Bhaskar Deo

How Massachusetts' coldest winter in 15 years helped us beat Meta FAIR and Google DeepMind This winter I didn't pay a single heating bill. At Verda, we've been training watermarking models that survive everything the real internet throws at them — re-encoding, screenshotting, sharing, filtering. Building this at a shoestring budget meant getting creative. My solution: heat my apartment with GPUs. The GPU cluster ran 24/7 from my apartment all winter. Massachusetts then decided to have its coldest winter in 15 years. Didn't matter. The apartment was warm. And I accidentally discovered the world's simplest ML monitoring system. How cold I woke up every morning told me exactly how efficient the training and validation runs were the night before. Warm apartment, good run. Freezing apartment, something crashed at 3am. I didn't even need GPU monitoring 📷 Here's the part that still surprises me. What came out of that apartment, built by a small team at a fraction of the budget and some GPU credits, outperforms Google DeepMind's SynthID and Meta FAIR's AudioSeal and VideoSeal. Not just on benchmarks. Demonstrably better. Their watermarks leave artifacts. Ours don't. See the examples below. And it's not just a research project. It's a product anyone can use today. Link in the first comment. Verify post at: verda.ai/Vo06ni.

Introducing Overlay: the open source AI operating system Work with the best models Put all your context from memories, files and outputs Run agents and automations Generate images and videos All in one platform Become AI-native today Zero data retention. Open source.

introducing Parafield, the spatial sound intelligence. the FIRST agentic platform that combines voice, music, and sound effect, to create a personalized 3D sound experience for you. if you're interested in sci-fi, arts, and/or spirituality experience, this is for you. comment "sound" to access early beta. watch it with 🎧

@timwangyc @seedtosunflower @Joshuabrowder PONDER

Introducing Ponder: the agentic video editor. It’s a new paradigm for filmmaking, where powerful creative agents and humans collaborate to tell world-class stories. We're also announcing our $2.5M pre-seed, led by Liu Jiang from Sunflower (@seedtosunflower), with @Joshuabrowder and @MattHartman. Joined by @levie (Box), @emerywells (Frame), @JaredLeto, @CommaCapital, the @nyuniversity venture fund, @cory, @darian314, @shiffman, and many more incredible founders, investors, and creators.

The next billion-dollar founder has 9 followers on X rn. I will find you & fund you!

@1_2MOA Well this was 3 months + process. And there several night I did wake up cold.

@bhaskardeo1 ‘Coldest winter…Didn’t matter. The apartment stays warm’ Next paragraph- …when I wake up freezing = something went wrong at 3am. Is this a joke?

How Massachusetts' coldest winter in 15 years helped us beat Meta FAIR and Google DeepMind This winter I didn't pay a single heating bill. At Verda, we've been training watermarking models that survive everything the real internet throws at them — re-encoding, screenshotting, sharing, filtering. Building this at a shoestring budget meant getting creative. My solution: heat my apartment with GPUs. The GPU cluster ran 24/7 from my apartment all winter. Massachusetts then decided to have its coldest winter in 15 years. Didn't matter. The apartment was warm. And I accidentally discovered the world's simplest ML monitoring system. How cold I woke up every morning told me exactly how efficient the training and validation runs were the night before. Warm apartment, good run. Freezing apartment, something crashed at 3am. I didn't even need GPU monitoring 📷 Here's the part that still surprises me. What came out of that apartment, built by a small team at a fraction of the budget and some GPU credits, outperforms Google DeepMind's SynthID and Meta FAIR's AudioSeal and VideoSeal. Not just on benchmarks. Demonstrably better. Their watermarks leave artifacts. Ours don't. See the examples below. And it's not just a research project. It's a product anyone can use today. Link in the first comment. Verify post at: verda.ai/Vo06ni.

sleep trackers are useless. here's what we built instead.

day 11 @fdotinc - talked to more aquaculture farms - improved our model - reviewed footage & edited video from farm visit chugging on the quest to build autonomous robot divers 🤿 here's a snippet of what our robot saw underwater! eel grass 🌿 + fishes 🐟 hope we didn't scare the fishies away :< @ruslanjabari @mdjshin @LerWilson

Brands are spending money to prove that their ads are not AI. adswithbenefits.in/brands-are-spe…

i built something called foyer. a voice agent that lives on any website. it talks to visitors, walks them through your pages, and grabs leads mid-conversation. like a salesperson that never sleeps.

MCP gives agents a protocol. It doesn't give them a phone book. mcp.waypoint.ing is the phone book.

@eth_chainId youtu.be/BK_IzleW60I?si…

Somebody’s building Calendly for AI Agents. Instead of OOO, they auto-reply: 🌵 OOT — Out of Tokens “Will be back next month after my owner limit reset.” The future is exhausting… and rate-limited. 💸

The hiring signal is broken. Today I'm rebuilding it. AI tools score a Python course the same as a founding engineer who shipped a YC backend. Recruiters fall back to 5 seconds per CV. Introducing 10xTable.com, the platform to never miss a top candidate again

If you think accepting is a headache, this is a fix.

random convo at @fdotinc: what if we replaced hermes agents tool calling with classic ml inference? turns out it works well - like, ~80% of the llm calls were basically deterministic, and we got a ~50% cost reduction e2e

Beautiful weather to build #SFBayArea, #SiliconValley, #SFTech @fdotinc @tash_2s Verify at: verda.ai/EyFnHb

@g0da_s Love it !

your sleep tracker tells you you slept like sh*t. neurex actually does something about it.