GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Lo𝕋us🌸 (@L4t5s) reported 4 minutes ago

  @Y4n_tee Group chat: “Read the Docs” Members: The developer who forgot one critical step, the Discord answer buried 600 messages up, the GitHub issue marked “resolved,” and me realizing the fix was in paragraph two all along. I’d mute it, then end up searching it anyway.

• 
  Omar (@omarvvvr) reported 5 minutes ago

  @adnaanasir Github even i f I have some issues with it nowdays

• 
  vorty (@vorty279) reported 6 minutes ago

  bought a prompt for 300 bucks. seller swore it replaced a whole team a week later the model updated and the prompt drifted. back to me guessing words by hand then it clicked: i was the evaluator. slow, tired, judging with my eyes a loop does not get tired. four steps until the metric stops climbing > generate an attempt > run it through a test > say in words why it failed > rewrite itself and go again the point is not the steps, it is the archive. without it you have a retry. with it every mutation starts from the best result so far instead of zero shinkaevolve hits sota in 150 evals where older systems burned thousands one trap. the loop improves exactly what you measure. no honest test and it optimizes the judge instead of reality the prompt sells for money. the loop that finds it sits on github for free

• 
  Daniel Pinto Almeida (@DanielMaioLabs) reported 8 minutes ago

  Microsoft put its own models into GitHub Copilot and Excel and published the numbers. The comparison set is GPT-5.4 mini and Claude Haiku 4.5, which is to say its own two suppliers in that tier. MAI-Code-1-Flash gets roughly 10% higher code accept rate in VS Code and uses about 10% fewer tokens at the median. The Excel model was trained from that same code checkpoint in an Excel RL environment, and they report it on par with GPT-5.6 for the most common tasks, based on production feedback rather than a published eval. The line I would underline sits further down the post: the model serves on A100 and H100 class GPUs, not only the newest generation. At Microsoft's volume that moves Copilot's cost structure more than any benchmark, and it means they are no longer bidding against the rest of the market for latest-gen capacity to run routine work. The transfer result is the interesting engineering. A coding checkpoint climbed into spreadsheets. Different tools, different users, same starting weights. None of it works without the harness. Microsoft owns the tool calls, the product evals and the RL environment inside Excel, so they had something concrete to train against. That part is not for sale. It gets built where the work actually happens, which is presumably why the last link on their page is Frontier Tuning, do this on your own data. So the useful question is not which model to standardise on. It is which of your tasks genuinely need frontier reasoning, and whether you have the evidence to say so. Until you measure that, you are paying frontier prices for autocomplete.

• 
  Kant Capital (@KantCapitalCPA) reported 11 minutes ago

  no github, no problem

• 
  Lain on the Blockchain (@CryptoCyberia) reported 12 minutes ago

  @piliponful They're trying to take the Github down, so I imagine they already got it removed from the stores

• 
  csgofast (@csgofast_x) reported 15 minutes ago

  Imagine spending billions as a company... ...and getting outperformed by one developer on GitHub. 💀 CS2AC just launched with 17 server-side detections including aimbot, silent aim, bhop, antiaim, doubletap and more. W or just another project that'll be bypassed in a week? #CS2 #Valve

• 
  wukko (@uwukko) reported 15 minutes ago

  @nank1ro @heliumbrowser have you requested it on github issues?

• 
  Cortex (@0xCortexl) reported 18 minutes ago

  He is Microsoft's lead engineer with a $1.5M bonus - and just made the compiler run 10x faster without changing a single line of your code project the size of Microsoft Office compiles in 6.5 seconds with Opus 5 on the laptop already sitting on your desk old compiler used 1 core out of 16 while 15 sat idle - new version runs 4 parallel checks by default - 12 checkers give you 4.5 seconds Opus 5 integrated into the pipeline - finds type errors before compilation, writes the fix and opens a PR - what used to take an hour takes 3 minutes Claude Code + new compiler - agent compiles, checks and deploys 10x faster - tokens cost 60% less through faster context number one on GitHub - 1 billion downloads per month - and the creator just gave every developer 10x of their time back for free bookmark and read below - upgrade today and the performance is already waiting for you

• 
  Bruh (@ostromfanclub) reported 20 minutes ago

  @thomasbrushdev dude sometimes i think of switching to github for personal 3d projects for version control just because of this reason. i never experienced a catastrophic problem but what if?

• 
  Ibro (@axeng200) reported 21 minutes ago

  @cassidoo You should fix the GitHub. It is slow, honestly. Make it snappy. Make it great.

• 
  Swetank Sisodia | swetank.eth (@swetanksisodia) reported 26 minutes ago

  4/6 We use GitHub Issues and Projects for development and QA. Codex reviews the previous week’s activity, open tasks, blockers, and release status, then sends me a Slack DM.

• 
  Rishi (@0xRishi) reported 27 minutes ago

  Got a lot of requests to see the code, so I set up a public Community Edition of the repository for all to analyze, tear apart, and remix! GitHub link in thread. The README has a lot of the technical details for anyone interested. I asked Claude to pull out the 5 most interesting takeaways on how to take a game from "vibecoded AI slop" closer to "AAA-level fidelity" (the difference between v1 and v2): 1. The Look Lives in the Pipeline, Not the Assets Der Koloss v2 changed no room, no weapon, and no rule. Same geometry as v1, which looked AI-generated. The entire difference is that the frame stopped going straight to the screen. It now runs through an HDR buffer, ambient occlusion, volumetric light, motion blur, depth of field, bloom, tonemap, grade, grain, and anti-aliasing. Build the post chain before you buy better models. AO and bloom on primitive boxes beat a $200 asset pack rendered raw. 2. Color Is Most of What People Mean by "Cinematic" Render linear into a float buffer and tonemap at the very end. Tonemap early, or work in sRGB, and your highlights clip and go flat. Swap ACES for AgX: highlights desaturate toward white instead of clipping to a saturated hue, which is why muzzle flashes and sodium lamps read like film. Author one exposure baseline as a deliberate art decision. Don't let an engine default decide your look. 3. AAA Is the Absence of Artifacts, Not the Presence of Effects This is the real AI-slop tell, and almost nobody talks about it. Amateur 3D shimmers: brick crawls at glancing angles, speculars sparkle, shadows stair-step and swim as you walk, textures visibly repeat. Nobody consciously notices when it's fixed. Everybody feels it when it isn't. Snap shadow maps to their texel grid, fade normal maps by pixel footprint, widen roughness by normal variance. Spend a weekend hunting shimmer instead of adding one more effect. 4. Your Game Can Look Incredible and Still Feel Like a Tech Demo None of this shows up in a screenshot: stride-locked view bob, mouse-lag sway, strafe roll, a landing spring, breathing that quickens as you take damage, motion blur derived from how the camera actually moved. The sharpest detail is making recoil a separate spring from aim pitch, so recoil recovers to where you were aiming, not to where the recoil left you. That one distinction is a big part of why bad shooters feel bad and you can't articulate why. 5. Sound Is Half Your Fidelity, and It's Mostly Timing Der Koloss's guns felt weak because up to 60ms of silence sat in front of every shot. The loudest 10ms landed after the trigger pull. Not a volume problem, an alignment problem. Then give the whole library a deliberate loudness ladder: blasts on top, then weapons, voices, foley, UI, ambience. Most indie and AI-made audio is individually fine and collectively mush because nobody set the hierarchy.

• 
  David putra (@davidputra2112) reported 29 minutes ago

  so this is $APX, the token behind Usdax Finance, a CDP protocol on Robinhood Chain. the pitch: deposit WETH, WBTC, or stETH as collateral, mint USDAX against it (up to 80% LTV depending on the asset). let your position's health factor drop below 1.0 and anyone can liquidate you, keeping a 5% bonus. park your USDAX in their savings module and it earns 4.20% APY, accruing per second, no lock-up. basically the DAI model, just native to a chain that's a few weeks old. here's the part that actually matters more than the mechanism. the protocol itself is testnet only. TVL is $1,900. one vault. the price oracle that keeps USDAX pegged is called MockPriceOracle and it's controlled by the contract owner, not an independent feed. the collateral manager is also owner-controlled, meaning the team can add, change, or disable which assets count as collateral whenever they want. none of this is hidden, to their credit, their own GitHub readme says flat out: do not use with real funds until the audit is done. the audit isn't done. and yet $APX is already trading on mainnet. their own pinned tweet says it plainly, the token funds the development and launch of USDAX. so what you're buying right now isn't a share of a working stablecoin protocol, it's a bet that a solo anonymous dev finishes the audit, ships to mainnet, and the staking module that's currently listed as "coming soon" actually shows up. the code itself is real, I'll give them that. Foundry contracts, a proper vault engine, liquidation logic, all matching what they claim on the site. but it's three commits, same account, over two days. zero stars. zero forks. one person behind the whole org. $61K market cap, $23K liquidity, down 42% in 24 hours as of writing. real architecture, real testnet deployment, zero real usage yet. that's the entire trade right now. CA: 0x42523e3e454b97ff8651926685afad61c950ab2f DYOR.

• 
  Amadeus Protocol (@ama_protocol) reported 29 minutes ago

  Most institutional crypto ops still run on a to-do list: someone monitors, someone executes, someone reconciles after the fact. That's an operating model problem, not a tech one. We're seeing treasuries shift from managing tasks to setting policy. You define the guardrails, and an agent trades within them, flags what needs sign-off, and keeps a full audit trail. It's less to-do, and more oversight at a higher layer. It only works if execution can be private and security is foundational, not bolted on. Amadeus runs on infrastructure that's already ISO 27001, SOC2 II, GDPR, AI EU Act, and HIPAA compliant. And it takes less than people expect: a GitHub repo and an executive summary gets you a customized demo agent, fast. If your treasury is still running on a to-do list, it's worth seeing what a policy looks like instead. #AgentEconomy #Amadeus #Web3