HNHacker News
TopNewBestAskShowJobs

nostrebored

2,711 karma · joined March 20, 2015

Opinions discussed here are my own.
submissionscomments
nostrebored··on I would like the value of my home to rise, while my property taxes fall
There is no inalienable right to live in a place because you were there first.

Housing is a fungible good. I have friends and family living in squalor because people have decided sentiment is a valid reason to keep people in these four walls instead of those four walls.

I think people’s primary residence should not be seen as assets in the way they are today at all. The only reason they are is substantial government intervention.

nostrebored··on Grieving the loss of details
Right, but performance testing has always been hard and LLMs are awful at it. So maybe your coworker was right?

I’m definitely on the “just use the clanker” end of the spectrum. But knowing when and is good. Constantly having to argue with people’s reposting of Claude’s idea of what’s wrong has left me burned out. At this point, if I have any idea about what’s going on and someone tells me “Claude says”, I will ignore them.

In this case, it sounds like you could have made a ten line repro that shows the problem. Why not just send that, something human interpretable?

nostrebored··on Typesafe AI raises $870M at $7.5B
Jev does not have the properties you’re talking about. It’s not very good at general knowledge, it frequently approaches chance level classification on business domains, and there is so much room for someone to do better.

Doing better, I suspect, will be throwing a bigger model underneath and doing something diffusion like (we’ll allocate a whole block to compute rather than just the constrained space) that doesn’t compromise on latency.

nostrebored··on Typesafe AI raises $870M at $7.5B
Yeah agreed that vision is a nice fit here. I’ve gotten some really impressive results by just throwing xgboost on 25 classification related questions.

System one models do not have an obvious use in voice ai, unless you are choosing from a list of potential utterances (eg embed -> top k -> choose from top k)

nostrebored··on Typesafe AI raises $870M at $7.5B
This is really not a good idea.

Modern stt is baking semantic EOT into the prediction. It makes sense. Conversation has many potential ends, and you have to deal with misprediction of current words that may signal completion, but mel spectrograms contain a ton of information as well.

In your version, we’d have to deal with partial transcriptions and a system that is not natively audio aware. It would also be much more expensive.

nostrebored··on I would like the value of my home to rise, while my property taxes fall
Why should this person take up a family home in a desirable area? It is inefficient. The pressure she is feeling to leave is the system working as intended.
nostrebored··on Programming Isn't Special
Idk I feel like it falls over in the same way. Poor understanding of abstraction, poor use of metaphor, an inability to really understand or plan through a thought from being to end.

People are okay with code that does stuff, however bad it may be.

But at some point it becomes impossible to understand and reconcile, which is something humans struggle with as well. Think of fantasy series that will never culminate, because there is no possible culmination.

nostrebored··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
plenty of harnesses use the claude subscription
nostrebored··on Decisions API is in public beta
i would be shocked if luna decides is less generally capable than jev. jev has failed to understand any novel domain I've given it. i have found that i use it only when "some data is better than no data"
nostrebored··on EmbeddingGemma 2: An open, lightweight multimodal embedding model
It's been awhile since I've been in the space, but Voyage was never a serious contender outside of super-niche business domains. I suspect that this compares favorably in 9X% of use cases
nostrebored··on LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents
… what large advances and at what cost? seems to me that muse 1.3 is kind of a thing. I doubt it will make meta very much money.
nostrebored··on Clef: Open-weight decision models, and new RL fine-tuning platform
Diffusion Gemma sucks to train if you are not already mostly in distribution. The reason diffusion Gemma is fast is a learned denoising that balances quality and speed. So the further your data is from what already happens, the slower it gets. In almost every case I’ve tried, you might as well just train Nemotron.
nostrebored··on ChatGPT Pro 500
Supposedly much less expensive per task (more token efficient and cheaper)
nostrebored··on GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price
I think you have probably started using OpenAI recently -- one draw used to be that it was really, really hard to ever hit limits. If you did, you probably had usage resets available.

I think this is still true provided you're not using Astra.

nostrebored··on Ember-1
I can’t help but think it’s more expensive tinker.
nostrebored··on Ember-1
Agreed, I think the only place where it’s still interesting is ui design. Visually kimi and muse feel much nicer than frontier models to me, but maybe it’s an artifact of everything terrible being Claude Design
nostrebored··on Mercury 2.5 LLM hits 770 tokens per second
depends on your use case I think. for small, transactional tasks it is quite strong and fast.
nostrebored··on Mercury 2.5 LLM hits 770 tokens per second
Try Celeris-magnus-1. We get similar speeds and it’s much closer to qwen 27B dense models.
nostrebored··on Jev in 25 Lines of Python
Yes, for vision classifiers we have in prod, I've seen a huge difference between A, B, C, 1, 2, 3 style answers and emitting a string. Even from just base model behavior pre-sft/rl. It was one of those obvious in retrospect moments.
nostrebored··on Grok 4.7
Can you point at the marketing in this release to show me where that’s the lean?

Many people are allergic to “AI safety” as they perceive it as an attempt to deceive them. When they ask a factual question and get back an unfactual answer, it makes them upset. None of the people I know who feel this way are searching out CSAM. They accurately perceive that there is a team that wants answers to come out a certain way.

For instance, I’m loosely connected to people who care about animal safety in AI research. There are absolutely people spending their time pushing a narrative about _the_ ethical way to interact with animals. Feeling this not being forced onto you is kind of nice.

nostrebored··on Grok 4.7
Ah gotcha, not on twitter so just hadn't seen them before!
nostrebored··on Grok 4.7
Is this an ad for Vals AI? Looking at their website, the rankings don't mesh with my observed utility for almost any model outside of fable and astra being good-ish.
nostrebored··on Gemini 3.8 Live and 3.8 Live Extended Thinking
In my experience actual experts don’t hedge because they have a perspective. They might say “I think”, but avoid weaseling.
nostrebored··on More questions about whether researchers can trust OpenAI with unpublished math
what benefit do they get from making the statement? they could just say nothing. saying it and having it be untrue opens them to legal issues that are not worth the risk for this nothingburger.
nostrebored··on Y Combinator Early Access Network
That’s a good question! A mix of both. There are definitely people who see it as a step on their validation treadmill. Most people want to make money. But a lot of people are just techno optimists who think they can do something.
nostrebored··on Growing proof that autonomous cars save lives
Huge amounts of insolvency is due to supporting this failed urban planning experiment. Urban centers are significantly more economically productive and significantly less expensive. Carism leads people into this trap when they should be advocating for significant expansion of dense cities and transit.
nostrebored··on Growing proof that autonomous cars save lives
From what I’ve been able to read it seems like it was a net positive public health intervention that’s been horribly dramatized.
nostrebored··on Mercury 2.5
latency, instruction adherence, reliable tool cools, conversationality are all in tension.

it's great when you can get a 170ms ttft. but if you have 700 ms endpointing on the stt side and 300ms ttfb on the voice side, then you haven't really made something super snappy.

nostrebored··on Y Combinator Early Access Network
Having been through YC, there's certainly some of this, but the overwhelming majority of founders were pretty down to earth. People generally just want to solve hard problems.
nostrebored··on What is nueralese and why is it bad
The objective of most large scale post training i am aware of is not to make think traces or to make them “more English and answer aligned”. It’s to achieve a result. That’s why I am calling it incidental
← PreviousPage 2 of 34Next →