HN user

pbd

169 karma

Software Engineer - https://www.linkedin.com/in/prabodh-agarwal/

Posts31
Comments44
View on HN
github.com 1mo ago

GoLongRL: Capability-Oriented Long Context RL with Multitask Alignment

pbd
1pts0
kenn.io 1mo ago

Agentic systems for what comes next

pbd
3pts0
www.youtube.com 2mo ago

Nvidia's Jim Fan on the End Game for Robotics [video]

pbd
3pts1
github.com 3mo ago

Create polished, pro-grade screen recordings – MIT Licensed

pbd
2pts1
github.com 6mo ago

Search by window title, app name, or text visible inside your windows

pbd
1pts0
github.com 6mo ago

A simple, efficient BPE training implementation in Rust with Python bindings

pbd
1pts0
alignment.anthropic.com 6mo ago

Bloom: An open source tool for automated behavioral evaluations

pbd
2pts0
github.com 8mo ago

LLM assisted book reader by Karpathy

pbd
2pts4
downdetector.in 8mo ago

Downdetector Is Down

pbd
4pts1
github.com 8mo ago

ADBC Driver Foundry

pbd
1pts0
github.com 8mo ago

The great convention of deep learning thinkers

pbd
2pts0
github.com 9mo ago

Deep learning framework built from the ground up in pure Go

pbd
1pts0
twitter.com 9mo ago

GitHub Action auto-redaction of Bearer Token malforms JSON payload in uv release

pbd
18pts1
www.youtube.com 9mo ago

A Tokyo commuters' poetic take on life [video]

pbd
3pts0
haveibeenpwned.com 9mo ago

Vietnam Airlines Data Breach

pbd
74pts34
www.generalcatalyst.com 9mo ago

Software for Hardware – Digitizing the Physical World

pbd
1pts0
alvinwan.com 9mo ago

How to tile matrix multiplication (2023)

pbd
82pts9
www.youtube.com 9mo ago

Visual representation of tiled matmul in CUDA

pbd
2pts0
docs.aws.amazon.com 9mo ago

AWS washing their hands this halloween

pbd
1pts1
twitter.com 9mo ago

Modular beat Nvidia's cuBLAS kernels on B200s in 170 LOC

pbd
3pts1
www.olcf.ornl.gov 10mo ago

Oakridge National Lab CUDA Training Series

pbd
5pts0
twitter.com 10mo ago

Lack of observability in FB chat births ReactJS

pbd
14pts1
news.ycombinator.com 10mo ago

LinkedIn seems to be leaking Google Docs

pbd
6pts5
www.youtube.com 10mo ago

Sutter Hill Ventures' Mike Speiser Creates Great Companies [video]

pbd
1pts0
news.ycombinator.com 1y ago

Ask HN: Is there any LLM based developer contributing to OSS project?

pbd
3pts0
news.ycombinator.com 2y ago

Ask HN: How to Create Snowflake Iceberg Tables on Cloudflare R2

pbd
1pts0
lists.apache.org 2y ago

Apache Datafusion Comet

pbd
1pts0
github.com 2y ago

Show HN: Ray Data Snowflake Connector

pbd
1pts0
github.com 3y ago

Show HN: Benchmark two strategies for a pivot operation in Snowflake

pbd
1pts0
medium.com 3y ago

Cooking with Snowflake

pbd
1pts1

RL researcher says LLMs won't work, LLM researchers say RL won't scale. Meanwhile, the rest of us are just shipping products with whatever actually works today. This is like watching database academics argue about ACID properties while MongoDB just took over the world.

Finally, a game that accurately simulates my daily productivity. I open it intending to play for 5 minutes and somehow 3 hours later I'm still there, having accomplished nothing useful, with a vague sense of dread and the feeling that demons are chasing me. The verisimilitude is uncanny.

I keep seeing Podman mentioned as a Docker alternative, but I'm unclear on when the juice is worth the squeeze. For someone doing typical web development (Node.js/Python services, Postgres, Redis), what specific problems would Podman solve that Docker doesn't? Is this more about security/compliance or are there developer experience benefits too?

From a systems design perspective, $3,000 per book makes this approach completely unscalable compared to web scraping. It's like choosing between a O(n) and O(n²) algorithm - legally compliant data acquisition has fundamentally different scaling characteristics than the 'move fast and break things' approach most labs took initially.

Eels are fish 11 months ago

I love these 'wait, what!' moments in biology. Thanks for sharing this - definitely going to be my fun fact for the week!

The power efficiency is fascinating - modern phones are basically ARM servers optimized for battery life. Pixel 5 probably draws <5W under load vs 50-100W for a typical x86 server. For a personal blog, that's 400-800 kWh/year savings. The environmental impact of reusing vs recycling electronics is under appreciated.

The 'dilution effect' is real - even with plenty of context space left, agents start losing track of their original goals around the 50k token mark. It's not just about fitting information in, it's about maintaining coherent reasoning chains. Single agents with good prompt engineering often outperform elaborate multi-agent orchestrations.

The timing mismatch is crucial - data centers can be built in 12-18 months, but new power generation takes 5-10 years minimum. We're essentially trying to scale AI demand faster than energy infrastructure can physically respond. This creates interesting arbitrage opportunities in power-rich but compute-poor regions.

GPT-4 at $24.7 per million tokens vs Mixtral at $0.24 - that's a 100x cost difference! Even if routing gets it wrong 20% of the time, the economics still work. But the real question is how you measure 'performance' - user satisfaction doesn't always correlate with technical metrics.

This data is incredibly valuable for both AI companies and publishers. CF gets unprecedented visibility into who's crawling what, when, and how much. Wouldn't be surprised if this becomes a premium product - 'pay for priority bot verification' or 'detailed crawl analytics.

The binary file problem is real - audio files balloon Git repos fast. Git LFS helps but adds complexity. I've had better luck with a hybrid approach: Git for project files/metadata/stems, cloud storage with versioned folders for final mixes. Keeps the repo lean while maintaining the branching workflow benefits.