Sam

Time's running out

"Cursed with the burden of knowledge that he was insufficient but lacking the agency and wisdom to amend it brought him misery to no end"

Just one shotted something I would have made in 2 days a couple of years ago It's so over

Game devs can now make generational wealth just by coddling our nostalgia. Only if..

What if I fall? Oh but my darling, what if you fly? So beautifully written ✨

@BellaBaddie__ No question about it and I know I'm lucky that I can say that

Be brutally honest. In your next life, would you choose your dad to be your dad again?

YOU WILL WIN.

@de3dsoul Seeing this brings back such good memories 😭

Bro: How r u so focused even under pressure? Meanwhile me in 2006:

surely it'll work out sooner or later

Keep trying anyway

@0xlelouch_ scale horizontally to increase worker count and concurrency, batch jobs where possible + optimize redundant work if any. Separate out slow and fast jobs, add backpressure and rate limits to protect downstream systems and finally fix the architecture so backlogs don’t choke system

Your background job queue has 1M jobs but workers are only processing 10 jobs per minute. How will you speed this up?

How about you pip install some personality bro

Other than giving it time, how do people deal with heartbreak?

@rasmalai We can't be stopped

@aditya4f samsaxena.vercel.app Not even close to the incredible portfolios in this thread but it's honest work

Show me your portfolio websites. I'll give honest feedback! I'm curious what types of portfolios people build!

i understand him now,

- Temperature: This doesn't filter the pool, it changes the shape of the pool before K or P get involved. It's like a creativity dial. Lower temp (<1) makes the model conservative and deterministic. Higher temp (>1) flattens the curve, giving rarer words more of a chance.

- Top P (Nucleus, let's say = 0.8): The smart filter. "Keep adding words until we hit 80% certainty." Better for balancing dynamism, as the pool shrinks or grows based on the model's confidence.

Every Software/ML/AI engineer is working on or with LLMs rn. But too often, the sampling parameters are treated like magic numbers. At least I did until I sat down to learn what they actually meant Did you know what Top K, Top P, and Temperature do to the token probability pool?