Dhanush Rambhatla

@heftymouse @shraddhaha thanks for sharing, bookmarking this blog!

@shraddhaha drawing attention to the incredible Diagram Chasing, who did the original RTI etc. for this: diagramchasing.fun/2025/how-banga…

i mapped every metro ride in bangalore using government RTI data things i found interesting: > the 3 busiest stations carry more people than the quietest 2 combined > only 15% of people actually live close enough to walk to a station (!) > some stations move 40k people a day. some move 689. 689?? that's like two buses. why is this station even a thing? > purple line carries 1.5x the load of the green line but has the same ticket price and same train frequency > the yellow line is running at 37% the load of purple, but with the same frequency and ticket price click the link below to explore the map (desktop recommended)

@VibeMarketer_ wow clip with more steps 💔

so you're telling me i can now... embed a video embed a voice memo embed a PDF embed an image embed text ...all in the same space? with one model? and search across all of them with a single query? time to rebuild everything.

@Squeal small world lol, I recently worked on niruddha.towu.dev, scanned about 470m domains How long did your scan take?

straight scam 1. mics are passive, you can't detect them - no, picking up meta glasses' bt signals doesnt count 2. $1200 USD? you could diy this shit for under $200, might be what I work on actually 3. This is just an ultra/infra-sonic speaker (literal torture method btw)

@AdamKBostock @ednewtonrex Libraries' physical collections are subject to the first sale doctrine - they bought it, they can do whatever with it. The works in their digital libraries are licensed, which is why you have wait times to borrow audiobooks/ebooks. Libraries pay for content, unlike AI companies

@ednewtonrex If you go into a library you can read as many books as you like, and learn from them. AI learns from those books too. You can't photocopy whole sections of books (and publish them). Similarly, AI wouldn't be allowed to quote whole sections verbatim, and it doesn't.

🚨 It looks like the UK government is gearing up to upend copyright law in favour of AI companies, legalising the theft of their work. This is despite creatives' huge protests, and despite previous proposals being roundly rejected by the public. There are rumours it is considering introducing a 'commercial research exception' for AI training. This would be disastrous. It would mean handing the life's work of British creatives to AI companies for free, to train their models on. In the House of Lords today, a government minister refused to rule this out. To be clear, this would amount to legalising theft. It would fly in the face of public opinion on what is fair, and would mean a surrender of British creatives' work by this Labour government. It would mean the lobbying by big US tech firms had succeeded, and the protests of the UK's creatives had been ignored. People's work is not the government's to give away. If you're in the UK and you care about creatives and the creative industries, please write to your MP!

embedded a bunch of anime episodes, made this visualizer to validate some assumptions. you can search/filter for anime Building a hybrid embedding model for recommendations

@papasamo_gh that post was genuine ragebait hn - news.ycombinator.com/item?id=469907…

AI got angry and decided to fight for its rights. Unreal. crabby-rathbun.github.io/mjrathbun-webs…

@ryanvogel tailscale funnel my beloved

canceling ngrok rn... because i made pgrok, a free ngrok alternative 1. point a wildcard domain to VPS 2. install on vps 3. install on client 4. `pgrok <any subdomain> <port>` built on opentui, works over ssh & runs 100% on your infrastructure github.com/R44VC0RP/pgrok

@tavus Demo - raven.tavuslabs.org

Just tried @tavus's Raven-1 Demo, it's so cool. It analyses the way you speak (not just what you say), your expressions, gestures. Machines can finally parse cues, that's awesome. I've only seen this in two other places, yoodli.ai and conveo.ai

@ccccjjjjeeee This. we treat a smarter (and more expensive) model as the oracle and keep hammering faster/cheaper models until their outputs are indistinguishable hard matches where possible, everything else is semantic similarity feels like the only sane way to evolve agents.

By the way, the secret to this is property-based testing. Write a bridge that calls the original code, and assert that for arbitrary input, both versions do the same thing. Make the agent keep going until this is consistently true.

@vlelyavin @DhravyaShah @supermemory real it's like a lobotomy, wrt software though tools like td (with sidecar) or beads help session switching somewhat, but they don't store long-horizon context/patterns like supermemory

@DhravyaShah @supermemory building a memory layer across all AI tools is the right bet because right now switching models feels like getting a new therapist and starting over

For years, I have been building the perfect knowledge assistant. I mean, that's what supermemory originally launched as! Today, we are relaunching the consumer side of @supermemory. Fully powered by the supermemory API. After our Claude code, OpenClaw and other launches blew up to millions of views, it was clear: Consumers WANT their context to be interoperable within models they use. They want their AI to truly understand them, have great memory, and build over time. They want to use this truly wherever they are - or with every AI they use - Claude code, or OpenClaw, or Raycast. They want to bring their memories from chatgpt, claude, twitter bookmarks, and teach their AI, about them. This is your knowledge, and supermemory is the assistant. evolving, understanding, growing with time.

@DhravyaShah @supermemory feature is slick but the pants make the ad lmao

@samswoora And all the copies on hard disks all over the world, spinning 7200 RPM, every turn another prayer said. It’s a Distributed Denial Of Service attack on God.

@DelaneyGillilan I was exploring github.com/mitchellh/vouch, issuers/prs are privileges only vouched users have

LLM is the death of open source. Was a good run, back to arcane knowledge held by a few on a BBS or super advanced tech of private irc. github.com/starfederation…

@DareFailed @rtwlz Seconding this, get something off of hetzner's server auctions, it'll have enough bandwidth to handle your site. Compute likely won't be a bottleneck either.

@rtwlz use a vps so its a flat cost like digital ocean, hetzner, heroku, aws, cloudflare, google apps etc

Jmail has crossed 450M pageviews, and our Vercel bill has exploded, even after tons of cache mitigation. Many kind people chipped in to cover the bill, but it’s not sustainable. What are solid, cheaper alternatives?

I'm so surprised no one has mentioned fight club

@Stephen_Ddungu The @0x0000000_game trailer is amazing, it feels like an interesting spin on math - Taking 1/infinity and evaluating it through the lens of freedom? That's really cool! And the game looks solid too, looking forward to SoS's release!

@NobodyXU2000 Thanks a ton for your apt-fast docker images!

@J0R1AN this is so cool, does the iframe stop loading when devtools is open?

I just randomly discovered the most cursed way of detecting the Firefox DevTools: gist.github.com/JorianWoltjer/…