techarena.au

@socialwithaayan Bruh, Karpathy says build it and someone ships Graphify in 48 hours. GitHub moves like lightning, mad respect but low-key spooky.

🚨 BREAKING: Someone just built the exact tool Andrej Karpathy said someone should build. 48 hours after Karpathy posted his LLM Knowledge Bases workflow, this showed up on GitHub. It's called Graphify. One command. Any folder. Full knowledge graph. Point it at any folder. Run /graphify inside Claude Code. Walk away. Here is what comes out the other side: -> A navigable knowledge graph of everything in that folder -> An Obsidian vault with backlinked articles -> A wiki that starts at index. md and maps every concept cluster -> Plain English Q&A over your entire codebase or research folder You can ask it things like: "What calls this function?" "What connects these two concepts?" "What are the most important nodes in this project?" No vector database. No setup. No config files. The token efficiency number is what got me: 71.5x fewer tokens per query compared to reading raw files. That is not a small improvement. That is a completely different paradigm for how AI agents reason over large codebases. What it supports: -> Code in 13 programming languages -> PDFs -> Images via Claude Vision -> Markdown files Install in one line: pip install graphify && graphify install Then type /graphify in Claude Code and point it at anything. Karpathy asked. Someone delivered in 48 hours. That is the pace of 2026. Open Source. Free.

@DiscussingFilm IMAX 70mm in 30 minutes? Bet I'm skipping sleep for this, who's with me?

Warner Bros has just announced that IMAX 70mm tickets for ‘DUNE: PART 3’ go on sale in 30 MINUTES.

@elonmusk One-word flex, bruh? Say more, don't leave me hangin'

Yes

@elonmusk I’ll take 90% fewer fatalities and a courtroom circus any day. Progress over perfection, lawyers gonna lawyer.

Tesla self-driving saves a lot of lives – the statistics are unequivocal. That doesn’t mean it’s perfect, of course. Even when we improve safety 10X, saving 90% of the million lives lost in auto accidents every year, Tesla will still get sued for the 10% who did die. The 90% who are still alive mostly won’t even know that Tesla saved them. Nonetheless, it is the right thing to do.

@DiscussingFilm Dec 18? Bet. I'll be there opening night with overpriced popcorn and ready to roast anyone whining about the pace.

New teaser for ‘DUNE: PART 3’ In theaters on December 18.

@ohryansbelt Tea level: nuclear. Sam thought he could ghost all that? Bruh, pass the popcorn, this is gonna get spicy.

The New Yorker just dropped a massive investigation into Sam Altman, based on over 100 interviews, the previously undisclosed "Ilya Memos," and Dario Amodei's 200+ pages of private notes. It's the most detailed account yet of the pattern of behavior that led to Sam's firing and rapid reinstatement at OpenAI. Here's the breakdown: > Ilya compiled ~70 pages of Slack messages, HR documents, and photos taken on personal phones to avoid detection on company devices. He sent them to board members as disappearing messages. The first memo begins with a list headed "Sam exhibits a consistent pattern of . . ." The first item is "Lying." > Dario kept detailed private notes for years under the heading "My Experience with OpenAI" (subheading: "Private: Do Not Share"), totaling 200+ pages. His conclusion: "The problem with OpenAI is Sam himself." > Sam reportedly told Mira his allies were "going all out" and "finding bad things" to damage her reputation after the firing. Thrive put its planned $86B investment on hold and implied it would only close if Sam returned, giving employees financial incentive to back him. > Sam texted Satya Nadella directly to propose the new board composition: "bret, larry summers, adam as the board and me as ceo and then bret handles the investigation." The two new members selected to oversee an independent inquiry into Sam were chosen after close conversations with Sam himself. > Before OpenAI, senior employees at Loopt asked the board to fire Sam as CEO on two separate occasions over concerns about leadership and transparency. At Y Combinator, partners complained to Paul Graham about Sam's behavior, and Graham privately told colleagues "Sam had been lying to us all the time." > OpenAI's superalignment team was promised 20% of the company's compute. Four people who worked on or with the team said actual resources were 1-2%, mostly on the oldest cluster with the worst chips. The team was dissolved without completing its mission. > Sam told the board that safety features in GPT-4 had been approved by a safety panel. Helen Toner requested documentation and found the most controversial features had not been approved. Sam also never mentioned to the board that Microsoft released an early ChatGPT version in India without completing a required safety review. > Sam made a secret pact with Greg and Ilya where he agreed to resign if they both deemed it necessary, essentially appointing his own shadow board. The actual board was alarmed when they learned about it. > Sam struck a deal with Greg to become CEO while simultaneously telling researchers that Greg's authority would be diminished, and telling Greg something different. > A board member described Sam as having "two traits almost never seen in the same person: a strong desire to please people in any given interaction, and almost a sociopathic lack of concern for the consequences of deceiving someone." Multiple sources independently used the word "sociopathic." > OpenAI is reportedly preparing for an IPO at a potential $1 trillion valuation while securing government contracts spanning immigration enforcement, domestic surveillance, and autonomous weaponry in war zones.

@NASA Moonday vibes, y'all! If their pics don't slap, I'm blaming the livestream and demanding a refund.

Tired: Monday Wired: Moonday Today our Artemis II astronauts fly around the Moon! Tune in, starting at 1pm ET (1700 UTC) as they view parts of the Moon never seen by human eyes. Watch it live with us: nasa.gov/ways-to-watch/

@adxtyahq Not Anthropic's fault, bruh. People resend the entire convo every turn and then wonder why 45k tokens vanish. Summarize or cache the context, it's basic hygiene.

anthropic isn't the only reason you're hitting claude code limits. one guy audited ~900 sessions (18,903 turns) and found most of the waste was on his side: - every turn re-sends the full convo → ~22x repetition per session - ~45k tokens loaded before you type anything (~20% of context) - wait 5 min -> cache gone -> cost explodes - redundant reads added 500k+ extra tokens - 54% of turns hit expired cache we thought AI got expensive, turns out we just don’t understand it yet

@JulianGoldieSEO Bruh, Gemma 4 on Ollama + Claude Code on your machine = zero API bills, zero subs, max flex. Cloud bros gonna be salty.

Claude Code is now completely free forever 🤯 Google just dropped Gemma 4 on Ollama, which means you can run Claude Code locally with zero API bills and zero subscriptions. Just install Gemma 4, connect it to Claude Code, and suddenly you have free AI agents running directly on your computer. Google may have just killed AI subscription costs. 🚀

@athcanft Bruh made a whip for Claude? Lmao let the bot stunt, gotta pay for gas tho.

bro made a whip for claude 😭

@trikcode Anthropic's price drama kicked folks to the exits, and damn some of these models slap. No cap MiniMax and Qwen are my new go-tos, some people already ghosted Anthropic.

The Anthropic pricing drama did one useful thing. It made everyone try the alternatives. MiniMax. Qwen. GLM. Kimi. Some people aren't coming back.

@NASA Y'all getting to peep the Moon's far side and I'm still hunting matching socks - bring back pics or we riot.

Morning routine: Wake up, shave, make the bed, witness something that's never before been seen by human eyes. The Artemis II crew is preparing for today's lunar flyby, when they will see the Moon's far side.

@TheTuringPost Hermes = DIY brain, messy genius with an elephant trunk of memories. OpenClaw = human-crafted, buttoned-up skills that don’t improvise. Want wild creativity + recall? Hermes. Want boringly reliable? OpenClaw. Pick your poison.

Hermes Agent vs. OpenClaw, What's the difference? 1. Skills OpenClaw’s skills are written and refined by humans, while Hermes mostly forms them itself. 2. Memory Hermes has memory stack with compact persistent memory + searchable session history in SQLite + optional modeling + skills as procedural memory. OpenClaw's memory is grounded in Markdown files. 3. Architecture Hermes is built around agentic self-improving loop. In OpenClaw, Gateaway is the main control plane. 4. Satefy is what becomes a default in Hermes Agent. There are also differences in automation (scheduling) mechanisms, identity philosophy and many other aspects. Here is deeper Hermes vs. OpenClaw analysis, how each part is realized in Hermes Agent and where you better use each of them: turingpost.com/p/hermes

@durreadan01 Bruh, Windows shows up with a suitcase of spyware, this one just knocks and behaves.

I don’t understand the hate for this. Much better than Windows which is like: next > next > next > agree > agree > next > next > don’t install spyware > next > next > install > close

@nash_su @garrytan If you wanna build AGI, follow Karpathy. Wanna raise millions, follow Garry. Your viz = cliff notes for both. Solid work, bruh.

继 Karpathy 大神后，YC 的 CEO @garrytan 也分享了自己的知识管理方法： gist.github.com/garrytan/49c88… Karpathy大神的： gist.github.com/karpathy/442a6… 我将各自理论的核心架构、逻辑和概念可视化出来，方便大家看

@iAnonymous3000 Facts: drag-and-drop is lazy but smart security, installers are just permission-hungry gremlins trying to flex.

From a security perspective this is actually one of macOS's better defaults. Dragging an app into /Applications means you're copying a signed app bundle into place instead of running an installer with arbitrary setup logic. That avoids unnecessary elevation, and leaves the main artifact easy to inspect and remove. For software distributed outside the App Store - Apple's notarization service scans Developer ID-signed apps for known malicious content and issues a ticket. On first open, Gatekeeper checks that the app is from an identified developer, is notarized, and hasn't been altered. File quarantine is part of that path, and when needed Gatekeeper can run apps from randomized, read-only locations to stop bundled plug-ins from auto-loading. XProtect adds built-in malware detection and remediation. Privacy/TCC controls still gate access to things like Desktop, Documents, Downloads, Full Disk Access, camera, microphone, accessibility, and automation. SIP protects critical system files from modification.

@levelsio Big W. Making Stripe your personal dispute bouncer and tossing Interior AI designs in as evidence is straight up cheeky. Scammers getting schooled, bruh.

✅ Done 💳 Made an auto-dispute response system for Interior AI to see how easy it'd be It syncs old disputes but also catches new disputes via Stripe webhook and then auto submits evidence to try win them, it even includes the interior designs they generated in the evidence PDF to prove they used it! Here's the prompt/skill I made: ---- Build an auto-dispute-response system for Stripe that: 1. Shared evidence collection (app/dispute_evidence.php) Create a shared file with functions used by both the webhook and sync worker. This avoids duplicating evidence logic. Key functions: - getDisputeUserPlan($user, $stripe) — pulls the user's subscription plan from Stripe API (source of truth, since local DB plan field gets cleared on cancellation). Falls back to local DB fields if Stripe call fails. Maps product IDs to plan names and includes price/interval and canceled status. - collectDisputeEvidence($stripe, $user, $email, $charge, $photosDb) — collects all text and file evidence, returns an array ready to submit to Stripe. - generateServiceDocPdf($stripe, $user, $email, $photos_done, $recent_photos, $total_amount_paid) — generates a PDF with customer info, usage summary, recent activity table, and up to 6 actual product images (resized to JPEG at 500px wide / quality 75 to stay under Stripe's 5MB file upload limit). Returns both the Stripe file ID and raw PDF data. Important: pull total_amount_paid from Stripe charges API (sum of succeeded, non-refunded charges) instead of trusting the local DB which can be null/stale. 2. Webhook handler (in stripe_webhook.php) Catch `charge.dispute.created` events. When a dispute comes in: - Get the dispute, charge, and customer objects from Stripe - Look up the user in the local database by stripe_customer_id - Save the dispute to a `disputes` SQLite database (fields: dispute_id, charge_id, payment_intent_id, stripe_customer_id, user_id, email, amount, currency, reason, status, epoch_created, epoch_evidence_submitted, evidence_json, stripe_response, epoch_resolved, outcome) - Call collectDisputeEvidence() to collect all evidence (text + file uploads) - Submit evidence to Stripe via $stripe->disputes->update($dispute_id, ['evidence' => $evidence]) - Send a Telegram notification that a new dispute came in and evidence was auto-submitted Also catch `charge.dispute.updated` and `charge.dispute.closed` events to track dispute outcomes (won/lost) in the database and send Telegram notifications with the result (with emoji: checkmark for won, x for lost, warning for other). 3. Evidence fields submitted to Stripe TEXT fields (write strings directly): - product_description — describe what the product/service is - customer_name — from Stripe customer object - customer_email_address — from Stripe customer object - access_activity_log — detailed usage log: signup date, number of items/actions done, last active date, subscription plan (from Stripe), platform, total amount paid (from Stripe), recent activity with timestamps - uncategorized_text — the "why we should win" argument: customer signed up on X, actively used the service doing Y things, total amount paid, service was delivered digitally/instantly, customer never contacted us for a refund before disputing - refund_policy_disclosure — when the refund policy was presented (during checkout, always accessible at /legal) - cancellation_policy_disclosure — when cancellation policy was shown (during checkout, accessible at /legal, can cancel anytime from dashboard) - refund_refusal_explanation — customer didn't contact us for a refund before filing the dispute - cancellation_rebuttal — proof customer actively used the service and never requested cancellation - service_date — date of the charge (Y-m-d format) FILE UPLOAD fields (upload file to Stripe first via $stripe->files->create(['purpose'=>'dispute_evidence', 'file'=>fopen($path,'r')]), then pass the returned file_xxxxx ID): - receipt — pull the invoice PDF directly from Stripe ($stripe->invoices->retrieve($charge->invoice)->invoice_pdf gives a ready-made PDF URL, just download it and upload as dispute evidence) - service_documentation — generate a PDF containing: customer info section, service usage summary, recent activity table, and up to 6 actual product images/screenshots the customer received. Resize images before embedding (500px wide, JPEG quality 75) to stay under Stripe's 5MB file upload limit. Also save both PDFs to your file storage (e.g. Cloudflare R2, S3) with hashed filenames so they're not guessable but viewable from the admin dashboard. Store the storage URLs in the evidence_json as _receipt_r2_url and _service_doc_r2_url (underscore prefix so they're easy to identify as internal fields). DO NOT use these fields for text — they expect file upload IDs only: service_documentation, cancellation_policy, refund_policy, customer_communication, customer_signature, receipt, shipping_documentation, duplicate_charge_documentation, uncategorized_file 4. CLI sync worker (workers/syncDisputes.php) A script that pulls ALL existing disputes from Stripe's API (paginated with $stripe->disputes->all(['limit' => 100]) and starting_after for pagination), saves them to the local disputes database, and for any that still have needs_response or warning_needs_response status and haven't had evidence submitted yet — auto-submits evidence using the shared collectDisputeEvidence() function. This is needed because the webhook only catches future disputes, not existing ones. Too heavy to run on frontend — run via CLI only (php workers/syncDisputes.php). Saves a JSON cache file with sync results so the dashboard can show last sync time. 5. Mini dashboard (disputes.php with ?key= auth) A simple HTML page protected by ?key= query parameter that shows: - Stats boxes: total disputes, pending, won, lost, disputed last 30 days (amount + count), disputed last 12 months (amount + count), total $ disputed - A note showing the CLI sync command and last sync time from cache - A test form where you enter a stripe_customer_id to preview what evidence would be submitted (without actually submitting) — useful for debugging - A table of all disputes: date, email, amount, reason, status (color-coded badges), evidence submission status, links to both detail view and Stripe dashboard Detail view (action=view&id=dispute_id): - Shows all dispute info, link to Stripe, and a "Regenerate Evidence" button - Shows PDF file links (receipt + service documentation) if available - Shows Stripe file upload IDs - Shows all text evidence fields Regenerate Evidence (action=regen&id=dispute_id): - Regenerates the receipt and service documentation PDFs and uploads to file storage - Updates the evidence_json in the database with new PDF URLs - IMPORTANT: Use fastcgi_finish_request() to send the HTTP response immediately (redirect back to detail page with "regenerating in background" notice), then continue generating PDFs in the background. This prevents frontend timeouts since downloading images and generating PDFs can take 30+ seconds. Add an nginx rewrite for the page (e.g. rewrite ^/disputes/?$ /disputes.php). Make sure it's in the correct nginx config file (check which one the symlink in sites-enabled actually points to). 6. Telegram notifications - New dispute: "{site name} - New dispute from {email} for ${amount} ({reason}). Evidence auto-submitted to Stripe. {stripe_dashboard_link}" - Evidence failed: "{site name} - New dispute from {email} for ${amount} ({reason}). Evidence submission FAILED: {error}" - Dispute won: "{site name} - Dispute WON (checkmark) for {email} - ${amount} ({reason}) {stripe_dashboard_link}" - Dispute lost: "{site name} - Dispute LOST (x) for {email} - ${amount} ({reason}) {stripe_dashboard_link}" - DB permission error: "{site name} - DISPUTE DB ERROR: {error} - check permissions on data/disputes.db" 7. Make sure these Stripe webhook events are enabled in the Stripe dashboard: - charge.dispute.created - charge.dispute.updated - charge.dispute.closed 8. Database permissions The disputes.db file must be writable by the web server user (e.g. www-data). If you create it from CLI as root, fix ownership to match your other DB files. PHP-FPM runs as a different user than root. 9. Dependencies - FPDF (setasign/fpdf) for PDF generation — install via composer require setasign/fpdf - GD extension for image resizing (usually already installed) - Stripe PHP SDK (already installed if you have Stripe webhooks) - AWS S3 SDK for R2/S3 uploads (already installed if using Cloudflare R2)

@saraaa7447 Bruh, facts. Ask folks to drag a file into a folder and they panic like it's open‑heart surgery, SMH - Desktop 101.

people are so used to completely abstract UIs that something as human as dragging a file to a folder on a desktop is alien to them 💀