Karnbir

Honestly in everything you do | CS @UBC

[Karnbir @KarnbirR]

@DevinOlsenn Prob the regular from voice that is in the iOS app. Nothing with reasoning ability. I remember asking it which months had the letter x in it, and it gave me random months.

[Karnbir @KarnbirR]

@JonBbC_TechGeek What do you think about a front air vent mesh grille? Tesla service suggested I don’t install it for warranty purposes https://t.co/qNlBFFYijd

[Karnbir @KarnbirR]

@Ari_on_twitr Cost of repairs on tesla is crazy! Gf wanted to buy her first car in 2023. Not only was the M3 more expensive but the insurance was also significantly more than our Elantra hybrid.

[Karnbir @KarnbirR]

@kane

[Karnbir @KarnbirR]

@anatolykim8 @gerrylum2 The current big model is a scaled up small model with no architecture changes. They will rework a special architecture just for the big model to make it really good! Source: https://t.co/VXfzeC1WsV

[Sawyer Merritt @SawyerMerritt]

I made a video compilation of @Tesla FSD (Supervised) avoiding accidents, pedestrians and animals. This technology is already saving countless lives.

Thanks for submitting your clips!

[Karnbir @KarnbirR]

@hamza_q_ Good stuff! Worked like a charm

[Karnbir @KarnbirR]

@DevinOlsenn @JoshWest247 lol, def AI, you can see the prime text fissile and morph. Plus this looks like driving on an invisible road.

[Karnbir @KarnbirR]

@DevinOlsenn @brian4591 I think this is only for the M3P. No one has posted the regular M3/MY animation yet

[Karnbir @KarnbirR]

@MrBeast 2456789 Pennies

[Karnbir @KarnbirR]

@FaZeApex @ycombinator FaZe Apex: Road to $1M ARR

[Karnbir @KarnbirR]

@Mallowed https://t.co/YzF3uYoSCe, get whatever is A tier and cheap in your area

[Karnbir @KarnbirR]

@wholemars Comma 4 has 3 TOPS of compute and will stop for a red light and go for a green. Rivian R2 has 200 TOPS of compute and won't even resume after a lead car departs?

[Karnbir @KarnbirR]

@aidaknx Windows Vista reborn!

[Karnbir @KarnbirR]

@simcity99 Nice refactor

[Karnbir @KarnbirR]

@macman222 @gerrylum2 @comma_ai already in sunnypilot

[kache @yacineMTB]

There is an autism lottery. Either you're autistic about something totally monetarily useless like sonic speedrunning, or you are autistic about something that makes a fuck ton of money like GPU schedulers

[Karnbir @KarnbirR]

@SheriefFYI What makes you say that? I was disappointed by its TAA implementation, FSR 3 baked in instead of FSR4, and DLSS seems to have a strong performance hit: https://t.co/CVNKZb3CbA

settings don’t scale well either. https://t.co/cjOTD5vqAK

[BenchmarKing @BMKing23]

There’s something wrong with the DLSS implementation in #007FirstLight.
In Quality mode, the game defaults to Preset K, and the internal resolution looks correct. but for some reason, DLSS is noticeably slower than FSR 3 and delivers much smaller performance gains than expected.

[Karnbir @KarnbirR]

@RodolfoYoshii @Chansoo @Tesla How many of those teslas do you think were using FSD? x)

[Mitchell Hashimoto @mitchellh]

I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.

As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!

I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.

It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?

88ms => 1.5ms
150K allocs => ~500 allocs

Incredible right? Nope.

My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.

This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.

The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.

Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.