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.
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.
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.
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.
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 hardware), and we don't know that Google won't send local conversation logs to their own servers (kind of surprised if they don't). So if you think you're having a private local conversation in Chrome, I wouldn't be so sure.
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 reason. Proton, wine, and the ecosystem have evolved a lot in the last couple years to the point that I'm surprised when a game can not run on linux. Occasionally, you need to look at https://www.protondb.com/ to see if there is some startup option that needs to be added to get the game to run. If you're into single player games, Linux is generally a really solid choice.
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 sorry anytime I see someone using Windows.
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't SOA, but it certainly can do many daily tasks. Even if the AI isn't running locally, hosted GLM 5.2 is in the same ballpark as Opus 4.8, and you can choose your provider.
I would wager frontier LLM achievements will plateau, and open-weight models will catch up. Then OpenAI and Anthropic will have significantly less value prop. It is hard to say when that may happen. People regard open-weight models about 4-10 months behind frontier models, so its not like those using open-weight models need to wait long to catch up.
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. Verifying your identity with a government ID and linking that to AI is the dream of a surveillance state. Agents running on your computer, accessing your internet accounts, access to your personal conversations with AI, and accessible by the government is just wild.
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. I don't know what percentage of earbud users depend on the earbuds for hearing assistance, but I would wager it is a small percent of the total of people who use earbuds. I doubt that I'm the only person who considers people who leave their earbuds while talking rude.
> 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 disturb" sign feels neglecting. People who might initiate conversation don't know if they will even be heard if they try to talk, so why bother. I would rather be alone than in a room of people who are actively ignoring each other.
I dislike the NC pause because it often awkwardly unpauses when someone is replying to you. I just pop the earbuds out when I start talking. To me, speaking with earbuds in is rude, and I want to show the courtesy to the person I'm talking to that they have my attention.
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.
The title is a bit misleading. The title suggests that Facebook is intentionally paying for a political interference campaign targeting Alberta. It sounds more like it is the monetary incentive structure that indirectly encourages this type of behavior.
It has been years since I signed into Facebook. The content on Facebook is just vile and disgusting, IMO. I remember signing in 3 years ago, and signing out less than a minute later because I couldn't look at it. It wasn't content from people I knew. It was all the thirst traps, rage bait, weird ads, and other garbage. I don't think normal humans still use Facebook, do they? I can't imagine regularly frequenting a sewer like that.
I think devs forget how efficient and blazing fast server rendered pages can be, and ultimately what a great user experience non-SPA applications can provide. It seems like the dev community has locked in on SPAs for everything. There is so much complexity and other overhead associated with SPAs. At the end of the day a browser is rendering HTML + CSS, JS can handle some additional interactivity. Presently, we have some very complex SPAs that are handling large amounts of state, large dependencies (js libs), often optimized assets, etc. I remember people bemoaning Flash apps. I kind of feel like SPAs are kind of becoming the new Flash app.
> this will create pressure on developers to memory-optimize their software
Ideally, yes, realistically, no. It is rare that I hear FE devs considering how much memory their apps are using. I really wish RAM use would be a much greater concern, but when I look at programs I normally run, I can tell RAM is not a concern (imagine me giving a dramatic accusing look at Docker Desktop, next-server, ...). RAM use for web pages is often not given much consideration either.
The Neo's value prop is great for many people. I keep needing to remind myself that most computer users can get by fine with 8GB or RAM, and that the I'm not the target market for products like the Neo. I do get nervous with how future proof 8GB of RAM will be in terms of total usable lifespan for the Neo. Maybe the idea is shortened timeline to obsolescence means more sales. Not digging on the build quality, but just if 8GB or RAM will still be fine 5 years from now.
* Openrouter.ai for a hosted router
* https://github.com/diegosouzapw/OmniRoute for a local router