@levpopov

Making browsers do weird things. Prev: transportation nerd (RoverRide YC W13 → Lyft Line), immersive language learning, early FB eng.

 San Francisco
Joined January 2008
  • Tweets141
  • Following498
  • Followers328
28Media
It's *not* a well-made keyboard, but I'm still going to be a bit sad when they all run out now that they don't make them anymore.
2
1
1,508
Many reasons to doubt VR, but complaining about a screen sitting too close to your eyes is not a good one. The typical focal length through the optics is much greater than what you keep your phone/laptop screens at. Your eyes are more relaxed looking through VR lenses!
1
1
763
This is also a periodical reminder that getting a larger monitor and putting it further away from you (deeper desk, setting up a keyboard tray, etc) is the best thing you can do to reduce eye strain.
1
660
If your agent recursively spawns itself, it’s an LLM MLM
1
716
Adventures in electronics component sourcing: My email thread with a supplier is now so long and has so many people on it that gmail web UI can no longer handle it.
1
596
GPT4 is impressive but painfully slow (we got spoiled with 3.5-turbo). The most jarring part is GPT taking a while to compute "obvious" tokens - 1) you often already know where it's going 2) *it already knows* where it's going. Yet it has to start from scratch on each token.
1
591
Every token taking the same amount of time is almost certainly an architecture bug. We are throwing away way too much useful computation between each inference step.
1
451
A naive fix could be a hybrid model. A large, slow supervisor that generates core tokens, and a fast "filler" model that speeds ahead when it has high confidence. Could probably even have the supervisor figure out if the filler model screwed up and output a "backtrack" token.
1
378
Not immediately obvious, but most 34" monitors are *smaller* than most 32" monitors
322
Welp, got my board back from PCBA with a component flipped upside down. At least it didn't immediately ignite on power-on.
1
1
5
422
Do you say "Thank you" to ChatGPT/Bing when they provide a good answer?
58%Yes
42%No
12 votes • Final results
1
1
427
Ted Chiang’s New Yorker piece is interesting but does not match my experience at all. 1. ChatGPT is not web, compressed - it creates genuinely new content. Just ask it to write a song/poem/etc about something absurd. The output is clearly not a rephrase of what it's seen.
1
2
620
more replies
4. Sure GPT hallucinates a bunch but it will get better with aux systems. We forget how humans misremember/hallucinate stuff *all the time* - we’ve invented technology called “writing things down” to help us.
1
351
5. Ted Chiang’s piece is still worth reading of course. Also, if you haven't seen it, his short story "Story of Your Life" (on which "Arrival" was based) is incredible and arguably better than the movie.
329
Trying to get your LLM-based app to do what it's supposed to do through prompt engineering:
1
315
Got my tiny @Raspberry_Pi board back from PCBA! Fully functioning RP2040 design at just 12x23mm (SD card in pic for size reference) - includes USB-C port and I2C out over FPC. RP2040 burning bootloader into the chip is amazingly dev friendly - just plug in USB to flash firmware
3
20
81
4,393
If you ever bought an SSD or an SD card on Amazon and started wondering if it's a fake - this tool is for you. A tiny 70 line python script to verify true storage capacity and consistency. github.com/levpopov/ssd-chec…
1
2
361
Might be having an existential crisis. ...am I a robot?
2
480
Load more