Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
lostmsu
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
15 ms
·
451.
▲
by
lostmsu
6mo ago
How about passing a job interview better than someone else?
452.
▲
by
lostmsu
6mo ago
> fake ... obviously not OK Obviously OK, and the puritans have to learn to live with it.
453.
▲
by
lostmsu
6mo ago
> They want people building integrations that are difficult to undo so that they lock into the platform. Ironically, they are now playing against their own models that can relatively easily build wrappers around any API shape into any ot
454.
▲
by
lostmsu
6mo ago
> Society evolves through epiphenomena caused by the behaviour of the majority I would argue plenty of significant societal changes were caused by the behavior of relatively small number of people. Even more so when you include instances
455.
▲
by
lostmsu
6mo ago
I hope you are not running models under Q8, preferably Q8 directly from the vendor.
456.
▲
by
lostmsu
6mo ago
Impossible is a strong word given our collective way of thinking has been reproduced with decent level of approximation.
457.
▲
by
lostmsu
6mo ago
Meetings? That's not interesting. I'm working on replacing myself (or rather my body) after death in a ship of Theseus manner.
458.
▲
by
lostmsu
6mo ago
I just add Yggdrasil to all my nodes. Removes the need to deal with nginx also.
459.
▲
by
lostmsu
6mo ago
Well no, Google at some point in mid 201x screwed up some of photos hosted on Photos. My personal backup has been flawless (so far). Would have spent a couple thousand $ by now, if stayed on it.
460.
▲
by
lostmsu
6mo ago
And AMD and Nvidia.
461.
▲
by
lostmsu
6mo ago
https://en.wikipedia.org/wiki/Execution_of_Charles_I_of_Engl...
462.
▲
by
lostmsu
6mo ago
Not sure why you are asking me about it. The point of my comment is that Iran will allow streaming what they'd want people to see.
463.
▲
by
lostmsu
6mo ago
Wouldn't they shut down the Internet for me but not for thee?
464.
▲
by
lostmsu
6mo ago
Any people or combatants? What about the other side? Do you expect the ratio to be different in the last 25 years?
465.
▲
by
lostmsu
6mo ago
Reminds? In a way that's basically a ripoff of the idea just applied to a different topic.
466.
▲
by
lostmsu
6mo ago
> Why are we outsourcing creativity and research and interest in discovery to an llm? This is also weird. I hate planning vacations, but I like going to them.
467.
▲
by
lostmsu
6mo ago
What's the professional human baseline?
468.
▲
by
lostmsu
6mo ago
This would be more interesting if the commits would list everyone who voted "yes" as authors.
469.
▲
by
lostmsu
6mo ago
One time retrieval is 1.5x times more expensive than HDD that would fit the data.
470.
▲
by
lostmsu
6mo ago
Transformers too. JEPA any day now
471.
▲
by
lostmsu
6mo ago
Are you considering batch inference?
472.
▲
by
lostmsu
6mo ago
You are anti-progress. Pro-humanity is not the same as pro-progress.
473.
▲
by
lostmsu
6mo ago
More like 6 months now. Qwen 3.5 is on Sonnet 4.5 levels
474.
▲
by
lostmsu
6mo ago
I think previous models could do hacking just fine.
475.
▲
by
lostmsu
6mo ago
You are reading the percentages wrong. Because 100% is maximum, you should be looking at error rates instead. GPT has 25% on Terminal Bench and the new model has 18%, almost 1.4x reduction.
476.
▲
Codexia: OSS Agent Workstation for Codex and Claude
(github.com)
2 points
by
lostmsu
7mo ago
|
1 comments
477.
▲
by
lostmsu
7mo ago
Agent Workstation for Codex CLI + Claude Code — with task scheduler, git worktree & remote control, skills management
478.
▲
by
lostmsu
7mo ago
It's unnoticed because it didn't. In Google's own benchmarks they are on par, and I've seen 3rd party benchmarks where Qwen beats G4 with high margin
479.
▲
by
lostmsu
7mo ago
That was 3.7. We might be past that.
480.
▲
by
lostmsu
7mo ago
AFAIK that model is pretty old, and it was explicitly trained for SVG generation. For other models the capability of generating SVGs of real stuff is accidental. Same as GPT-5.x and Sonnet 4.5+ being able to generate MIDI music.
More ›