ricardobeat
ricardobeat on Hacker News
Did you use Fable, set Opus 5.5 to xhigh, or use 'ultracode' / workflows? That's the only way this will happen.
In my experience a single 5 hour allowance in the 20eur Claude plan equals about $10 in Deepseek 4.1 Flash API usage. DS is a lot more token hungry and takes 3x-4x longer. The subscription is way cheaper, probably around $200/month worth of DS, just stick to Opus low and Sonnet for most tasks.
This is a misunderstanding of the purpose of UUIDv7. The whole point is to keep rows physically close, so you can very quickly extract an entire time range from a number of database shards, with minimal reads and compute. And be able to generate them without a centralized server. That's all there was to it, and something you'd only worry about when serving billions of records every hour.
They are also not supposed to be used as public IDs, just like you wouldn't use auto-incrementing IDs that way.
(I wrote a library for this 13y ago [1], around the time Twitter, Instagram et al started reaching enough scale to think about this kind of optimization)
[1] https://github.com/ricardobeat/unik
> we figured out how to train Dessa on only one NVIDIA GB200 NVL72.
This is the crazy part. It seems pretty smart, and LLMs require 1000x more compute to train. Can this kind of efficiency somehow be achieved with text too?
I had the lowest plan ($20/month I think) and it literally gave me fifteen, maybe twenty minutes tops, using the cheaper K3-256K variant. Basically enough to fill the context about 1.2x and that was it.
It's a really smart model but way too expensive. The same $20 goes 100x further with Claude. I used it for the initial task planning while the plan didn't expire, then handed off to DeepSeek or Minimax.
Netflix has "optimized" their app to the point where you're cycling between the same 30-40 recommendations on every screen, just grouped a little different. The catalog shrinks every year. And they invest large amounts into producing their own algorithm-based content instead of acquiring more.
It is not. The exact wording, as it sits in my inbox:
> We’ve clarified that all Service Plans may include: (i) promotional content, (ii) sponsorships, and (iii) advertisements before/after playback of Content and in channels, live/as-live, special events, and any third-party services content.
Emphasis on "before/after playback of Content" and live etc. So they are opening the door for ads everywhere, not just on live TV or Hulu.
How high can it go? It's already €16/month over here for 4K. This is getting into cable channel territory, except it had a lot more variety.
But that’s the gist of the question, can’t their internal frontier models provide that infinite attention? That is what their PR implies.
This is the central plot in the “Life 3.0” book that came out a few years ago.
[SPOILER] After the singularity, AI is able to dominate all markets – starting by gaming and entertainment – in order to generate cash to rent more servers to improve itself.
Crystal never aimed at Ruby interoperability, it's
inspired by Ruby's ergonomic syntax.
You should learn this language if you want the nice developer experience, great stdlib, managed memory, native performance. Similar reasons to why you'd learn Go.
Items 1-5 at the end of the article basically describe a supervised learning process. Data + labelling, updating weights, calibration.
Yes, using an LLM for confidence scoring while generating prose is a bad idea. I expected a mention of decision models here, as that exactly what they are trained for.
It’s also incredibly annoying to navigate on mobile, due to everything being in separate pages and quite slow to load.
This variety of slop is really easy to automatically detect (for now).
A note on Meta AI was making the rounds the past few days: it will happily generate images featuring the right-wing candidate, logo or party number, while consistently refusing to generate anything related to the opposition Lula/PT/13.
They claimed it was a 'technical error'.
In September, as the elections geared up, they also suspened Lula's campaign account due to another 'technical error'. There are also thousands of complaints from users having their posts blocked.
It doesn't have to be one or the other. Having a machine that does most of the coding for you means you can have more of the collaborative, human input. You can focus on the UX, you can fix all bugs, you can experiment with new ideas faster than ever. There will be zero excuses for poorly done work - think of all the half-assed corporate apps you have had to use during your career. I think this is the start of a great era in software development.
That’s not the case anymore. I’ve have completed projects in several slightly obscure languages (including ones I created) and frontier models have almost no trouble at all, other than having to reference documentation very often.
Your comment makes little sense - is it about free will? That has no relation to the quality of comments.
This used to be a place for intelligent commentary, learning new things, not just participation, rants or funny remarks.
How about a sous vide bath? Could go as long as 72 hours before it turns into complete mush.
I'll watch the video in full when you read the paper. It starts with obviously flawed math: mortality rate decreases + 'lives saved' doesn't have to add up, and a flat death rate does not equal "no change" because of aging and other factors. It then jumps into mischaracterizations of the study, completely ignoring the data, "it's just a lot of numbers to make it look sophisticated" and false allegations like that excluding high income countries (which do not receive USAID) is manipulating the results. The author clearly has no idea what the scientific method is just randomly finding "holes" in the paper.
The 90M figure is probably exaggerated, but this video is completely nuts. This is exactly why the internet has failed us.
HN is not a place where people post content
made for HN (or at least it didn't use to be). Relevant links are posted. They get upvoted if they are interesting. Articles from many other fields like physics, cryptography, hardware design, crocheting, nuclear safety, farming, etc, are posted very often and can still be an interesting read for people not familiar with the subject matter. Finally, engineers have a wide variety of backgrounds, and I bet a good chunk can read math notation or drive a soil compactor - there is no "us".
They might not all be a good read for you in particular; the choice is yours to make an effort, or "move on" as someone else suggested. You're being downvoted because you are making it sound like you are entitled to an easier explanation, an hypothetical version of the article that caters to you in particular. There are many ways you could have framed your comment differently without that baggage.
You'll have to disagree with a
lot of scholars, in addition to this HN comment, to make your point. This is the '90 million' study:
> This finding indicates that 91 839 663 (95% CI 85 690 135–98 291 626) all-age deaths, including 30 391 980 (26 023 132–35 482 636) in children younger than 5 years, were prevented by USAID funding over the 21-year study period. USAID funding was associated with a 65% reduction (RR 0·35, 0·29-0·42) in mortality from HIV/AIDS (representing 25·5 million deaths), 51% (RR 0·49, 0·39–0·61) from malaria (8·0 million deaths), and 50% (RR 0·50, 0·40–0·62) from neglected tropical diseases (8·9 million deaths) [1]
30B+ is pocket change for the US, but makes a massive impact in poorer countries.
It also had a soft power effect that the current US government seems completely oblivious to [2].
[1] https://www.thelancet.com/article/S0140-6736(25)01186-9/full...
[2] https://www.jstor.org/stable/23353912
[3] https://time.com/article/2026/05/15/usaid-shutdown-rise-glob...
They said that only the "thin upper door trim" is glued on.
My bad, maybe I read it wrong or the comment was edited, but I was referring to LLMs in general. You won't get a fair coin flip from the model either way.
Try asking it to write a program to predict coin flips. That's what it is trained to do.
In a private benchmark I have, it performs much, much worse than Qwen 35B A3B for general knowledge. Still, probably beats Qwen 9B at much higher speed.
Both of those are phishing attacks. You are right about NPM though - a good reminder to avoid banking on developer machines.
> fake or infected Zoom binaries, Xcode binaries, etc.
There are so many hoops to go through to pull something like that off, it's basically impossible. It would be a "the great train robbery" kind of effort. I also don't think an npm package could alter an existing app or obtain screen access otherwise, unless you're running `sudo npm install` routinely.
Some form of phishing would be way more likely to succeed, luckily you can use your own brain to protect against those attacks.
It's a problem similar to "give me a random word" always returning "zephyr" and "luminous", or guessing the numbers 42/27.
It also affects writing, if you ask for a story with a minmal prompt something very similar comes out of all of them. Often fixated on clocks.
Is it though? There is nothing you can solder onto it that will gain any meaningful access and go unnoticed. Tampering with the supply chain is a lot more difficult, easily detectable, and wouldn't scale. I can use 100% open source software, in an OS where it's signed and completely safe from interference - except the NSA of course, but they are not interested in your money.
For example, in MacOS it would be extremely hard to "sniff the display" as in this attack, unless you take obvious steps to authorize screen access.