Vivek Aithal
@nuwandavek
Followers
2K
Following
12K
Media
150
Statuses
1K
all things @minusxai (yc s24); ex-research @comma_ai. Revealed preferences greater than greater than Stated preferences
Joined July 2015
how is @claudeai's claude code so damn good and how can you recreate its magic in your own llm agent/workflow? intercepted all cc calls over the last few months and wrote up a 2k word guide on this! (all prompts, tools linked in comments) main takeaways (0/3): 0.
10
34
172
this is 100% correct. - text to sql is solved. but that was never the problem - when someone says “active user” they have more assumptions in that statement than they know themselves understanding data is fundamentally understanding extreme nuances about your company
Not as smart as the people in this thread (I don't know what encoder-decoder or multi attention layer mean). But working in this field for the last few months on the application layer and there are 5 key dimensions that we have learned about this problem: 1. Users need insights,
0
1
3
there’s obviously no way this animal learnt this by RL-ing in the real world. nice intuition from @karpathy from the recent podcast. the beaver here learnt low level motor control/physics from trial and error. there’s something else entirely wrt its dam building goal
This beaver was orphaned and rescued as a newborn, Watch the incredible instinct to build a dam, even though it’s never seen it’s parents build one.
0
0
6
cc @billy_agi 🙏
Build a humanoid robot yourself! 🪚 OpenArm is an open-source humanoid robot. It comes with full CAD, control code, firmware, and simulation tools, everything needed to build, modify, and operate it. The arms are designed to be compliant and backdrivable. Teleoperation is
0
0
0
crazy controversy! on a recent podcast appearance emma said she believes llms will scale to AGI with just more test time compute and dissed jk rowling as being a skeptic. rowling responded with some harsh words for her former colleague. “can emma tell me what the llm’s goals
0
0
5
pre-training as a poor mans evolution to learn soft constraints and make RL actually sample efficient is the best case scenario for llms right?
Finally had a chance to listen through this pod with Sutton, which was interesting and amusing. As background, Sutton's "The Bitter Lesson" has become a bit of biblical text in frontier LLM circles. Researchers routinely talk about and ask whether this or that approach or idea
0
0
2
super impressed by all the demo videos from @corl_conf but I hate seeing cameras right at the grippers in most robotic arms. why is this the case? seems super unnatural
1
0
0
this is fantastic! if you want what ramp did inside your company, @minusxai can help! :) remember, llms cannot do magic. if you shape your data a bit and provide enough context, you can get a pretty magical data experience!
In the last four weeks, @tryramp Research (our AI analyst agent in slack) has answered 1,476 questions, compared to 66 in our help data channel. This is one of my favorite projects that we've shipped. Our team built an agentic solution based on tools to index domain docs, to
0
1
6
great (at times a bit awkward) interview! is intelligence finding verifiable proxies for your goal? short of this, any intelligent seeming behaviour from LLMs is an open loop "best effort". nothing actually intelligent in the world (bees to pugs to humans) behaves this way.
.@RichardSSutton, father of reinforcement learning, doesn’t think LLMs are bitter-lesson-pilled. My steel man of Richard’s position: we need some new architecture to enable continual (on-the-job) learning. And if we have continual learning, we don't need a special training
1
0
3
kee doing kee things! very very cool!
Gemini Robotics 1.5 just dropped! ✨ A🧵on VLA progress: For far too long roboticists have silo-ed action and reasoning modeling. Thinking VLA unifies them—bringing the robotics modeling problem closer to the physical AGI problem and exploiting synergies in the middle.
0
0
2
gave a talk at @lossfunk a while ago about self driving cars in general and some stuff I worked on at @comma_ai. was a great turnout! i hope someone has installed openpilot and fixed the metadrive sim!
New talk dropped on our channel! 👇 listen to @nuwandavek on how to build a self-driving car in your garage! it's full of insights from his work at @comma_ai
1
0
9
recently I wrote a blogpost about Claude Code (how it is so good, and how you can give the same experience in your AI agent) and it went sort-of-viral-ish. will be doing a deep dive into it tonight (8pm IST) in a session with @GrowthX_Club. I'll be covering why simple loops >>>
2
1
6
just got into an auto rickshaw ~50yo driver: I don’t know where this place is me: here, ill put it on your google maps ~a dozen ai girlfriend apps. ooooof. crazy that this is prob his first intro to ai tech.
0
0
3
If you're building AI agents (especially coding agents), stop what you're doing and read this: https://t.co/Rrx9mR7qMe
@nuwandavek distills key insights perfectly. Having built complex agent graphs with LangGraph, I can say this hits all the right notes.
minusx.ai
Claude Code is the most delightful AI agent/workflow I have used so far. Not only does it make targeted edits or vibe coding throwaway tools less annoying, ...
0
2
4
i've been super turned off by marketing-speak ("do content marketing, bro"). as a sign of protest, i corrected too much into not publishing anything(??). response to the claude code blogpost has been really overwhelming! will def write more about stuff i find interesting! ❤️
2
0
20
til translate post on x is powered by grok! it crashed out while translating haha
0
0
1
If you're on the engineering org discord, @nuwandavek is going to be live at 1530 today to discuss this blog post.
lmao. if anyone needs a push to finish a side project/blogpost, commit a deadline to @dementorSam. he will carrot and stick you relentlessly till you finish it. much appreciated, boss! ❤️
0
3
7
lmao. if anyone needs a push to finish a side project/blogpost, commit a deadline to @dementorSam. he will carrot and stick you relentlessly till you finish it. much appreciated, boss! ❤️
I appeal to everyone here to push @nuwandavek should write more! His hit ratio on HN is nuts!
1
0
11
til ~2016 google maps team literally strapped cameras to sheep 🐑 to map out the Faroe Islands. and they called it sheep view lmao this was such a fun and playful company. who comes closest to this now?
3
1
21