Hacker News Reader

k__on Pi.dev: You Said No MCP

What would you say is a good harness with subagents?

k__on Livenerf: Has Opus 5.5 been nerfed yet?

Another reason not to use proprietary models for your products

k__on Jeeves. Reasoning improves Jev-like decision models

Not entirely.

I think, it's a bit much to call this "no hallucinations".

Technically true, but in practice you could still choose the wrong result or the probabilities can be off.

k__on Jeeves. Reasoning improves Jev-like decision models

The whole "no hallucinations" premise is based on that.

Like, yeah, you don't hallucinate, but only because you force the user to decide in the end.

k__on Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms

Then again, it's only a very small number and fixed set of tokens for the output.

k__on Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms

No need, as that functionality can run locally no problem.

k__on Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms

Von 1.2 had a better Doom score :D

https://github.com/wfzyx/von

k__on 13 Months Sober (2025)

Yeah

None of my grand parents made it over 65, all of them heavy drinkers.

So, I'd probably have that going for me.

k__on 13 Months Sober (2025)

I'm always a bit disappointed when I look at the effects of staying sober.

All these people writing about how they turned their life around, sleeping better, thinking clearer, getting more healthy, etc.

I stopped drinking 3 or 4 years ago and I don't feel any better for it.

However, I only drank like one or two nights a month, when going on a party, so most of my life I was sober anyway.

The only thing I noticed is that I'm now more awkward in social situations. I mean, sure, I might be healthier than before, but I don't have any numbers for that. It's just a thing I assume to be true.

k__on Ember-1

Half the tokens can also mean quicker task completion

k__on What would a serious AI product look like?

An AI product should probably start with knowing what model (e.g., arch, version, quant, etc.) you're actually using. Opaque providers make that quite hard.

People are constantly complaining about GPT/Claude constantly changing under their apps without notice.

k__on Ember-1

Vertical integration.

k__on Ember-1

With DeepSeek's pricing, no other value prop has been great.

k__on Ember-1

If they can keep up.

The lock-in is less pronounced as it is with AWS or MS.

k__on Turning GLM-5.3-Flash into a Jev-like decision model

Is it non-autoregressive?

k__on Ollaya – Ollama for open-source, Jev-style decision models

Can any model give you a reasonably correct answer on that amount of data?

k__on Ollaya – Ollama for open-source, Jev-style decision models

But aren't BERTs tiny compared to a LLM and can be trained cheaply with the help of an LLM?

k__on Gravity seems holographic. What does that mean for reality?

As far as I understand it,the relative time as we know it, is just something that emerges from the fact that all particle interactions travel with constant light speed.

So if anything moves, the interactions don't change their speed, they stay constant, but the delta to the movement changes, and thus time dilates.

Would be interesting to know if there is some absolute/fundamental global time that has nothing to do with the time we perceive.

k__on Kev: Tiny Jev-like family of decision models built on top of Qwen3.5

Half-OT:

Is Jev a decoder (e.g., BERT) or is it some kind of encoder (e.g., GPT) that just happens to be trimmed down to only outputting a handful of tokens for the answers and their probability?

k__on A Necessary History of the Oddest Letter: W

Fair, but Y isn't used much in German.

k__on Qwen Image 2.1

That's how they're sold, isn't it?

k__on OpenJev

I think so, yes.

However, it might have fewer restrictions than a BERT and/or is smarter (whatever that means).

k__on Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data

Does that mean, it's not compressed anymore?

k__on Vibe Coding is the new Internet Dating?

I didn't have the impression anyone felt ashamed of dating online back in the days and I don't have the impression vibe coders feel any shame today.

k__on OpenArch – PyTorch implementations of modern LLM architectures

I was yesterday years old when I learned that those open weight models need custom code to run.

Somehow I expected inference engines are generic LLM runtimes that can execute any weight.

So, to get this right.

Someone trains a model.

They release the weights and a reference implementation of the model architecture.

Then a provider has to host this model either by running inference via the reference implementation, an open source implementation, or build their own.

Does this mean, providers don't just differ in quantisation and configuration, but also in inference engine implementation?

k__on DeepSeek v4.1 Flash

Update:

I'm using it right now and it's noticeably faster.

I'd also say, it seems smarter, but I think that's because of some harness updates I installed. (I haven't used pi for almost a month)

k__on DeepSeek v4.1 Flash

So, while the throughput was 400-500tps in beta its now ~150tps on OpenRouter.

I was hoping for a bit more, but it's still 100% faster for a very good price, so I won't complain.

k__on iPhone Duo

Nice form factor.

I have a Samsung Fold 4. The front screen is too narrow and the folding screen doesn't open all the way anymore.

Not what I expected from a 1400€ phone.

k__on DeepSeek launching v4.1 flash cheaper and more capable than v4 pro

At least they can use another provider or self-host.

k__on DeepSeek launching v4.1 flash cheaper and more capable than v4 pro

Beta testers report >400 TPS.

https://www.geeky-gadgets.com/deepseek-v4-1-flash-review/

I hope some of those speed increases will make it to production.