Hacker News
X / Twitter
I GOT ONESHOTTED BY OPENPILOT
___
"delivery bots suck! i can do so much better"
... right?
spent two years nights and weekends building prototypes
raspberry pi, cheap android phone, jetson with csi cam and lte hat ... i tried it all!
then ... finally, i bought a Comma 3X
it aint cheap! but had everything i needed
- 360 fisheye cam
- LTE
- controls
- self driving model
i ported over the low-latency teleop system i had been developing entirely into openpilot
and it was ripping!
but ... would the model actually work on such a small car!??
wait ... sick!
its working on a benchtop looking at stock driving youtube
and it sort of works going down the hallway ...
ok, let's take it the streets!
... calibration going well
... hands off the wheel
... ok, lets hit the engage button
... oneshotted ...
Sometimes most of my interface imagination looks like this in my head, shows up vividly.
Probably because I've collected more than 10,000 artifacts around this style.
Credit goes to all the respectful owner of all of this beauty. https://t.co/FNqM6MJ1Lg
🥰Super excited that SceneWeaver (https://t.co/uAutDQ6Ay5) won the best paper award at the IROS25 RoboGen workshop. SceneWeaver provides an agentic framework for tool-based 3D scene generation, given a language description as input, you can generate or edit a corresponding details with lots of details.
How do you use coding CLIs for prompt optimization?
1. Write your evals results into SQLite.
2. Give your agent a CLI tool to run the evals. On top of that, give your CLI access to the function that has the prompt.
3. Includes a prompt table that includes metadata on the number of tokens and the model
And then tell the CLI
"Your job is to run the evils and inspect the error modes. Use SQLite to inspect the data for failures and successes. Then rewrite the prompt and try to get a better performance on the evil while minimizing the number of tokens required to do so. Keep going until you cannot improve the metric without increasing the number of tokens in the prompt. Give me your final findings in an https://t.co/cCIEibkRZy and your recommendations on next steps."
A redditor used spaced repetition to teach his kids how to read. Results:
- 4yo daughter could read at 5th grade level
- 3yo son could do 12x12 times tables https://t.co/JTIxcBsUhW
🚀Our Code and Weight is public available. We provide detailed examples, and try out at https://t.co/0jE0CrSTLY https://t.co/uBtgFpRhyQ https://t.co/l3GP1LooJX
A great guide by @dammit_jeff https://t.co/ns7wRGGFdT
I cherish any time I have an excuse to mock up ideas in Excalidraw, such a lovely tool. https://t.co/4i6xNocyBk
We have a few of these browser agent products now - Perplexity Comet and Brave Leo and Claude for Chrome have been around for a few months now
Anyone finding value in them? What have you used them for that genuinely saved you time?
We’re proud to share that 100 Lost Species has been awarded Site of the Day on @awwwards https://t.co/4ceKI5F9yS
Inspired by recent convos on ViPE, VGGT, and MapAnything