HNHacker News
TopNewBestAskShowJobs

blfr

8,622 karma · joined January 7, 2014

genetyczny.belfer@gmail.com

meet.hn/city/52.2319581,21.0067249/Warsaw

submissionscomments
blfr··on Fable and the end of the free lunch
What are all these rote coding tasks people do that they can farm it out to lesser models?
blfr··on Fast and Hard Code
Try Fable and make it write a ton of tests. I generally don't trust benchmarks, as they don't match my experience, but this one[1] very much does.

[1] https://www.primeintellect.ai/research/nanogpt-speedrun

blfr··on Wi-Fi 8 is the first wireless upgrade in years that isn't chasing speed
It's the first because we finally caught up on speed. We have it better than I imagined possible.

Last year, I saw 1600 Mbit/s from the Internet at a normie's place with an ISP-provided router. Still have a cat 6 cable around the living room but haven't used it in the last 2 years. Not exactly wifi but I routinely get 600-800 Mbit on 5g networks in crowded areas.

blfr··on There's no reason for software to be slow anymore
Yes, but it's also built by people who would rather tinker with AI than build MacOS apps. Intrinsic motivation is very hard to beat, especially in subtler areas like good UX or performant software.
blfr··on Civic Hygiene – avoid building technologies that could be used by a police state (2013)
Libertarianism has its moments but it generally hasn't caught on because state capacity is unique and useful. And virtually every last bit of it can be used to facilitate a police state because it's... a state.

Sure, we could all live without IDs and use motio belli for defence but this arrangement will be soon replaced by someone more organized because there is a huge return to that organization despite its risks.

blfr··on Win-V combo from Windows on Ubuntu
I didn't even know super+v on Windows. Not to sound all entitled about opensource software but can we get a screenshot on GNOME? Maybe on both?
blfr··on AI usage patterns in software teams
It's hard to measure ROI so we generally wouldn't. Even McKinseys and BCGs have a hard time showing ROIs.

Sure, you could say all digital transformations are nebulous snake oil and don't deliver but I just don't think this is the case.

blfr··on AI usage patterns in software teams
> Usage does not correlate with valued output or ROI

This sounds possible but how can we know? It could just as well correlate. Effort of all kinds correlates with success even when it's not obvious or not... linear.

blfr··on GPT-5.6 Sol Pricing Cut by 50%
There is a massive difference even between Opus and Fable, same provider, before various harnesses and other optimizations come into play. Don't be deceived by rankings and benchmarks, try for yourself.
blfr··on India has paved the way for charging merchants a fee on UPI transactions
Standard VAT in Poland is 23% so taxes usually dwarf fees.
blfr··on India has paved the way for charging merchants a fee on UPI transactions
Only the interchange is capped. The merchant has to also cover network and processor fees. And then there are some uncapped transaction supposedly so the fee covers this risk. Comes out to 0.6-0.7% for a smaller merchant with the card present.
blfr··on India has paved the way for charging merchants a fee on UPI transactions
Isn't 0.5% very competitive and basically nothing compared to taxes on the transaction?

Here in Poland we have a budding "save cash" movement. And they make some good points about freedom and privacy but are the loudest about the processing fees when mostly it's a way for smaller merchants to avoid paying taxes. Which is fine, I think they should be exempted anyway, but let's not pretend that it's the 0.5% in fees rather than the 30% in taxes.

blfr··on On AI regulation and messaging
I am definitely not impressed by Opus. Quite the opposite usually.
blfr··on On AI regulation and messaging
What's moving the goalposts? I am very much amazed at what Fable can do. I push its code straight to prod.

But I am just as amazed with how little real life consequence it seems to have! Even software houses were hit more by interest rates than by this magical revolution.

If I couldn't directly observe Fable in action, I wouldn't believe in AI.

blfr··on Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing
The main reason we don't see much quality degradation in LLM writing output is because they're already poor writers. This is the load bearing reason.

I was bulding a small interpreter and writing an article in ~markdown yesterday with Fable. And while it codes like a pro, it writes like a sixth grader.

Let's see how these watermarking stats hold up if/when llms start writing well.

blfr··on On AI regulation and messaging
First, it's not three years since. The real improvements in programming ability arrived in the last 6-8 months.

Second, the Internet didn't show up much in GDP and similar measures either!

But your point stands. Where are the amazing digital products/stuff? I get that it might take time to arrive as we scale up compute and learn new paradigms. But so much infra already exists (deployment pipeliens, everyone reachable on a smartphone) that we should be seeing something.

blfr··on The Life and Death of Direct File [pdf]
They lost 18 months in decision-making, compressing their development timeline.

I am not seeing it happen yet but I wonder if this will be the main AI SDLC improvement: you can analyze a ton of docs, build prototypes smoothing the decisions for non-technical stakeholders, and compress development so much that it might actually absorb all the other inefficiencies.

blfr··on Asus Bike Booster
There's usually a fairly small number of criminals in the population and they have short careers. Otherwise, the whole enterprise of society wouldn't be viable.

So once you take them off the streets for a couple of years, or decades for the more hardened ones, it's fine. It has an added benefit of breaking criminal networks so the future adepts have no place to fence the stolen goods.

blfr··on The Problem
They don't have a great track record with their predictions.

https://foom.hyperplex.org/

Doesn't exactly move the needle on the odds of our extinction but neither does anything they publish, apparently.

I personally sleep soundly because I don't think there's any going back. We will either escape drudgery or die. Which is fair and as it should be.

blfr··on Yadda 3.0.0: BDD in the Age of AI Agents
Very impressive that it was done with Opus 4.8.

My experiences with Opus have been such that I always max out my Fable allotment but rarely exceed 40% of the remaining limit on Claude.

blfr··on Ordinary Abundance
The same on a smaller scale happened with modern consumer computing. We like to complain but

* have OLED screens and dark modes which are amazing after sundown,

* virtually everything works on Linux, especially all effectively mandatory/government services (at least here in Poland),

* and with some hardware selection Linux works everywhere,

* batteries last ages and everything can be charged via USB-C,

* smartphone cameras now finally exceed my own vision, have amazing zoom and take great nighttime photos (although not of moving stuff yet),

* the ecosystem allows to magically summon a car,

* can have an e2e encrypted multimodal (video, photos, documents, text) conversation with virtually everyone we know, wherever they are in the world, live or asynchronously,

* not to mention LLMs just doing things for you like a tireless assistant.

All of these were a dream when I first set up a home server in 2004. Some like OLED were a dream I didn't even know I had, the LCD glow just annoyed me.

blfr··on Gemini 3.7 Flash
I find Fable completely unbeatable for anything code-related. It's the only frontier model that seems to come with sane defaults.

If it implements something simple like a file export, it just knows that the file should have a meaningful name. Vibecoded feature beats most software's lazy "untitled.png".

So, yes, I want the smartest model even for simple stuff. Maybe especially for simple stuff because the tokens burned will be trivial so the cost doesn't give lower models a comparative advantage.

blfr··on Tuxedo No. 2 – Cocktail recipes
Very cool. French 75 is more interesting in its original formulation based on cognac rather than gin though.

https://tuxedono2.com/french-75-cocktail-recipe

blfr··on Anthropic CEO reportedly worried new hires only care about money
People give him a lot of shit for these antics. The "regulate us!" cries, disaster-marketing of escaped models, now this. I am slightly annoyed to be paying for a product (Claude Max) with such eyeroll worthy marketing.

But it works. Fable is truly amazing, maybe not head and shoulders above the rest but at least a head. Apparently, this neurosis-driven development does attract AI talent.

It suprises me. It shouldn't because the same target demo has been following Eliezer Yudkowsky for decades now but it still does. I just don't have a theory of mind where this is an attractor.

blfr··on Oracle cut its Always Free ARM limits to 2 OCPU / 12GB, enforced Aug 18
Yes, it definitely worked. I had three servers with them for years and only added a card this year because it lifted bandwidth limits on the ARM.
blfr··on Oracle cut its Always Free ARM limits to 2 OCPU / 12GB, enforced Aug 18
I'm a PAYG subscriber and the notice on my account says

"Always Free Ampere A1 Compute entitlements have changed. This update does not affect your paid resources, existing paid workloads or bill. Contact support for further questions."

blfr··on AI Mania: From Tulips to Tokens
We created the content their systems are built on, yet they won’t even share how much energy each query uses, how much water goes to cooling

"You read my blog so now show me your utility bill" is... an interesting argument.

blfr··on Is the Industrial Revolution a good precedent for explosive growth today?
When I was a kid, my dad would schedule a monthly visit to the bank. This is one afternoon freed. With 21 in the month, solid 4% of afternoons saved to do something else. Plus the peace of mind because you can check whatever you want whenever you want.

With e-commerce my shopping is down to just groceries. Clothes, electronics, tools, spare parts, furniture, all show up at my door magically without hunting them down in stores. I remember a small repair being half a day. It's an hour now.

My dad certainly did not and could not have maintained a 12 cafe outings a month habit. I can do it with a newborn. Still probably deflationary compared to his activity.

blfr··on Is the Industrial Revolution a good precedent for explosive growth today?
I think the speedups and conveniences of modern IT and the Internet have been massive: I can bank from my pocket, review the best photos out of 100k+ just when riding a tram, get virtually any good produced anywhere delivered to my doorstep in 1-3 business days (and it will be the exact kind and make I chose, not something that happened to be available), read virtually any book ever published while my appointment is delayed.

However, almost all of these are purely private benefits that do not show in any ledger. This is at least in part why the Internet seems to have had such small impact on gdp.

How about AI? So far, even more private benefits than the Internet.

blfr··on Is the Industrial Revolution a good precedent for explosive growth today?
Will we see any growth acceleration? The Internet doesn't seem to have had this effect. AI may turn out to nothing in terms of gdp growth or even be hugely deflationary.
← PreviousPage 4 of 34Next →