mrinterweb

5 ms·

1.
by mrinterweb 7d ago
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.
2.
by mrinterweb 7d ago
> 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
3.
by mrinterweb 12d ago
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,
4.
by mrinterweb 13d ago
I saw the version of this video with Paul Rudd (Celery Man) https://youtu.be/a8K6QUPmv8Q?si=TWmoNhxYAPp73TKg
5.
by mrinterweb 14d ago
I'm really curious how GML-5.3-flash would do. Very affordable, and it seems to do pretty well with 3D modeling.
6.
by mrinterweb 21d ago
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-m
7.
by mrinterweb 21d ago
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.
8.
by mrinterweb 21d ago
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.
9.
by mrinterweb 1mo ago
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&#
10.
by mrinterweb 1mo ago
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.
11.
by mrinterweb 1mo ago
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 handi
12.
by mrinterweb 2mo ago
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 HT
13.
by mrinterweb 2mo ago
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.
14.
by mrinterweb 2mo ago
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
15.
by mrinterweb 2mo ago
Two that I use are: * Openrouter.ai for a hosted router * https://github.com/diegosouzapw/OmniRoute for a local router
16.
by mrinterweb 2mo ago
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.
17.
by mrinterweb 2mo ago
Open weight models are much more auditable than closed models, but could still hide backdoors that could be near impossible to detect.
18.
by mrinterweb 2mo ago
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 publ
19.
by mrinterweb 2mo ago
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 lik
20.
by mrinterweb 2mo ago
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
21.
by mrinterweb 2mo ago
I'm a big fan of local models, and moving inference from the cloud to local machines is great, but there's a couple potential problems with this. LLMs take significant (V)RAM resources to run (which is in short supply on consumer
22.
by mrinterweb 3mo ago
PC gaming on linux these days is joy these days. The main games you're likely to struggle with are games that require some windows kernel level anti-cheat software running, so some online multi-player will not be playable for that reas
23.
by mrinterweb 3mo ago
I few months ago, I backed up my windows gaming machine and overwrote the partition with CachyOS. Haven't looked back. Gaming performance and compatibility has exceeded my expectations. Just a much better experience overall. I feel sor
24.
by mrinterweb 3mo ago
I have a lot of hope for local AI. Local model intelligence has come so far from where it was just a few years ago. It is about a model's intelligence density now for local AI. Models like Qwen 3.6 are truly capable. Sure Qwen 3.6 isn&
25.
by mrinterweb 3mo ago
The biggest concern is identifying "who". If the US government says only US citizens can access a model, how do they enforce that. Anthropic and OpenAI will use Persona (a company funded by Peter Thiel) to verify user identity. Ve
26.
by mrinterweb 3mo ago
I don't call people rude for doing it. Maybe I should. I consider it rude. Maybe I should inquire if the earbuds are for hearing assistance then mention they are rude if they say no. Not really my style, but at least it would be direct
27.
by mrinterweb 3mo ago
> And people now don’t feel neglected when you keep the Pods in your ear. I disagree with this. Pods in ears are essentially a "do not disturb" sign for most people. Being around people who regularly have the "do not distu
28.
by mrinterweb 3mo ago
MV3 is not an improvement over MV2 for the purposes of ad blocking.
29.
by mrinterweb 3mo ago
This would have made my life so much easier. I wrote a medical scheduling calendar application about 1.5 years ago.
30.
by mrinterweb 3mo ago
It kind of sucks, but I get the silent change. If a user was trying to use the model for something untoward, having a rejected prompt would just give signal to train on how to eventually successfully bypass security measures.

More ›