was working with mediapipe BlazePose, which gives 33 pose points in world space, but wanted "the pose to always point forward" (virtually this prompt exactly)

it one-shotted 600 lines of code which did the job perfectly. it understood from context the center of body, how to calculate the body normal, to rotate each point around that, all while handling edge cases to avoid errors. would've takens me hours if not days to tweak it manually to work.

another example just from today:

I merely selected entire json file, two word prompt: "generate schema"

one-shotted 600 lines json schema, unsolicited 200 lines typescript schema, 150 lines python dataclass model and a README!!! completely unsolicited!

(cursor agent mode)

Have you checked it yet?

Last time I did that from an Open API file rather than transform it, as I was lazy, it hallucinated a bunch of properties and left a load off too.

That was a year or so ago.

a lot changed in a year

very insightful to read the comments here.

it seems quality isn't universally declining, but the variance in quality is increasing, and finding good quality is nearly impossible in the influx of product offers. reviews are fake, to most short term profit margins are more important than reputation.

"unlocking revenue stream" is wild way to say theft

capcut will fulfill all your video editing needs for the near future, music, ai dub, etc.

now you only need inspiration I'd go for fireship-style videos and just start pushing content

tailscale to expose a port is such overkill. just use ngrok for that.

For simple problems use simple tools. I believe the simplest tunneling tool out there is https://pinggy.io . Tailscale is for a different use case. Not just exposing one port to the interenet.

or one of the many alternatives - https://github.com/anderspitman/awesome-tunneling. I will advocate for zrok.io as I work on its parent project, OpenZiti. zrok is open source and has a free (more generous and capable) SaaS than ngrok.

this is amazing. I've been swearing at diffs which I now know are myers algorithm and histogram diffs are exactly what I've been wanting!!

also perplexity.ai (pro subscription)

Not rolled out on the phone app, but having the option on desktop so soon was a pleasant surprise.

this is really nice. what's your MRR? do you know anifusion?

then how do you discover new content to sub to?

"The benefits become noticeable only at p999 onwards; for p90 and p99, the performance is almost the same as SQLite."

I hate to be a hater, and I love sqlite and optimizations, but this is true.

The benchmark seems a bit weird. Fetch 100 results from a table with no filtering,sorting,or anything? That feels like the IO is going to be really small anyways.

they compare threads and coroutines for limbo. threads have much worse p90 latencies since they context switch.... im not sure they can draw any conclusions except that coroutines are faster (of course)

So, it's almost useless.

So this isn't faster for people running a monolith on one machine. This is only gives faster tail latency in congested multitenant scenarios. So only a narrow gain in a narrow scenario. Cool and all, all progress is good progress, but also not relevant for me or a lot of people.