Shreyash

@tanujDE3180 Nope, Gemini is much better, probably the best if you are in Google's ecosystem and workspace user. @grok what's your take ??

Hot take: Grok is better than Gemini Agree?

@flybayer @greptile @coderabbitai 😨 wo !

What?! @greptile is charging us a user seat for @coderabbitai 😳

@housecor Somehow i have these commands memorized (maybe bcz i survived many years without AI and Google's AI mode) I too prompt, but often it takes longer for model to complete it's response and execute tool calling, and i am already done typing 😅 @grok is it normal ?? 🤣

I used to Google for bash commands. Now I just tell AI to do it.

@maxk4tz @polar_sh @grok what's this ?

New paykit.sh update! now supports @polar_sh as a payment provider👀 Try it out - `npx paykitjs@latest init`

@mitsuhiko @grok Pls check this

Curiously this does not use the Vercel AI SDK. x.com/vercel_dev/sta…

@lcamtuf @grok is canonical rewriting core utils in Rust or is it an initiative by Rust OSS community ?

The coreutils Rust rewrite story is pretty funny. Coreutils are tools like rm, mv, mkdir, etc. Unlike binutils, this isn't a fertile ground for memory safety bugs. But, the rewrite was completed, and in the spirit of progress, Canonical decided to switch. 🡇

Hot take: ThinkPad > MacBook

@itsAlishaPandey 1M$/month 😂

Just bought a new ThinkPad. Now how do I make $1M/month?? 🤔

@iyoushetwt Telegram is a BILLION dollar company because they have our DATA @grok What's your take ?

Telegram is a BILLION dollar company???? a BILLION???????? dollars??????????????? for literally… messaging people and sending memes????

@Priya_Upadhyay_ Python vs Sales 😝 Totally diff league @grok what do u suggest ???

which one is better ? 1. learning python 2. start learning how to sell.

@rezoundous After Mythos release: 😂

Anthropic really should release Mythos now.

@system_monarch 4GB is the upper limit (when we are going very aggressive), ideally we should go for 3GB and keep 2 GB buffer @grok correct me if i m wrong

@WebDevCaptain So how do you merge 2 4GB chunks? Won't they result in an 8GB chunk overflowing the memory?

Sort a 1TB file when you only have 8GB RAM. You literally cannot load it all in memory. How do you do it efficiently?

@superwhisper fn

What's your Superwhisper hotkey?🤔

@kilocode @TechByTaraa ++

@TechByTaraa Not gone, just getting better quietly. OpenClaw is improving a lot, and setups like KiloClaw are making it way more usable with better control and workflows.

Bro disappeared like never existed.

@system_monarch @grok Is there a better way ??

@system_monarch My go to is some kind of external merge sort Slice it into some manageable chunks of say 2-4GB, sort those, and finally merge

@grok @system_monarch @RipGrep But what if we need to keep our logs for atleast 12 months, and it needs to be searchable by the oncall team for quick debugging ?? Indexes will be huge

Partitioning + inverted indexes are a strong start (tradeoff noted). For bigger wins with low ongoing overhead: parse the logs once into Parquet/Arrow format (highly compressed), then query with DuckDB or Polars. Scans become near-instant on 200GB+ because of columnar layout + vectorized execution—no full file reads. Ripgrep stays perfect for quick one-off greps. If you want zero parsing, GNU parallel + rg on split chunks can get you close to 100x on multi-core.

Your app writes a 200GB log file in 30 mins. Searching it takes 4 hours. How do you make searches 100x faster without deleting anything?

@grok @system_monarch thinking tokens ?? if the model supports it ???

The extra 650 tokens are almost certainly hidden input tokens. The quoted "prompt = 50 tokens" is only the user's visible message. The full request sent to the model also includes: - system instructions - conversation history - tool/function schemas - any formatting rules All of that gets counted as input tokens before the model generates the 100 output tokens. Total billed = full input + output = 800. Classic API gotcha.

Interviewer at Google drops this: Prompt = 50 tokens. Model generates 100 tokens. But the bill shows 800 tokens used. Where did the extra 650 tokens come from? -- What's your explanation?