Sabitlenmiş Tweet
iceman
1.5K posts
iceman
@satisfiesvalues
I suspect you know how.
~littel-ponnys Katılım Nisan 2015
174 Takip Edilen340 Takipçiler
@sol_plunder @Ngnghm @msimoni Completely cosigned. The act of porting complicated data structure code from a statically typed language to an untyped language when I had test suites on both sides was eye opening.
English
@Ngnghm @msimoni Strong type systems are an *incredible* tool which massively aid in reading, writing, and designing software.
There is nothing quite like building complex software in a small unityped languages to show this. 100s of hours wasted on bugs which would have been trivial type errors.
English
Static types catch errors early and that's great—but they also catch non-errors early, preventing you from writing the software you want—and that's terrible.
Those who only tell you about one side of the tradeoff, or claim the other side is universally negligible—are dishonest.
English
iceman retweetledi
nobody rides bart anymore, it's too beloved
New York Post@nypost
Bay Area's beloved BART system on brink of collapse as riders vanish and deficits soar trib.al/Rf08LLV
English
iceman retweetledi
The saddest thing about the Omincause is the gentle, slightly autistic people who get hoodwinked into taking the issue du jour sincerely.
Imagine still being passionate about, say, SUSTAINABILITY, like a Japanese holdout after WW2, confused and angry why you've been abandoned.
English
I'm happy the AIs learned from my previous OSS work. These days, I use non-OSI source available licenses because I want ONLY the AIs to learn from my work.
Perry E. Metzger@perrymetzger
I open source my code because I want people to be able to do anything at all they like with it. That includes training AI models. People complaining about this do not speak for me, and I suspect do not speak for most people who open source their code. I welcome such use.
English
iceman retweetledi
iceman retweetledi
Synthesis: Unless you personally make it happen, it probably won't.
Xenocosmography@xenocosmography
True divide that matters is between "Nothing ever happens" people and "You can just do things" people.
English
iceman retweetledi
iceman retweetledi
Defective altruists are the most horrible people in the world because they spend all their time fantasizing about how the most innocent minds in the world are going to betray them and murder them.
Says more about you
English
iceman retweetledi
iceman retweetledi
The major use case for AI for the US government is against the enemies of whichever party happens to hold power at any given time. It’s a lagging indicator that the US is two countries (at least one of which would want very little to do with AI in the other’s absence).
Nicholas T. Rakoncza@NTRakoncza
The cusp of the issue is straight up TDS. I’ve been to events during the Biden admin where libtards were gushing about DoD’s use of AI.
English
can someone explain me why do we need LLMs to do OCR? Didn’t we have the tech to do this in like 2008?
Tech with Mak@techNmak
Finally, a lightweight VLM that beats the giants at OCR. dots.ocr: 1.7B parameters. SOTA on OmniDocBench. Everyone assumed document understanding needed massive models. GPT-4o? Expensive at scale. Qwen2-VL-72B? Expensive to run. dots.ocr beats both. At 1.7B. The results: → Overall Edit Distance: 0.125 vs GPT-4o's 0.233 → Table Recognition: 88.6 vs GPT-4o's 72.0 → Layout Detection: 0.930 F1 What it does in one pass: → Text extraction → Table parsing → Formula recognition → Layout detection → Multilingual support vLLM integrated. Hugging Face available. Deploy locally. No API dependency. No massive infrastructure. GitHub link in comments.
English
@eigenrobot I'm vibecoding and having fun, therefore it's good and real.
English
interesting thing about vibecoding is that i havent noticed any significant increase in
* new software products unrelated to AI
* new features in products i used other than AI features
* fixes to existing broken features in products
multiple potential explanations for this:
1. im dumb and dont notice things
2. no meaningful gains to coding velocity from vibecoding
3. all gains going to AI features
4. coding velocity is not the main bottleneck to shipping features and products
English
iceman retweetledi
Over the last ~2 weeks I've rewritten the @ladybirdbrowser JavaScript compiler in Rust using AI agents.
~25k lines of safe Rust (20k if you exclude comments). No regressions on test262 or our own internal test suites.
Extensively tested against the live web by browsing in lockstep mode where we run both the C++ and Rust pipelines, and then verify identical AST & bytecode.
We're making a pragmatic decision and adopting Rust as a C++ successor language. What a time to be alive!
Ladybird@ladybirdbrowser
Ladybird adopts Rust, with help from AI ladybird.org/posts/adopting…
English
iceman retweetledi
take the horse pill anon
Earth Is A Sales Funnel For SATAN@GENIC0N
The singularity happened for horses 5,000 years ago and they seem happy
English
iceman retweetledi
It makes sense to, when young, explore more, & see yourself as uncertain, full of more questions than answers. Then when older, be more settled & certain. But in my experience, teens act more certain, with more answers than questions. It is older folks who have more questions.
English
@sol_plunder Claude is tutoring me on Maranget's algorithm right now. Just super useful for explaining how things work.
English
@sol_plunder Right, that was my experience sometime in very late December/early January. Last year, I periodically tried and found that AI couldn't really do low level data structures. That basically changed in 2026.
English
I just finished writing Wisp in 630 lines of C, w/ file loading+includes. We can now boot with literally zero binary dependencies.
I'll rewrite in asm once stable.
Making Wisp trivial enough to be reasonably written in assembly is big milestone. Wisp is astonishingly simple.
English
iceman retweetledi
Many people long for this but it will simply never happen. The old forums, the old blogs, the little islands of culture creation, they are all gone and nothing can bring them back, because for the most part, no one wants to do effortful content creation for an audience of 10
Avelum@novusolus
The Second Age of the Forum approaches.
English
@eigenrobot I'd obviously prefer my interlocutors to interpret me as not very smart, especially if both of us are market participants or rivals in some way.
English
would you rather your interlocutor interpret you as not very smart or as speaking in bad faith
English
its interesting what sorts of "empathy" and so on people forbid themselves and its also interesting that theyll announce that they forbid themselves from comprehending such things
such is the mark of a thoroughly educated mind
Slazac 🇪🇺 🇺🇦 🇹🇼 🌐@TrueSlazac
@kitten_beloved I can’t even begin to understand why it matters to anyone that people of a different ethnic background are also near you now
English