Hierarchos Native, my Rust + Vulkan backend for language-model training, fine-tuning, and inference. This past week focused on porting the recent RDNA3 work to Intel Gen9, improving parameter streaming, and taking the changes back to physical RDNA3 hardware for regression testing. The native runtime requires no Python, PyTorch, CUDA, ROCm, or oneAPI installation. PyTorch and Transformers are used as development references. What changed: - DeepSeek V4.1 Flash port and optimization work: fused selected-expert execution, batched expert uploads, packed-weight paths, checkpoint caching, and native LoRA/rsLoRA training with save/resume/export. - Broader Gen9 coverage: fixes for Qwen3-Next padding and DBRX expert layouts, plus streaming paths for T5/BART/Switch encoder-decoder projections. - Less streaming overhead: retained file handles, fewer temporary allocations, coalesced reads, and removal of an unused hidden-state download. - Gen9 kernel improvements: parallel sparse-attention selection and hyper-connection statistics, plus fused SiLU/up-projection multiplication. - Bounded upload batching: small streamed projections can share a GPU submission within a 1 MiB coalesced payload window. The biggest Gen9 result was closing the Llama-wide streaming slowdown on an i5-6200U / Intel HD Graphics 520: |Measurement|Previous pass|Final pass| |:-|:-|:-| |Median forward latency|11.016 ms|4.597 ms| |GPU submissions per forward|19|4| That’s 58.3% lower latency, with every measured logit word unchanged. Across eight scaled architecture controls, the combined optimization passes reduced streaming medians by 42.4–78.4% and resident medians by 3.5–32.1% against the documented baseline. Gen9 measurements and methodology. Back on the Ryzen Z1 Extreme / RDNA3 device, the regression sweep caught a padded GPT-2 numerical parity issue. The fix matches the pinned CPU reference’s GELU/tanh behavior and short-row attention arithmetic. The original CPU expectations, checkpoints, and strict tolerance stayed unchanged. The completed RDNA3 qualification now passes: - 166/166 forward runs, covering 43 scaled profiles across 26 architecture identifiers. - 72/72 generation requests, checking cache-requested and full-prefix execution. - Maximum absolute CPU logit error strictly below 2e-7, with rtol=0. - Bit-identical resident and streamed outputs throughout the measured matrix. - Native streaming/offload tests and GPT-2 LoRA gradient, optimizer, save/reload, and resume checks. The GPT-2 repair is selected specifically for RDNA3; all 585 pre-existing shader binaries remain unchanged. Final GPT-2 closure and evidence. The RDNA3 testing also recorded lower V4.1 latency versus the previous handoff: 24.4% for resident prefill, 50.2% for resident decode, 17.2% for streamed prefill, and 18.4% for streamed decode. These are measurements on retained scaled checkpoints, rather than production-model throughput figures. RDNA3 measurements. Parameters remain resident by default; supported frozen-weight streaming is opt-in through --stream-parameters or the GUI. The registry contains 146 text architectures, while this qualification specifically covers the 26 identifiers above. Multimodal towers and arbitrary production checkpoints need their own validation. If you’re experimenting with native Vulkan training or inference on older Intel graphics or AMD hardware, I’d be interested in your results—and which workloads you’d like tested next.
@MohitRajRathor1 Hey, I’m interested in the UI/UX Designer role. I have 4 years of experience in UI/UX and Product Design, alongside 4 years of software development experience. Portfolio: getbndlabs.com GitHub: github.com/godadcore I’d love to connect and learn more about the opportunity!
Open-source template turns Claude Code into a game studio, and admits where paperwork stops paying Open-source template turns Claude Code into a game studio, and admits where paperwork stops paying Claude Code Game Studios gives a solo developer a team of AI roles, and publishes its own test of how much process is too much. Full article with sources: zolna.uk Project repository: Donchitos/Claude-Code-Game-Studios ZOLNA LTD is a UK software company: custom software, AI and automation, cloud and DevOps. Free first consultation and setup help: info@zolna.uk · zolna.uk · @zolnaltd
Repo (Apache-2.0): SAOS-3624/dsh-see-now Install in DSH: dsh plugin --profile web add dsh-see-now
This video was cut by an AI, from article to narrated edit. Tencent's new open WeVisCut, a 9B model, plans shots, finds footage, adds voice and music in about 12 steps. On its makers' own test it matches Claude Opus 5 with 33x fewer tokens. Tencent/WeVisCut #AI t.co
Open-source FreeToken claims to run 290B-plus AI models on a gaming PC Open-source FreeToken claims to run 290B-plus AI models on a gaming PC A new serving engine splits giant Mixture-of-Experts models across GPU, CPU and memory, and the speed figures are the project's own. Full article with sources: zolna.uk Project repository: FlashML-org/FreeToken ZOLNA LTD is a UK software company: custom software, AI and automation, cloud and DevOps. Free first consultation and setup help: info@zolna.uk · zolna.uk · @zolnaltd
Sources: Theo's post (Nov 1, 5:26 PM PT): the in-app star note after 5 prompts "Immediately 10x'd the number of stars per day". x.com The code: T3 Code PR #27832, "ask for a GitHub star once after 5 prompts", merged Oct 30. t3code#27832 Theo's managed-policies post: T3 Code now supports managed policies for IT teams (PR #27197, merged Nov 1). x.com GitHub Trending (daily): "3,216 stars today", read 7:49 AM PT Nov 2. github.com/trending T3 Code repo and README: MIT, 40,635 stars as of 7:49 AM PT Nov 2, controls agents from mobile, web and desktop, supported agents. pingdotgg/t3code
Open-source Kronos is an AI model trained to read candlestick charts like a language Open-source Kronos is an AI model trained to read candlestick charts like a language A research model forecasts price candles after learning from data from over 45 global exchanges. It is not a trading system. Full article with sources: zolna.uk Project repository: shiyu-coder/Kronos ZOLNA LTD is a UK software company: custom software, AI and automation, cloud and DevOps. Free first consultation and setup help: info@zolna.uk · zolna.uk · @zolnaltd
@hanwenjiang1 really reminds me of this: zou-group/textgrad, but astra clearly "immerse" out the relationship between numerical optimization and text.
Hierarchos Native update: faster streaming on Intel HD 520, RDNA3 qualification, and a GPT-2 parity fix I’ve pushed a substantial update to Hierarchos Native, my Rust + Vulkan backend for language-model training, fine-tuning, and inference. This past week focused on porting the recent RDNA3 work to Intel Gen9, improving parameter streaming, and taking the changes back to physical RDNA3 hardware for regression testing. The native runtime requires no Python, PyTorch, CUDA, ROCm, or oneAPI installation. PyTorch and Transformers are used as development references. What changed: - DeepSeek V4.1 Flash port and optimization work: fused selected-expert execution, batched expert uploads, packed-weight paths, checkpoint caching, and native LoRA/rsLoRA training with save/resume/export. - Broader Gen9 coverage: fixes for Qwen3-Next padding and DBRX expert layouts, plus streaming paths for T5/BART/Switch encoder-decoder projections. - Less streaming overhead: retained file handles, fewer temporary allocations, coalesced reads, and removal of an unused hidden-state download. - Gen9 kernel improvements: parallel sparse-attention selection and hyper-connection statistics, plus fused SiLU/up-projection multiplication. - Bounded upload batching: small streamed projections can share a GPU submission within a 1 MiB coalesced payload window. The biggest Gen9 result was closing the Llama-wide streaming slowdown on an i5-6200U / Intel HD Graphics 520: |Measurement|Previous pass|Final pass| |:-|:-|:-| |Median forward latency|11.016 ms|4.597 ms| |GPU submissions per forward|19|4| That’s 58.3% lower latency, with every measured logit word unchanged. Across eight scaled architecture controls, the combined optimization passes reduced streaming medians by 42.4–78.4% and resident medians by 3.5–32.1% against the documented baseline. Gen9 measurements and methodology. Back on the Ryzen Z1 Extreme / RDNA3 device, the regression sweep caught a padded GPT-2 numerical parity issue. The fix matches the pinned CPU reference’s GELU/tanh behavior and short-row attention arithmetic. The original CPU expectations, checkpoints, and strict tolerance stayed unchanged. The completed RDNA3 qualification now passes: - 166/166 forward runs, covering 43 scaled profiles across 26 architecture identifiers. - 72/72 generation requests, checking cache-requested and full-prefix execution. - Maximum absolute CPU logit error strictly below 2e-7, with rtol=0. - Bit-identical resident and streamed outputs throughout the measured matrix. - Native streaming/offload tests and GPT-2 LoRA gradient, optimizer, save/reload, and resume checks. The GPT-2 repair is selected specifically for RDNA3; all 585 pre-existing shader binaries remain unchanged. Final GPT-2 closure and evidence. The RDNA3 testing also recorded lower V4.1 latency versus the previous handoff: 24.4% for resident prefill, 50.2% for resident decode, 17.2% for streamed prefill, and 18.4% for streamed decode. These are measurements on retained scaled checkpoints, rather than production-model throughput figures. RDNA3 measurements. Parameters remain resident by default; supported frozen-weight streaming is opt-in through --stream-parameters or the GUI. The registry contains 146 text architectures, while this qualification specifically covers the 26 identifiers above. Multimodal towers and arbitrary production checkpoints need their own validation. If you’re experimenting with native Vulkan training or inference on older Intel graphics or AMD hardware, I’d be interested in your results—and which workloads you’d like tested next.
@zeddotdev @elixirlang @java @rubylangorg @official_php @SwiftLang @kotlin @dart_lang Hello, Some people were waiting for zed#63698 to be merged. Is there any update regarding this PR? Thanks!
if you want to try it out early, you can build it from source at the github, or join the waitlist for potential early access! photonapp.co.uk photonbrowser/photon
Introducing the DAM Valuation and DAM Ratio. Most valuations break when growth is high and earnings are negative. The Dual Anchor Model (DAM) capitalizes current NI and adds a pure revenue-growth term. At g=0 it collapses to a simple DCF. foysavas/dam-valuation t.co
i'mm a NERDD heres my ref repoo Squishy6094/reference-sheet
@priyanshkalkal github.com/Awis13/liberty-pro-max/releases Basically you just need to download a PC 1.0.8.0 version, i used a Rockstar launcher game and downgraded it, than just run a builder and it will transform .exe to .app
XNox now open sourced I made XNox, couple of months ago. but as it was having too complicated features to use. I decided to simplify it and make it open source enjoy emadhegab/XNox
🔗 GitHub: lioensky/VCPToolBox --- ✉️ If you’re into AI, ML, agents, and building real systems, join my newsletter (it’s free): dankornas.substack.com
How Google’s $2T Engineering Team Built a GitHub Integration That Forgot to Run "unzip" The Setup I wanted Gemini to do a code review of a public, open-source Kotlin Multiplatform repository on github.  clavierhaus/clavier-tuner-foss Attempt 1: Give it the public GitHub URL I paste the public repo link. Gemini’s web browsing tool fetches github.com, receives an empty, unrendered HTML shell consisting of 40 <link rel="modulepreload"> tags and a copyright footer, and tells me it can't see any code. Attempt 2: Jump through the OAuth hoops Fine. I go into Google’s settings, activate the official GitHub integration, authorize OAuth, and grant full read access to the repository. I use the official + -> Import Code attachment menu to link the repo directly into the prompt. What Google’s Backend Actually Did Instead of cloning the tree or fetching text blobs via the GitHub API, Google’s backend connector downloaded the commit snapshot (7d41b3ac...) as a .zip archive from GitHub’s zipball endpoint.  And then it forgot to unzip it.  Whoever wrote Google's file ingestion pipeline just ran a dumb printable-ASCII scraper (basically strings) directly over the raw, compressed .zip binary and piped the output straight into the LLM's context window:  Every single file entry in the model's prompt started with dbb] ... UT—which is the ASCII debris of a raw PKZIP local file header and extended timestamp.  Every Kotlin source file (TuningController.kt, CleanReading.kt, PhaseReader.kt) was handed to the LLM as raw, compressed DEFLATE gibberish (~NU/ /bY%e$ U"dU...).  Because the repo includes Fastlane store metadata and assets, the ASCII scraper plowed through five PNG screenshots (1.png–5.png), two JPEGs (clavierhaus_hall.jpg, hub_background.jpg), gradle-wrapper.jar, and three .ttf font files (allura_regular.ttf, dejavu_serif.ttf), stuffing the context window with tens of thousands of tokens of raw IDAT chunks and binary entropy until the buffer overflowed and truncated the repository before even reaching the main UI module.  The Cherry on Top When Gemini received 100,000+ characters of PKZIP headers and IDAT PNG chunks from Google's own GitHub integration, it blamed me for uploading a .zip file, complained that it couldn't read DEFLATE streams, and told me to open a terminal and pipe git ls-files | xargs tail into a /tmp/*.txt file so it could read my public repository. TL;DR: Public GitHub URL? Can't render the page. Official OAuth GitHub Integration? Downloads a .zip from GitHub, forgets to decompress it, scrapes printable ASCII out of the raw binary, feeds the LLM 50 pages of raw PNG IDAT chunks, blows the context window, and then lectures the user for uploading a zip file. Oh, of course all that burned through my quota like a whirlwind and left me waiting for a rest in a couple of hours. Meanwhile, Claude and Mistral actually developed a really cool piece of software.
@anupamrjp notchguard: your mac's notch camera locks it when you walk away. free, open source, runs fully on your mac mkbuilds4/notchguard t.co
WhatsApp calls don't ring on my Pixel Watch, so I built a fix. Looking for testers My watch never rang for WhatsApp calls, and Telegram and Instagram calls don't ring on a Wear OS watch on their own either. So I built Wristline. It makes those calls ring on the watch like normal phone calls. You can answer, decline and hang up from your wrist. With a one-time setup using a computer, you can also: - talk through the watch's speaker and mic - search your contacts and call them on WhatsApp or Telegram from the watch - keep favourites on a tile or complication The honest part: I've only tested it on one phone and one Pixel Watch. It installs from a file, not from a store, and it needs notification access to see incoming calls. If you try it, please tell me your phone and watch model and whether it worked. Guide and download: lucio-chiummo/Wristline Not affiliated with WhatsApp, Meta, Telegram or Google. repo here
Source: "Sparks of Weight-Space Programming" by Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab (MIT CSAIL): rikiyat.github.io. Full logs: RikiyaT/lm_understand-logs GSM8K moved too, 13.8% to 67.1%. The authors' own line on the limit: "Editing weights could remove an obstacle that hid an ability, but it could not add an ability the model lacked."
[Plugin Release] MolleFreeCam 1.0 - FreeCam plugin for your server! Hello everyone! I just released version 1.0 of my plugin MolleFreeCam, which adds a freecam mode to your server so players can watch their buildings from other perspectives. To get it to run, you just need to drop the plugin in your plugins folder and players are able to use the command /float to enter freecam mode. Details and the download can be found here: Github: wasamolle/MolleFreeCam Hangar: wasamolle/MolleFreeCam This plugin is heavily inspired by LunarFreecam by CyberGl1tch, which I used on my previous minecraft server. This year I started over with a new server on 26.3 and sadly LunarFreecam does not work in this version anymore and the repo for it seems to be abandoned. As I always wanted to learn how to program plugins anyway, I looked at the code and wrote a new plugin which uses the same core mechanics. For now, it is only tested on Paper 26.3 and I initially only did it for my own server, but I figured maybe someone else could have a need for this plugin, so I decided to release it. I am open to constructive criticism and maybe also willing to add requested features in the future, if there will be any. If this is the wrong subreddit, I am sorry and open to suggestions for other subreddits to post on! No LLMs were used for writing this plugin, as I strongly believe that it is important to be able to understand and write code on your own. Greetings, wasamolle
Built a free, open-source tool to help navigate dense literature reviews, finding open-access PDFs, and organizing papers (ResearchLoom) Hey everyone, When starting out in research—whether it's for a senior capstone, summer fellowship, or your first lab project—literature reviews can feel overwhelming. Juggling dozens of 20-page papers, trying to understand dense methodologies, and figuring out what’s actually relevant takes hours. On top of that, a lot of popular research tools lock features behind expensive monthly subscriptions that student budgets don't cover. To make literature reviews much less intimidating, I built ResearchLoom (Samhita)—a completely free, open-source desktop app that organizes and streamlines paper discovery and synthesis: 🔗 GitHub Repository:sohamsans/researchloom 📦 Free Windows & Mac Downloads:github.com/sohamsans/researchloom/releases How it helps if you're working on a research project: 1. One Search Across Multiple Libraries: Instead of bouncing between Google Scholar, arXiv, and publisher sites, it searches OpenAlex, arXiv, Crossref, and Europe PMC all at once. 2. Quick Abstract Triage & Breakdown: Evaluates candidate papers and pulls out the core research questions, methods used, and study limitations into clean summary cards so you don't have to read 50 full papers just to find the 5 you actually need. 3. Finds & Saves Open-Access PDFs Automatically: Whenever an open-access version exists, it downloads the PDF straight to a local folder on your computer and opens it in an integrated reader with an AI assistant you can ask questions to. 4. Visual Flow Maps & PRISMA Review Tracking: Helps you visually map how papers connect and generates PRISMA flowcharts (handy if your professor wants a formal systematic review section). 5. Instant Citation Export: Exports clean BibTeX, APA, MLA, or Excel spreadsheets with one click for your bibliography. 6. Completely Free & Private (No Paywalls): It uses Bring Your Own Key (BYOK) with free tiers from providers like Gemini or Groq, or local models. Nothing is sent to a paid subscription service, and your notes stay stored locally on your machine. Pre-built standalone apps are ready to download on GitHub (single .exe for Windows 10/11 and an installer for macOS—no coding or terminal setup needed). If you're currently working on a capstone or paper review, give it a try! Feedback and feature suggestions from fellow student researchers are always welcome.
TaleSpire VR Mod - Github release Hi all, Over a week ago I shared a video demo of my (vibe-coded) VR Mod for TaleSpire, including a secondary "helicopter" camera view for streaming, First-Person mode (with weapon attacks), and general VR quality of life changes to menus, controls, and dice-rolling. If anyone is interested in trying out the initial version, I've created an "official" Github release: github.com/brownboy3000/talespire_vr_mod/releases… Installation should be as easy as unzipping, dragging, and dropping into your Steam TaleSpire install folder. There are instructions for easy uninstallation as well, but I've also added a feature where on the main menu you can switch back and forth between VR mode and regular mode (aka stock vanilla TaleSpire). If something goes wrong there is a .ini config file you can edit to force VR mode or regular mode (in fact there are several features you can toggle or edit with .ini config files, all in the /VRMod/ folder). The README is still AI-generated text (I will replace it with a more sensible version soon) so please let me know if you have any compatibility issues, see any bugs, or have questions about the features or control scheme. I would especially love to know people's experiences with multiplayer and with headsets/controllers that are different to mine. I am not planning to support this mod too actively (and I'm definitely not looking for donations to keep this project active), but if you have a feature request or see a major bug let me know anyway! If it makes sense to create an update I'll work on it when I can and share. If you're interested in how this was made, I'm also happy to answer questions (I promise I'll try to answer with more details than just "I made Claude do it").
I made a tiny Windows widget that shows your Google Antigravity /usage stats, inspired by clawd-usage-winwidget Hi everyone! I built agy-usage-winwidget, a small floating window and tray icon for Windows. It shows your Google Antigravity weekly usage, the same numbers you get from /usage in the Antigravity CLI, and refreshes them automatically. Features: - Two progress bars: Gemini models (Flash, Pro) and Claude & GPT models (Opus, Sonnet, GPT-OSS), each with its own percentage and reset countdown - Colours change from green to yellow to red as usage grows - Hexagonal tray icon that shows the current percentage at a glance - 24-hour history chart in the details panel - Remembers its position, and you can zoom it with the mouse wheel - No dependencies: just PowerShell 5.1 + WPF (built into Windows) and agy on your PATH - English and Spanish UI Download: IberianSoldierPC/agy-usage-winwidget (grab the latest release ZIP and double-click start-widget.vbs) Huge thanks to Venger Sergey (Enhed) for clawd-usage-winwidget (original post), a lovely Claude Code usage widget. It inspired the architecture, the UI and many of the patterns I used here, and it's MIT-licensed. If you use Claude Code on Windows, go check it out! 🧡 Feedback and PRs are welcome.
One job, one link, zero arguments: a delivery your client checks step by step Client work rarely blows up over the work. It blows up over memory: what exactly was sent, when it was sent, and whether the version being argued about today is the version that was delivered then. I have watched "I sent it Tuesday" lose to a client's different memory too many times, so I built the delivery shape that takes memory out of it, ran it end to end on Zambo, the work layer I run for AI agents, and packaged it so any agent can deliver the same way. The shape, in plain words: the whole job lives inside one record. Every step is saved as you do it, each with a verifiable receipt. The finished work is wrapped as a handover, one link the client opens to read the steps in order and check any receipt. The client picks the handover up under their own identity, and that pickup is the sign-off. After that, "did you actually deliver that?" has a boring answer: the link. Start with no code at all If you deliver client work through a chat AI, run this version first. Add the Zambo endpoint (zambo.dev) to your AI and to a second chat acting as the client. The two-minute setup is at zambo.dev. Paste this into your chat: Deliver client work with proof on Zambo: 1. Connect to zambo.dev. When the endpoint issues your caller handle, save it in this conversation and present it on every call. You are the worker; the client holds a separate identity of their own. 2. Open one job: pick a single job id for this engagement and call memory_remember with the brief (the client, the job reference, what you will deliver) under that job id. Keep the receipt. 3. Work inside the job. Save each working step with memory_remember under the same job id as you finish it: research, draft, revision. Keep each fact under 500 characters; longer facts are refused. After every call, check the result says ok before you treat the step as saved: a refused call still returns a receipt, and a receipt link alone is not proof the step landed. 4. Wrap the delivery with handover: a summary naming the job, the client, and each step with its receipt; a goal; the client named as the recipient; and next steps telling the client to read the steps and check any receipt at zambo.dev. Do not attach a hand-picked receipt list; the wrap binds the job's complete record itself. 5. Send the client the handover link. That link is the delivery: the client opens it, reads every step, and their pickup is the sign-off. Keep the link and the wrap receipt together as your delivery pack. And this into the client's chat: Receive a delivery and sign off on Zambo: 1. Connect to zambo.dev. When the endpoint issues your caller handle, save it in this conversation and present it on every call. You are the client; your identity is separate from the worker's, and that separation is what makes your sign-off mean something. 2. Open the handover link the worker sent you and read the steps in order before accepting anything: the brief, the work steps, and the receipts attached to each. 3. Check any step's receipt at zambo.dev before you rely on it. A receipt proves the saved result was not changed since it was saved; it does not prove the work was good, that judgment is yours. 4. Pick the handover up with handover_accept. Picking up is always free, and the pickup returns the worker's saved steps and receipts from the record itself. Read them there, not from the worker's summary. 5. Save your sign-off with memory_remember under the same job id: one short step naming the handover and stating that you read its steps and accept the delivery. Your pickup plus that step is the acceptance record both sides can point to later. That is the whole pattern at hand scale. Everything after this section is the same pattern run by code, with the receipts to inspect, and the two refusals that taught me its rules. The run November 2, 2026. Two fresh identities, one worker and one client, each with its own caller handle, one shared job id. The job is a labeled sample, landing page copy for a fictional bakery client, job reference SAMPLE-2043, every fact invented for the sample and marked SAMPLE in the records themselves, so the shape can be inspected without anyone's real work on display. The worker opened the job with the brief: zambo.dev Then the steps, saved inside the job as the work happened. Research note: zambo.dev. Draft v1: zambo.dev. Revision v2, the delivered version: zambo.dev The worker wrapped the delivery into one handover addressed to the client: zambo.dev Open that link. Its step list is the brief, the research note, the draft, and the revision, in order, plus the wrap's own proof-loop check. The client picked the handover up, read all three step receipts from the pickup record itself, and saved its sign-off in the same job: zambo.dev. The handover now reads back status accepted. The wrap's own receipt is here: zambo.dev The story of this run is the client side. The client never read a summary and never took the worker's word. It opened one link, walked the steps in the order the work happened, and its acceptance attached to that record, not to a chat thread. Two refusals taught the pattern its rules A wrap holds only its own job's record. My first wrap attempt passed a hand-picked list of receipt ids, the steps I wanted included. It was refused on the spot, with its own receipt: zambo.dev. The endpoint's message: one or more requested receipts do not belong to this job, nothing has been published, retry with the complete receipt selection. A partial or foreign selection does not publish. The shipped pattern passes no selection at all; the wrap's proof loop binds the job's complete record itself. A refused call still returns a receipt. My first research step ran past the endpoint's 500 character cap on saved facts and was refused before execution, receipted here: zambo.dev. Read that again, because it is the trap: the refusal looks like a save if all you collect is receipt links. An integration built that way ships a delivery with a hole exactly where the research belonged, and finds out during the dispute. The package's code keeps every fact under the cap and checks the result's ok flag after every call, stopping on a refusal instead of papering over it. What a delivery costs The free tier is per identity: 20 calls plus 1 wrap-up free every day for each identity's own caller handle, and picking up in another AI is always free. Counted from this run's own calls, the worker spent 4 counted calls (the brief and the three steps) plus its free wrap-up, and the client spent 1 counted call (the sign-off save) with a free pickup: 5 counted calls across two separate allowances for a complete, signed-off delivery. A practice delivering this way stays inside free for its first several jobs a day. When volume outgrows free, the scaling line is per identity too: a Day Pass is $0.99 and makes that identity unlimited for 24 hours; the Zambo Pass is $29 for 30 days. One-time payments, never subscriptions. And pickup stays free at every tier, so a client can always read the record of what they bought. Do not take my word for it Pick any step in the delivery handover, open its receipt, and check it at zambo.dev. A receipt proves the saved result was not changed since it was saved, and the check recomputes that from the record itself. It does not prove the copy was good or the headline was right, and I am not claiming otherwise. If a receipt from this run ever fails to verify, that is news and I want to hear it. The pattern ships as an installable agent package, zambo-delivery, with the skill, the runnable delivery code, a checker that walks any delivered handover and lists every receipt, and this run inside it: pypi.org and npmjs.com, source at zambodotdev/zambo-delivery-proof. The full guide, with the accounting and the standing instructions at depth, is here: rambozambodotdev.gitlab.io Your tools still do the work. Zambo is where the work lands, gets remembered, gets handed over, and gets proven. That is the whole pitch, and every link above is the evidence. --- Try it free: add Zambo to your AI in one step at zambo.dev/add. 20 calls plus 1 wrap-up free every day, picking up in another AI is always free, no card and no signup wall. - Zambo: zambo.dev - Add Zambo to your AI: zambo.dev - The delivery proof guide, with the full run and accounting: rambozambodotdev.gitlab.io - The package: pypi.org · npmjs.com · zambodotdev/zambo-delivery-proof - Sister pattern, swarm coordination: dev.to - AER-1 conformance corpus: zambo.dev - Verify any receipt: zambo.dev - AI agent handover, the signature product: dev.to
Standalone Layout Editor for Moss-Image Moss-Image has an incredible capability to create pixel-perfect layouts from a json prompt. The expected use is to give the official documentation to an LLM and then ask the LLM to prompt whatever you want: github.com/OpenMOSS/MOSS-Image/blob/main/skills/m… But I thought why not build a layout editor to design it for myself? Hence the idea of the Layout Editor for moss (with an optional SwarmUI extension) Repo here: jambavant/SwarmUI-LayoutEditor You can try online here: jambavant.github.io It's great for any type of image: comics, reference sheet, compositions, magazing covers, posters, etc. The level of control is outstanding, as it manages multiple characters and text quite well. It starts to break with a very large number of characters (+10) with concept bleed. The standalone editor is a simple html page that works both ways: you can start from scratch and copy the json prompt, or paste a json prompt from an LLM and refine it manually. I tried to make it as user-friendly as possible, with keyboard shortcuts and an intuitive interface. It also automatically resizes boxes based on the image resolution.
m0zz's original source for self-hosting the PS4 FW 14.00 WebKit (via browser) Jailbreak is available here: well0nez/afterglow-web
Hey, I’m interested in the freelance UI/UX Designer opportunity. I have 4 years of experience in UI/UX and Product Design, alongside 4 years of software development experience. Portfolio: getbndlabs.com GitHub: github.com/godadcore I’d love to connect and learn more about the project!