Hacker News: mrinterweb

New comment by mrinterweb in "Mercury 2.5"

mrinterweb — Wed, 09 Sep 2026 15:58:42 +0000

The benchmark comparison to other models was suspiciously missing. The self-comparison is a good representation of progress, but it is light years behind frontier models. Speed is great, but wrong is much worse than slow, IMO.

New comment by mrinterweb in "Claude, change the "Add to Cart" button to blue"

mrinterweb — Wed, 09 Sep 2026 15:34:13 +0000

> triggered memories

Yeah of 10 minutes ago. It is shocking how long some seemingly simple things can take. I know there are some things I can do faster than the LLM and some things it can do faster than me. The amount of rambling BS is the exhausting part.

New comment by mrinterweb in "The largest electric aircraft just flew [video]"

mrinterweb — Fri, 04 Sep 2026 17:13:04 +0000

I think lithium air batteries will be the real inflection point for aeronautics. Li-air has potential density 12 kWh/kg. CATL is focused on this tech. May be some years before they are able to achieve the full potential density. Still, as the other commenters said, the efficiency of electricity make make up that difference.

New comment by mrinterweb in "GPT-6 Astra"

mrinterweb — Thu, 03 Sep 2026 20:05:59 +0000

I saw the version of this video with Paul Rudd (Celery Man) https://youtu.be/a8K6QUPmv8Q?si=TWmoNhxYAPp73TKg

New comment by mrinterweb in "Fable 5.1 World Modeling"

mrinterweb — Wed, 02 Sep 2026 22:22:34 +0000

I'm really curious how GML-5.3-flash would do. Very affordable, and it seems to do pretty well with 3D modeling.

New comment by mrinterweb in "GLM-5.3-Flash"

mrinterweb — Wed, 26 Aug 2026 19:25:18 +0000

That's wonderful. I was going off an older version of the Artificial Analysis page for GLM-5.3-Flash https://artificialanalysis.ai/models/glm-5-3-flash. The page is updated now to show that it does support multi-modal image inputs.

New comment by mrinterweb in "GLM-5.3-Flash"

mrinterweb — Wed, 26 Aug 2026 16:26:12 +0000

I really wish GLM models had vision capabilities. I've worked around that in the past to use a vision MCP in my harness that GLM can call. It is not the same, but it allows the model to query images.

New comment by mrinterweb in "GLM-5.3-Flash"

mrinterweb — Wed, 26 Aug 2026 16:23:34 +0000

Give it a couple days, and there will be plenty of other inference companies hosting it. Don't like z.ai's TOS? Use the model on a provider with TOS that you agree with.

New comment by mrinterweb in "Why does Opus 5 feel worse to work with?"

mrinterweb — Fri, 14 Aug 2026 18:28:53 +0000

The pain points in the article do not bother me. I'm bothered by Opus 5's verbosity. It is so long-winded and you have to read through verbose outputs to mentally distill what is important. It is exhausting. I don't think I've ever started skim reading LLM output more that I do with Opus 5. I use the caveman skill, and I think that does help some.

New comment by mrinterweb in "Beating GPT-5.6 Sol on retrieval with 100x cheaper open models"

mrinterweb — Wed, 05 Aug 2026 20:13:00 +0000

Exactly. There could be a lot of value for inference companies to do this. Could save a lot of money being able to hand off highly repetitive known tasks to far smaller specialized models.

New comment by mrinterweb in "Beating GPT-5.6 Sol on retrieval with 100x cheaper open models"

mrinterweb — Wed, 05 Aug 2026 19:03:07 +0000

There is so much opportunity for purpose built models like this. Ideally a harness should spin up a subagent to offload to targeted models for specific tasks like this. I know this is not a novel idea. Claude code does some of this by handing off the "explore" agent work to haiku. I just love seeing that specialized LLMs are being developed.

New comment by mrinterweb in "Solid Queue 1.6.0 now supports fiber workers"

mrinterweb — Sat, 01 Aug 2026 17:43:22 +0000

This looks fantastic for a common async workflow I use. I often use one job to fan out multiple individual http request jobs. The reason I prefer jobs for this is easy and consistent retry logic, and durability. I want to make sure those HTTP requests eventually go through. Fibers would be much better suited for this. So much of work that goes onto work queues is IO bound, and fibers are a great fit for that.

New comment by mrinterweb in "Solid Queue 1.6.0 now supports fiber workers"

mrinterweb — Sat, 01 Aug 2026 17:37:15 +0000

That ruby fiber vs go goroutine benchmark is interesting. The 4-10x memory use doesn't surprise me, but the near performance does. I'm guessing there is more of a gap with the p50.

New comment by mrinterweb in "Gemini Robotics 2 brings whole body intelligence to robots"

mrinterweb — Thu, 30 Jul 2026 23:42:57 +0000

Look at robotics demonstrations on YouTube from one or two years ago, and compare that to where we are today. It seems reasonable to me that general purpose robotics may be able to outmaneuver your average human in a few years. At least one robot has surpassed what I could do https://www.youtube.com/watch?v=Xha2TJ_1C6Q

New comment by mrinterweb in "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"

mrinterweb — Tue, 21 Jul 2026 23:13:17 +0000

Two that I use are:

* Openrouter.ai for a hosted router

* https://github.com/diegosouzapw/OmniRoute for a local router

New comment by mrinterweb in "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"

mrinterweb — Tue, 21 Jul 2026 23:08:24 +0000

The don't only host open weight models. Also, why not promote this. If Fireworks thinks this big news might convert some new business doesn't make it not true.

New comment by mrinterweb in "Who's afraid of Chinese models?"

mrinterweb — Mon, 20 Jul 2026 22:53:13 +0000

Open weight models are much more auditable than closed models, but could still hide backdoors that could be near impossible to detect.

New comment by mrinterweb in "China’s open-weights AI strategy is winning"

mrinterweb — Mon, 20 Jul 2026 19:47:06 +0000

Undercutting US dominance in AI is huge for China. If the entire narrative is that you have to use Anthropic or OpenAI to access a decent model, then China's AI labs are sitting on the sidelines as some third rate solutions. China publishing the model weights of models comparable to the frontier proprietary models drastically undercuts closed labs dominance. Maybe these Chinese AI labs don't have the billions infrastructures some of the US players do, but they don't have to if the model is open weight. Many inference provider companies around the world have hardware that can run these models and they will happily run frontier class models for people. Starting in 7 days, people will have the option of which of many providers they want to use to access K3.

Making frontier grade models a commodity will make a competitive market where companies compete for business by improving their quality and decreasing their prices. The cost to access frontier grade models will continue be driven down the more competition that enters the market. This commoditization will challenge the valuations of Anthropic and OpenAI.

New comment by mrinterweb in "Kimi Work"

mrinterweb — Mon, 20 Jul 2026 18:51:53 +0000

As soon as the K3 weights are published on July 27th, there will be many US providers hosting the model. I realize model hosting doesn't really apply to kimi work specifically, but in terms of accessing K3, there will be US options likely in 7 days. Just look at OpenRouter to see the providers hosting K2.6 https://openrouter.ai/moonshotai/kimi-k2.6.

New comment by mrinterweb in "SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence"

mrinterweb — Wed, 08 Jul 2026 20:34:25 +0000

I really don't want harness lock-in. I am trying to decouple myself from Claude Code now. I love the model of OpenRouter and being able to switch models at will let's your harness focus on your personal tooling and you can easily switch to the flavor of the month LLM with a single slash command instead of rewiring your entire workflow to use a harness to use a model.

I like Cerabras, but I really wish they would make more of their hosted models generally available.