Pranab Salian

Google engineer explained how to fine-tune a tiny LLM from 46% to 90% accuracy on your phone in 21 minutes - better than $1500 on-device AI bootcamps. pick Gemma 270M -> generate synthetic task data -> fine-tune with LoRA -> quantize to int4 -> deploy to Pixel and hit 2000 tokens per second. That loop is how a 270M model beats a 70B one on your task, running fully offline in your pocket. Gemma 270M + synthetic data + LoRA + int4 quantization + on-device runtime - that's the stack. Watch and save it, then fine-tune your own tiny agent tonight.

if i ever see a marketing email using a root domain, i always mark it as spam because clearly whoever is in charge doesn't know what they're doing.

The new development cycle in one image.

Please submit your sitemap to bing webmaster and indexnow. You will thank me later

Investigation: Banned in Europe, Scanning Faces in India Cant sell it in the EU. But Spanish firm Herta's facial recognition software is being used to scan millions of citizens across railway stations, temples and more. With @investigate_eu @_infoLibre 1/5

Instead of watching another Netflix series tonight, watch this 16-minute talk from a Jane Street AI engineer explaining how the hedge fund uses LLMs to make $30B+ every year. It’s the clearest, most practical look I’ve seen at how a top quant firm actually builds custom LLMs and agentic tools for developers, especially when public models fall short on something like OCaml. Whether you’ve never built AI tooling before or you’re already shipping agents, you’ll walk away with concrete ideas on making LLMs reliable in production. Bookmark it & watch the talk today - it might be the most valuable thing you learn about real-world AI engineering all month.

@thdxr Ok I just need a couple more days. It’s been almost finished for two years now.

Someone should make an app to record meetings with automatic AI transcripts. Underexplored space rn.

I love and hate prepping someone for a deposition. To make them understand how a deposition works, the first question I ask is, “Do you know what time it is?” Nearly every single time, the person will either check his or her watch, or look up at a clock. Then, they will say “Yes, it is 10:30.” At this point I just smile and shake my head. “No.” Then, I explain why their answer is wrong: I asked if they knew the time. Not what time it was currently. Me: “So what would your answer be then?” Witness: “Yes.” Me: “Wrong.” They didn’t know the time, because they had to look at a watch or clock. If you are ever deposed and do not have a bitch like me prepping you, remember this key: Listen to the question and answer only what is asked. And only what you actually know or recall. No charge for this advice.

Cloudflare tunnels is one of the best things to ever happen to Indian homelab community

discovered dolby atmos remasters recently and I regret to inform you that music apparently has been in black and white this whole time. even on my budget, sub $1k setup (2x sonos era 300s) albums i've heard 500 times suddenly grew a third dimension, almost making me cry. WHY IS NOBODY TALKING ABOUT THIS. switched from spotify to apple music purely for this and I'd do it again in front of daniel ek's family. its insanely good - do yourself a favor and go hear it.

Tuesday morning good news: Water stocks in the seven lakes supplying to Mumbai’s needs have touched 29% crossing four lakh million litres stock from 16.92% of 2.44 lakh million litres a day ago. Triple digit rains in all our lakes almost.

#PowaiLake used to overflow after nearly half the monsoon was over. Later Atleast a month. This time it overflows within 3 days of rains after 20 days delay in monsoon. We should be worried. When we argued with @mybmc about heavy silting we were told it is very costly! Saving environment is costly, degrading it is not. 🥲

Citizens can now report illegal hawking across Mumbai through the BMC’s WhatsApp chatbot, with the civic body introducing a dedicated complaint option in compliance with directions issued by the Bombay High Court. The new feature has been added to the BMC’s WhatsApp chatbot on the number 8999-22-8999, allowing citizens to upload the location and a live photograph of unauthorised hawking.

@plivo @bevenky what's this about?

@plivo what does one need to do to be able to sign up with you? The cx.plivo.com/signup form doesn't accept any work email address - I've tried .in and two .com so far.

@schoudhury1991 @sidhant_sarthak Generally this is a positive thing for procurement - unless there are details showing otherwise.

@sidhant_sarthak sanjibani.github.io/cppp-tender-re…

transparency needs to be accessible. from today it is. within the past 2 weeks, around 1 crore 66 lakh records were scraped from the government's CPP Portal. The Nation's procurement database is now publicly available to all. link below

@Semper_Viventem It's just a setting. Just ask it not to if you don't prefer it to do something.

Claude Code automatically adds itself to the commit metadata "Co-Authored-By: Claude Opus <noreply@anthropic.com>". I strongly disagree with crediting Claude as a commit co-author. It is a tool I'm paying for, not a copyright holder, not an author. We don't credit Visual Studio.

The wait for 'The Winds of Winter' has officially reached 5,458 days with no updates on its release George R.R. Martin published the first five 'A Song of Ice and Fire' books in exactly 5,458 days

@timh022 @guinnesschen Perhaps this depends on your tasks. I find it pays to be exceedingly precise in prompts. LLMs can give you a completely different quality of output when you change a word or two, especially on complex or technical tasks.

@prbsl @guinnesschen You don’t have to be clear that’s the whole point. The llm will still understand

Bro it’s June 2026. Stop hand editing your prompts. Hold down the dictation button and ramble for 10 minutes. Give the model every fragment, caveat, example, and vibe in your head. It is literally a large language model. If it’s superhuman at anything, it’s reconstructing latent intent from language.