5 Photos and videos
Cal Lavicka retweeted
Jun 9
More AI-generated code doesn't make your team faster. It might actually slow you down.
634
2,454
19,364
6,551,680
Cal Lavicka retweeted
May 4
We just moved into our first office in the Mission. My only requirements were that it be a single-story building and formerly a garage
May 4
It could actually be a significant problem that Europe doesn't have enough garages. This sounds like a joke, but I'm serious. Garages let you work on stuff that doesn't matter yet, which is how big things often start. The outliers of ideas need the outliers of space.
1
2
15
1,341
Cal Lavicka retweeted
Apr 22
Can a language model learn, end-to-end, what to keep in its own KV cache and what to throw away? Can it learn to forget while it learns to reason?
Deep learning's central lesson: capability emerges from end-to-end optimization, not heuristics/strong inductive biases. But for efficiency, we rely heavily on hand-designed approaches.
🗑️ Introducing Neural Garbage Collection (NGC): we train a language model to jointly reason and manage its own KV cache, using reinforcement learning with outcome-based task reward alone. No SFT, no proxy objectives, no summarization in natural language.
New paper with @jubayer_hamid, Emily Fox, and @noahdgoodman!
30
132
904
164,748
Apr 22
idk aws, should I use the "enhanced" builder, or the "advanced" one? tough choices...
1
2
75
Cal Lavicka retweeted
Apr 8
We're hosting a happy hour for ~50 developers on Accel's rooftop tonight at 6:30 and have a few spots left. DM me to get added or RSVP at the link in the comments
1
3
10
719
Mar 20
"Strong preference for software that reduces chaos instead of adding more of it" everyone should strive for this
Mar 20
Products are built by humans, and I'm leaning into that.
Added a new section to the Solo homepage.
3
95
I've been seeing a lot of folk on the timeline say IDEs are dying. But I don't get it. Are you guys not reviewing the code your agents generate 😅
Mar 19
I hear the same. The IDE is dying.
Everyone's managing agents now--mostly local, some cloud.
Honestly Devin from @cognition is extremely slept on as the SOTA SWE agent.
68
21
706
56,443
Mar 18
When you give every engineer a magic button, it separates the ones who truly care about the craft and the ones who were never going to succeed anyways
Mar 18
The biggest issue for me with agents is that they are hard to resist. But then you can build yourself some shit into the codebase that you get to regret in record time. And no, I don't think you can vibe yourself back to sanity with better models.
1
38
Mar 17
There certainly are times that they speed me up, but I am scared of what our codebase might have looked like if I didn't keep such a close eye on the code outputs
Mar 16
im fully convinced that LLMs are not an actual net productivity boost (today)
they remove the barrier to get started, but they create increasingly complex software which does not appear to be maintainable
so far, in my situations, they appear to slow down long term velocity
2
54
Cal Lavicka retweeted
Mar 5
Wouldn't it be easier to review the before/after of a UI component than the underlying code? 🔜
2
2
10
439
The <Suspense> is over.
Solid 2.0 Beta is now released (next tag on npm). 🎉
github.com/solidjs/solid/rel…
32
112
818
117,965
Feb 15
Code review is the superpower that is still needed
The main problem i see with LLMs is that if I care about the code output as supposed to just the end result they still really suck
2
101
Feb 10
If only there was a better way to read it...
Feb 9
if you're not reading every line of code going into your codebase, you're not going to make it
1
4
158
Cal Lavicka retweeted
Feb 9
if you're not reading every line of code going into your codebase, you're not going to make it
109
33
916
71,515
Feb 6
the struggles when your cofounder decided to have the wrong name
1
5
320
Cal Lavicka retweeted
Feb 5
Code quality, like the quality of who you hire, compounds. Investing in it early is critical, particularly since both engineers and agents are memetic and will learn from existing code
Feb 5
Code quality matters
1
4
216
Feb 5
LLMs suck at creating tests. Their tests are too basic and they cheat all the time, validating buggy behavior to get 100% test coverage rather than flagging real bugs.
So, I created an opencode plugin to fix this
1
6
18
3,495
Feb 5
In my experience, it helps create actually useful tests, and helps me specify intended behavior. You can even ask it to update JSDocs, or spawn subagents that don't have this restriction and can fix the bug.
1
4
159