Konstantin

998 posts

Konstantin banner
Konstantin

Konstantin

@kauffinger

I enjoy solving problems in Laravel, NVIM, and LLMs Tech Lead @ InnoBrain https://t.co/puXQPhxCx3

Katılım Haziran 2020
455 Takip Edilen229 Takipçiler
Konstantin retweetledi
Mitchell Hashimoto
Mitchell Hashimoto@mitchellh·
Mind boggling to me that I can make a thing faster and there's always people that ask "but why?" What kind of mentality is that? The pursuit of excellence does not need justification. Also, I find in so many cases, we can't know the impact of an improvement until we do it. For example, one I've talked about before: Ghostty's high IO throughput has enabled terminal program (emulator and TUI) fuzzing at a speed thats incomparably fast to prior solutions. This has resulted in upstream patches to resolve issues in popular projects like btop, tmux, and more. Speed enabled that anecdotally example that lifted the tides of adjacent communities that don't rely on Ghostty technology at all. I didn't predict this. Make things better because they can be better and let the results naturally play out.
English
169
447
5.7K
393.7K
Konstantin
Konstantin@kauffinger·
@awesomekling @antirez Agreed, the more experienced are the more knowledge is encoded in sentences that look completely normal to you.
English
0
0
0
17
Andreas Kling
Andreas Kling@awesomekling·
Yeah, I think you're right about this. That said, older guys like you and me (who have spent decades building our mental models) can operate very effectively in this new environment with low friction. I don't know how new developers will build the skills and experience needed to move completely freely. Not that I'm worried about it. Humans always find a way :)
English
7
0
106
11.5K
Konstantin
Konstantin@kauffinger·
@jessethanley @p_millerd Makes sense. Might be a difference in how we work then. I like having the agent go off for longer to parallelize my work. Not a fan of a fast agent loop. Prevents flow state for me. E.g. I let it work while reading the docs of one of the dependencies top to bottom.
English
0
0
0
15
˗ˏˋ Jesse Hanley ˎˊ˗
˗ˏˋ Jesse Hanley ˎˊ˗@jessethanley·
If you are doing deep work, day to day, then it becomes obvious that Fable/Sol is too slow to work properly with. If you are doing research/greenfield work then yeah sure throw it at a cloud agent and come back later. Fable and Sol are far too slow and breaks me out of flow. Analogy that might be helpful: a semitrailer is capable at dropping kids off at school but an wagon/SUV is likely a better fit.
English
1
0
1
41
Paul Millerd
Paul Millerd@p_millerd·
fable creates an incredible amount of work for a simple bug fix
Paul Millerd tweet media
English
3
0
2
890
Konstantin
Konstantin@kauffinger·
@jessethanley @p_millerd Fair. Just a time tradeoff for me. Whenever I actively prompt I want to put in minimal effort and the patch to be correct when I check it. Do you have a rough rule of thumb how you decide which model to hand what? E.g. for research work I let Fable delegate to smaller models.
English
1
0
1
29
Konstantin
Konstantin@kauffinger·
@jessethanley @p_millerd Fable is probably quite depressed about not being allowed to dispatch sonnet to update my README. But tbf, i never run into stuff like OP. I drive Fable on high with very succinct prompting & no manual skill activation and it judges the amount of work to put in very well.
English
1
0
0
33
Konstantin
Konstantin@kauffinger·
@jessethanley @p_millerd On subsidized cost i just see no reason to use less intelligence than the max i can get at any point in time
English
1
0
1
30
Konstantin
Konstantin@kauffinger·
It’s funny to me that people who run unsupervised agent loops with models that explicitly say that all data is retained at least for 30 days are up in arms about grok cli taking 30% more of the relevant data than their agent normally would.
English
0
0
1
45
Konstantin
Konstantin@kauffinger·
@joshmanders While there is a class of software on which code barely matters, I think a whole lot of people just really want to believe reading code is not necessary.
English
0
0
0
59
Josh
Josh@joshmanders·
I just want to put on the record, I don't give a fuck what achievements someone says, I will never follow their "stop reading code" advice. And ya'll jumping through hoops to try and defend them is glizzy gobbling 9000 behavior. Have some self respect.
English
8
2
58
1.4K
Konstantin retweetledi
Armin Ronacher ⇌
Armin Ronacher ⇌@mitsuhiko·
One thing I know for sure: If you end up in hospital from stress because of tokenmaxxing and FOMO, you're ngmi.
English
27
20
697
36K
John Koster
John Koster@johnmkoster·
"i solemnly swear i am up to no good"
John Koster tweet media
English
4
0
10
2.6K
Konstantin
Konstantin@kauffinger·
@zeeg We have a lot of projects bounded by scope and scale, often consumers of APIs. Depending on how you look at it, I am usually happy with the code to further iterate on getting things right. No one-shotting stuff, obviously.
English
0
0
0
30
David Cramer
David Cramer@zeeg·
How frequently do you estimate a model produces a patch you're honestly happy with, or at least have confidence in no longer term issues? Real world scenarios, not your hobby projects.
English
32
1
31
23.6K
Tilly The Coder
Tilly The Coder@tillythecoder·
When was the last time you had a code editor open?
English
5
0
0
400
Konstantin
Konstantin@kauffinger·
Can we please stop cutting off y-axes?
Sophia Yang, Ph.D.@sophiamyang

✅State-of-the-art performance on R2R-CE 79.4% Success Rate on validation seen 76.6% Success Rate on validation unseen ✅Operates from a single RGB camera, with no LiDAR or depth sensors ✅8B model, built in-house and trained entirely in simulation ✅Runs on wheeled, legged, and flying robots, and generalizes across robot sizes ✅Robust to differences in camera intrinsics ✅Token-efficient training via prefix-caching

English
0
0
0
78
Konstantin
Konstantin@kauffinger·
If you're an actual designer, I am interested in what you think!
English
1
0
1
285
Konstantin
Konstantin@kauffinger·
Using impeccable with fable is pretty nice. Also using flux, btw
Konstantin tweet mediaKonstantin tweet mediaKonstantin tweet media
English
4
6
138
188.3K
Konstantin
Konstantin@kauffinger·
Reminder that you do *not* need to be paid actual hard dollars to have a conflict of interest.
English
0
0
0
29
Konstantin
Konstantin@kauffinger·
People with a platform getting access to models months ahead is horrible for the public discourse. Same situation as with games journalists receiving review copies ahead of time. Conflict of interest is real no matter how much people say they are neutral.
Pietro Schirano@skirano

I can finally talk about 5.6. I’ve been testing it for months and, without exaggeration, it’s the best model I’ve ever used. Fast, smart, genuinely creative, and you guessed it, they finally fixed front-end design. I haven’t needed to check the code I’ve written in two months.

English
1
0
1
195

Keşfet

@awesomekling @antirez @jessethanley @p_millerd @joshmanders @johnmkoster @zeeg @elonmusk