HNHacker News
TopNewBestAskShowJobs

dragochat

46 karma · joined May 26, 2025

crafting https://www.facebook.com/ai.drago.chat
submissionscomments
dragochat··on Migrating to the EU
...also migrating AWAY from Fastmail (Australian) and TO an European provider sounds like a very bad idea - I'd kind of want both the US and the EU legally away from my coms at all costs (!)
dragochat··on Migrating to the EU
how about the OPPOSITE problem: _anyone knows of any non-EU AND non-US email providers_? with email accounts as the roots of trust for many things, i'd really wanna know how can I get a trustworthy one not-attached to eithern an unstable system (US), or a very overregulating one like the EU juristictions...

and ofc, non-CN too

dragochat··on Mistral AI Releases Forge
where sample notebook/script? where github? where signup?

...learn a thing or two from NVIDIA or gtfo

dragochat··on Gemini 3.1 Pro
...you sound like a typical opus-person :P Just use anthropic's flagships if you want good instruction following, focus in long convos, and proper understanding of guidance-when-wrong.
dragochat··on Gemini 3.1 Pro
> feel more work to get actionable insight

WHAT?! I find that exactly the nice sharp formatting are what makes it EASIER to get actionable insight from it...

(Plus the weird-but-cute unrequested analogies are nice to occassionally elicit a smile and keep you motivated :P)

dragochat··on Introduction to Software Development Tooling (2024)
obligatory link to the famous very similar resource - MIT's The Missing Semester https://missing.csail.mit.edu/

...I'd be curious if anyone has went through _both_, unlikely as that may be, and could give some comparison :P

dragochat··on Show HN: AlgoDrill – Interactive drills to stop forgetting LeetCode patterns
...the f?! why are we interviewing ppl for things like this?!

you either:

(a) want DEEP understanding of math and proofs behind algorithms etc.

(b) can get away with very high level understanding, and refer to documentation and/or use LLMs for implementation details help

there is no real world use case for a middle-ground (c) where you want someone with algo implementation details rote-memorized in their brain and without the very deep understanding that would make the rote-memorization unnecessary!

dragochat··on Supercookie: Browser Fingerprinting via Favicon (2021)
Well, we'd probably agree on most things... and re the photography example, afaik model release forms work similarly in the EU and US, right?

Now website code does typically run on your device, but I'd say that once you're a paid logged in user you clearly accepted to run it, under the conditions of it staying in its browser sandbox so... if you think it's "malware" then just stop being a customer. Otherwise software has a right to monitor its own operation.

...but yeah, maybe I missed the context a bit, a tracking pixel style tool will likely be used to track not customers but leads, so I do get your point, it gets trickier there and maybe privacy laws have a point there (as long as they stop there... hint: they usually don't!)

dragochat··on Supercookie: Browser Fingerprinting via Favicon (2021)
+ as a bonus we'd also incentivize businesses to internalize their marketing and related tech operations (since sharing data with 3rd parties would not be allowed), same for AI-customizations etc., forcing them to tech-ify and become more tech-savy businesses instead of externalizing all such things to evil big tech (eg. a clothing store chain could compete not only by producing better clothes, but also by developing better monitoring and generative AI for human-in-the-loop hyperpersonalization, spreading tech out... instead of outsourcing these to tech or big-consulting companies as they do now when the too-little-data they so collect anyhow is otherwise easily share-able to third parties)
dragochat··on Supercookie: Browser Fingerprinting via Favicon (2021)
If you visit my eg. physical clothing store I'm allowed to monitor your in-shop behavior to better optimize my store for your needs. Same for a restaurant etc. That's how _you_ get _much improved services_ and I get _happier customers_.

Ofc I'm not allowed to freaking resell that data. THIS is the problem in online: releseling and data-brokers. Just KILL these categories of businesses off completely and make _them_ criminal (like even give f prison sentences to their operators).

We should get back to our sanity in ONLINE. As long as you're on _my (online) property_ and using _my services_ I can of course see EVERYTHING you f do, and should stop pretending I don't (as a business, ofc - anonymization exists and not any random employee can access any customer's data, probably should never access both data and identity correlated unless they're actively investigating some serious fraud). As long as I'm not sharing this data with anyone else, I should be 100% allowed to use every drop of this data to improve my services to you and totally differentiate myself from the incompetent competition that can't properly do this.

Data privacy (from EU's GDPR to... everything else) only helps big corporations fend-off competition from small startups or boutique shops that could easily out-compete them by offering hyper-personalized hand tailored micro-optimized experiences for their smaller number of customers based on the loads of data they collect from them. In the EU I've only ever seen these kinds of laws severely hamper small boutique or family businesses that wanted to hyperpersonalize to everyone's gain while big corpos easily surf around them with their teams of lawyers.

...we've all been brainwashed by this privacy psyop to sheepishly "fight for our privacy" in ways that are detrimental to us and only help our corporate oligarch overlords maintain an even tighter grip on power, while offering us worse and worse services. Wake the f up, DATA IS MEANT TO BE USED to IMPROVE goods and services, not remain uncollected or sit unused!

dragochat··on Yann LeCun to depart Meta and launch AI startup focused on 'world models'
the fact that a not-so-direct experience of reality produces "good enough results" (eg. human intelligence) doesn't mean that a more-direct experience of reality won't produce much better results, and it clearly doesn't mean it can't produce these better results in AI

your whole reasoning is neither here not there, and attacking a straw man - YLC for sure knows that human experience of reality is heavily modified and distorted

but he also knows, and I'd bet he's very right on this, that we don't "sip reality through a narrow straw of tokens/words", and that we don't learn "just from our/approved written down notes", and only under very specific and expensive circumstances (training runs)

anything closer to more-direct-world-models (as LLMs are ofc at a very indirect level world models) has very high likelihood of yielding lots of benefits

dragochat··on Rio Terminal: A hardware-accelerated GPU terminal emulator
...how do you search (history) effectively in ghostty? or are you not dependent on that feature?
dragochat··on Rio Terminal: A hardware-accelerated GPU terminal emulator
Quick question: can I JUST simply C-f (or smth else) to SEARCH the terminal for text without any extra complications? Like in macos default terminal and other sane ones?

...or is this another Ghostty geeksfest that wants me to C-S-whatever to dump history into a file then pipe it into a pager etc etc etc?

[+EDIT: ...apparently not liking ghostty gets you downvoted too? weird amounts fanboysm and hype for an underfeatured and poorly documented terminal]

dragochat··on Distillation makes AI models smaller and cheaper
YES

We'll always find uses for more intelligence if it keeps getting more and more general (I don't like the term AGI bc. I think the "G" there is quantity not a quality, and humans are very low on generality too compared to what could be mathematically and physically possible for intelligence in our universe).

...we won't stop until the planet is papered with compute hardware UNLES we accelerate space development too (that's why SPACE is CRUCIAL!) and go grind the asteroid belt into thousands of datacenters too, then on and on.

There's a whole yummy lightcone that awaits to be eaten :P

dragochat··on Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
great, so pretty soon it will be prevented or illegal to even finetune models above a certain cap threshold - dog forbid you... UNalign it (-:
dragochat··on Open source can't coordinate?
nah, we all know the "real-life CDE" is called either OpenGL (see ImGui https://github.com/ocornut/imgui, egui https://github.com/emilk/egui etc) or... HTML/CSS/JS (see Electron, Tauri etc)
dragochat··on Open source can't coordinate?
exactly, you can't standardize on a solution before any good one exists in the first place
dragochat··on MicroExplorer: Lisp-machine and Mac hybrid – it existed [pdf]
would really want to hear stories from ppl who actually used one of these
dragochat··on Marines being mobilized in response to LA protests
> Everything has beginning and end

Most things of real interest do not tbh. The point should be to learn how to play infinite games. Sure, there's hard-to-impossible prices to pay, and/or to force others into paying. But what if it can be made to work, won't any price be worth it for that... ?

The epicenter is where the fun is at, even if the price might be sacrificing things of value for your average population and some unquantifiable measure of fluff like national-identity or whatever people imagine it's real nowadays.

I'm an unashamed globalist, but would rather have a "world village" with the US (despite that US being maybe quite different from what most gen-pop wants) at its heart :P Imagine all the people...

dragochat··on Go is a good fit for agents
good point, might be worth a revisit
dragochat··on Apple introduces a universal design across platforms
uhm, maybe no... data-drive in UX means sliding towards the lowest common denominator, optimizing at first for the dumbest user, then later giving in to dark patterns and quasi-scamming

there's room for creativity in UX, lots, just not at the "how does the texture of a button feel and flow" - need to move HIGHER level, towards eg thinking of experience minimizing cognitive load, increasing synergy and augmentation ppotential etc etc ...the ceiling is waaaay higher than most UX ppl think

dragochat··on Marines being mobilized in response to LA protests
> The planned decline of America won't be like other countries

Maybe bc americans WON'T and SHOULDN'T settle for a decline - they should violently rebel against this mindset and claw they way UPWARDS - there's more room for more growth, even if you lose #1 status and have to settle for #2 for a while you can still catch up etc.

It's good that at least the US and China are NOT infected with this degrowth and "cyclical history" mindvirus that seems to be doing the rounds in Europe and elsewhere... keep being a bastion of endless progress brothers, fight the good fight! There's a whole light cone to eat/infect (if not for us the for the successors we'll build)! Whoop, whoop!

Jokes aside though, most of the open world we live in today owes its existence to ideas, mindsets, $$$ and tech exported from the US, and I'm sure there's way more cool stuff to come from you once you properly clean up the parasitic individuals and institutions that have infected your society. Purge on and keep growing, fight for a deservedly big chunk of the Dyson sphere and beyond!

dragochat··on Go is a good fit for agents
...could we just get Go's GREAT concurrency model and decent standard lib, but in a language that is less horrible than Go (like with decent type system, enums, expressions based grammar, pattern matching etc etc)?

pretty please :P

we all yearn for a good static language, and most of us would kill for "something like Rust (good type system, syntax, tools) but without ownership / linear-typing - just a good GC, all-on-the-heap and a dash of nice immutable datastructs"...

dragochat··on Apple introduces a universal design across platforms
this++

We need more UX that people can "settle into" instead of the constant assault of superficial change that drains energy from everyone's ongoing effort to adapt to exponentially increasing fundamental change!

dragochat··on Marines being mobilized in response to LA protests
as a non-American reading this, only reaction can be... "WTF?!"

can you elaborate on that?

centralized systems for identity are used all over the f world, and de-facto you have them in the US too (hello credit cards, hello driver licenses, hello... SIM card?)... they can be used as _tools_ by regimes with fascist tendencies, but their existence alone is quite neutral

ALL countries with functional governments NEED centralized identity systems to function (hello... IRS?), and they all get them even if you like to pretend they don't exist.

...why not just accept they are there, and focus on properly securing them from attacks, making them unalterable by corrupt officials (from simple checksumming and write-only permanent archives to full-on blockchain solutions), and preventing unrestricted access to them, all technically doable.

If you PRETEND you don't have a centralized identity system, don't you just leave all these problems unfixed and just go on with a broken system forever?

dragochat··on A deep dive into self-improving AI and the Darwin-Gödel Machine
I guess you could prove the conformance of a particular implementation if you'd implement separate Plan & Implement stages + a "superior" evaluator in the loop that would halt the evolution at a certain p(iq(next_version) > iq(evaluator)) as an "outer halt-switch" + many "inner halt-switches" that try to detect the arising of problematic behavior of particular interest.

Ofc it's stochastic and sooner or later such a system will "break out", but if by then sufficient "superior systems" with good behavior are deployed and can be targeted to hunt it, the chance of it overpowering all of them and avoiding detection by all would be close to zero. At cosmic scales where it stops being close to zero, you're protected by physics (speed of light + some thermodyn limits - we know they work by virtue of the anthropic principle, as if they didn't the universe would've already been eaten by some malign agent and we wouldn't be here asking the question - but then again, we're already assuming too much, maybe it has already happened and that's the Evil Demiurge we're musing about :P).

dragochat··on Nitrogen Triiodide (2016)
> There is no risk of terrorists using NI3 because anybody who made it in sufficient quantities to do serious damage would succeed only in blowing themselves up: those who do so are humorous, not terrible.

...not gonna say more, but, buried in the info and video there IS actually an idea for overcoming this nasty limitation (if you can mostly live up with "quasi-random detonation time", which could be acceptable for _some_ nefarious uses). Tbh I'd be more curious if any current gen LLM can figure it out.

dragochat··on Claude 4 System Card
...wonder how much effort is put into making these models seem _more_ dangerous than they actually are tbh, for the hype value. I mean, it's clear the writers of the report overemphasized its blackmailing ability etc. etc. These companies are gaming our fears and attention for engagement, and it's more problematic _because_ the substrates of these fears are in fact real too.
dragochat··on Chomsky on what ChatGPT is good for (2023)
...for the lulz try asking ChatGPT "what is Chomsky (still) good for?"
dragochat··on Lisping at JPL (2002)
true Evergreen ...nostalgic memories of stories from old 20%-time-and-don-t-be-evil Google days. Sigh. I still hope AI takeoff somehow revives Lisp-as-a-human-friendly-computer-language or something in the vein of this.
← PreviousPage 2 of 3Next →