HNHacker News
TopNewBestAskShowJobs

criemen

395 karma · joined June 28, 2025

submissionscomments
criemen··on Sonnet 5.5
The latest Haiku release is almost a year old. Clearly they don't care about the small-but-capable part of the market at all.
criemen··on Ember-1
> to what end?

I'd expect that their business strategy is to compete in more markets, and if successful, they can capture more value. This is the "easiest" for them as they already have GPUs, a training environment etc. For that platform it's not the worst if there's an internal customer team that can help shape the future and provide immediate feedback, and if it results in a good model, even better.

Other things I'd not be surprised they offer in the future in the same vein: A multi-model harness, coding agent (cloud and local), and maybe at a later point in time even a CPU-only cloud compute product.

criemen··on Analyzing Frontier Model Progress with My Favourite Game: Prince of Persia
My apologies, and it's too late to edit the parent now.

To be fair, though, I mainly ran this experiment to distinguish models, not to see if they were better than humans or not, so for that purpose it didn't really matter if the basic conditions were the same or not.

criemen··on Analyzing Frontier Model Progress with My Favourite Game: Prince of Persia
Thanks! I went back, and had a look at both the contest rules and the highest scoring submission again. Turns out I benchmarked against infinite games, whereas the contest measured a 5min time limit. So, Astra solved a different problem.

Comparing strategies, though, Astra does something pretty different from the highest scoring submission. It doesn't predict the RNG, instead it reacts to the visuals on-screen, planning ahead by estimating velocity of all objects on screen. Unlike the winning solution, it actively flies the ship, whereas the winning solution basically only rotates and teleports. So Astra behaves more like a regular "perfect" player would, rather than one that breaks the PRNG.

criemen··on Analyzing Frontier Model Progress with My Favourite Game: Prince of Persia
A German magazine (c't) ran a Asteroids programming contest in 2008 - create a client with access to the emulator output of the game that provides keyboard inputs to control the game.

The highest scoring submission that won the contest had a high score of around 137k. Last week, I had GPT-6 Astra, Sol and Luna implement and hill-climb on this task, as I wanted to see how big the difference in smartness is. Luna implemented something, but never exceeded ca. 20k points, with a large variance. Sol got something in the area of the humans implementation.

Astra, which finished fastest, had a highscore of around 1.7Mm when the game seemed to fairly reliably crash. On the way, it disassembled parts of the ROM to extract information about the game.

I didn't do a ton work to document and measure the specifics, but it was very impressive.

criemen··on What even is an OS now?
My colleague Steve Sanderson showed off VibeOS already back in June https://www.youtube.com/watch?v=7NfyZhV1dKM

Sounds like this is a more serious and phone-centered take on what Steve demoed. We sure live in interesting times.

criemen··on Too AI; Didn't Read
I hope that labs are actively working on the voice and writing styles of their models. I don't want to read slop in large part because their prose is overly verbose and complicated. If the models were writing simple sentences in mainstream English, mirroring the vocabulary used by my colleagues, I'd mind less than the current status.

Surely that's something that can be trained for.

criemen··on Typst makes big strides
To paraphrase Churchill, LaTeX is the worst form of typesetting experience except for all those other technologies that we've tried.

Glad that that's no longer the case!

criemen··on Japanese used bookstores see 5x sales surge as books are being bought by the ton
They're targeting non-fiction books, so romance novels would be out.
criemen··on The current balance of power in open models
If you don't believe AWS follows their ToS, you shouldn't use any cloud provider for CPU/data storage/anything else either. That's not a mainstream position in the industry.
criemen··on AI coding has made CI a bottleneck, so we reworked ours to keep up
well for test runs you kinda need the binary to run, but you're correct, if the job is just "does this build" no download necessary.
criemen··on AI coding has made CI a bottleneck, so we reworked ours to keep up
If you don't need to rebuild anything, bazel can fetch only the final artifact (not the intermediates) from the remote cache.

Also, if you have persistent CI workers with a persistent bazel instance, you save on some network roundtrips, but that's obviously harder to set up and make bulletproof.

criemen··on AI coding has made CI a bottleneck, so we reworked ours to keep up
Nothing substantial, no. I've never personally experienced builds to be so slow to begin investigating bazel as an option for JS/TS.

Tsgo, oxlint, caching dependencies etc. what linear outlined in their blog post would be more impactful for the average TS project I've worked on.

criemen··on AI coding has made CI a bottleneck, so we reworked ours to keep up
I lead a bazel conversion for a pretty complex piece of software written in 5+ programming languages and shipping native binaries to all 3 major OSes a few years ago, and it took multiple years to get it done.

For a less complex project (1 programming language, still shipping to all 3 major OSes), with my knowledge and agents I got the bazel conversion done in 2 weeks.

The setup cost for bazel just went down by a lot, and I don't think the industry as a whole is aware of that yet.

criemen··on AWS says it can't restore some data from mideast facilities struck by Iran
Do cloud providers even share data center locations so you can assess the "far enough" bit yourself?
criemen··on An atlas of periodic solutions to the three-body problem
While I love the books as much as everyone else, the pedant in me has to point out that the Trisolareans have a four body problem at hand (3 suns and their planet), not a three body problem.
criemen··on Things That Annoy Me About Cars
> Using a phone as a key is super convenient, and having a little credit card sized NFC key works great as a backup. I've rarely had any problems with it.

The author elaborates that it only works intermittently for them, providing a long list of links to similar complaints?

criemen··on The GDR and Vietnam: From Fake Coffee to Coffee Empire
It's even funnier: Having both "democratic" and "republic" is redundant - by definition, a republic is democratic. Only non-democracies feel the need to double down on the democratic part.
criemen··on I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes
I believe this is part of the complaints of new models taking longer/requiring higher spend - they go the extra mile on verification, regardless of whether their change is correct already or not. So on problems that an earlier model one-shotted an answer to and did some lighter verification, the newer models might take longer to come back to the user due to running all the tests for your software they could find.
criemen··on Europe EV Sales BEVs Jump 50% & Reach 26% Market Share
Sweden, Finland and Iceland would be part of any definition of Europe? They're all EU countries, and (irrelevant to that) are now all NATO members too.
criemen··on Benchmarks in Leipzig
Partially, 2.2 Submission workflow W2 deals with this:

> Stage W2 The five project-active models, see Table 2, attempted the question. Their answers were compared to the original answer by an LLM judge. If at most three models answered correctly, the contributor could proceed.

So "trivially contained in the training data" is excluded, as then all models could/should easily come up with the solution.

criemen··on DeepSeek reasonix, DeepSeek native coding agent with high caching and low cost
> Ah, reminds me of good old "There are only 2 hard problems in computer science: cache invalidation, naming things, and off-by-1 errors."

You quip, but LLM KV caching (from the harness side) is quite easy: You get a cache hit on stable prompt prefixes, period. That means you want to keep the prefix stable, and only append at the end of the conversation. Made up example: Don't put the git branch name into the system prompt part (that comes first), as whenever the branch name changes, that'd trigger a cache invalidation of the entire prompt.

Getting this right requires some care to not by accident modify the prefix, basically, and some design on communicating the things that can change (user configuration, working dir, git information, ...).

criemen··on Project Glasswing: An Initial Update
> There's also a runaway effect of model improvement from the discovery, triage and fix data. This is likely already the most potent corpus of curated offensive data ever assembled and will only get better.

But that corpus of data is accessible to all competitors, American or not. I don't believe that this can't be replicated. I'd posit that there's enough annotated data out there (CVE+patch), only increasing thanks to Mythos, that if you specifically RL for this scenario, you can improve your models performance on finding vulnerabilities without access to Mythos.

criemen··on Cursor Introduces Composer 2.5
Well is that a statement about the quality of Opus 4.7 or about compose 2.5? :P
criemen··on Measuring Claude 4.7's tokenizer costs
> Is there an equivalent ultra-high-end LLM you can have if you’re willing to pay? Or does it not exist because it would cost too much to train?

I guess at the time that was GPT-4.5. I don't think people used it a lot because it was crazy expensive, and not that much better than the rest of the crop.

criemen··on Measuring Claude 4.7's tokenizer costs
> it's them trying to push the models to burn less compute

I'm curious, how does using more tokens save compute?

criemen··on Škoda DuoBell: A bicycle bell that penetrates noise-cancelling headphones
Pretty cool if true!
criemen··on No, it doesn't cost Anthropic $5k per Claude Code user
> What people don't realize is that cache is free

I'm incredibly salty about this - they're essentially monetizing intensely something that allows them to sell their inference at premium prices to more users - without any caching, they'd have much less capacity available.

criemen··on Giving LLMs a personality is just good engineering
I tried ChatGPT over the holidays (paid) vs. claude.ai (paid). After trying some prompts that worked well on Claude in ChatGPT, I understand why people are so annoyed about AI slop. The speech patterns in text output for ChatGPT are both obvious and annoying, and impossible to unsee when people use them in written communication.

Claude isn't without problems ("You're absolutely right"), but I feel that some of the perception there is around the limited set of phrases the coding agent uses regularly, and comes less from the multi-paragraph responses from the chatbot.

criemen··on MacBook Air with M5
> Out of curiosity, what are some good use cases for a MBP now with the MBAs being so powerful?

Local software development (node/TS). When opus-4.6-fast launched, it felt like some of the limiting factor in turnaround time moved from inference to the validation steps, i.e. execute tests, run linter, etc. Granted, that's with endpoint management slowing down I/O, and hopefully tsgo and some eslint replacement will speed things up significantly over there.

← PreviousPage 2 of 5Next →