You would think they’d at least experimint with a few lines before they cast admonishmint at we poor souls who are so pArched for a good line.

That’s because these are GUI users, the real TOS is written in an arcane format only viewable by a cat or a bat.

iPhone notification summaries were made with GPT3.5 I believe (maybe even the -turbo version).

It doesn’t use reasoning and so when using very short outputs it can produce wild variations since there are not a lot of previous tokens in order to direct the LLM into the appropriate direction in kv-space and so you’re more at the whims of temperature setting (randomly selecting the next token from a SOFTMAX’d list which was output from the LLM).

You can take those same messages and plug them into a good model and get much higher quality results. But good models are expensive and Apple is, for some reason, going for the budget option.

Anyone learning a new language massively benefits from being able to speak with native speakers.

That being said, LLMs are better at languages and translation tasks than any pretty much anything else. If you need vocabulary help or have difficulty with grammar they’re incredibly helpful (vs Googling and hoping someone had the same issue and posted about it on Reddit).

I mean, if you can afford a native speaker tutor that is the superior choice. But, for the average person, an LLM is a massive improvement over trying to learn via YouTube or apps.

Thanks a ton, saves me having to navigate the slopped up search results (‘AI’ as a search term is SEOd to death and back a few times)

I dunno what card you have now, but hybrid CPU+GPU inference is the trend days.

That system has the 3080 12GB and 64GB RAM but I have another 2 slots so I could go up to 128GB. I don’t doubt that there’s a GLM quant model that’ll work.

Is ollama for hosting the models and LM Studio for chatbot work still the way to go? Doesn’t seem like there’s much to improve in that area once there’s software that does the thing.

Yeah, you do want more contextual intelligence than an 8B for this.

Oh yeah, I’m sure. I may peek at it this weekend. I’m trying to decide if Santa is going to bring me a new graphics card, so I need to see what the price:performance curve looks like.

Massive understatement!

I think I stopped actively using image generation a little bit after LoRAs and IP Adapters were invented. I was trying to edit a video (random meme gif) to change the people in the meme to have the faces of my family, but it was very hard to have consistency between frames. Since there is generated video, it seems like someone solved this problem.

Thanks for the recommendation, I’ll look into GLM Air, I haven’t looked into the current state of the art for self-hosting in a while.

I just use this model to translate natural language into JSON commands for my home automation system. I probably don’t need a reasoning model, but it doesn’t need to be super quick. A typical query uses very few tokens (like 3-4 keys in JSON).

The next project will be some kind of agent. A ‘go and Google this and summarize the results’ agent at first. I haven’t messed around much with MCP Servers or Agents (other than for coding). The image models I’m using are probably pretty dated too, they’re all variants of SDXL and I stopped messing with ComfyUI before video generation was possible locally, so I gotta grab another few hundred GB of models.

It’s a lot to keep up with.😮‍💨

They’re overestimating the costs. 4x H100 and 512GB DDR4 will run the full DeepSeek-R1 model, that’s about $100k of GPU and $7k of RAM. It’s not something you’re going to have in your homelab (for a few years at least) but it’s well within the budget of a hobbyist group or moderately sized local business.

Since it’s an open weights model, people have created quantized versions of the model. The resulting models can have much less parameters and that makes their RAM requirements a lot lower.

You can run quantized versions of DeepSeek-R1 locally. I’m running deepseek-r1-0528-qwen3-8b on a machine with an NVIDIA 3080 12GB and 64GB RAM. Unless you pay for an AI service and are using their flagship models, it’s pretty indistinguishable from the full model.

If you’re coding or doing other tasks that push AI it’ll stumble more often, but for a ‘ChatGPT’ style interaction you couldn’t tell the difference between it and ChatGPT.

Still deflecting. The one-liner gaslight doesn’t erase the DARVO playbook you just ran. Anyone scrolling can see it; my work here is done.

Textbook DARVO sequencing.

(https://en.wikipedia.org/wiki/DARVO)

On top of that, hypersensitivity, projection, victim-role adoption, blame pre-loading.

While these are effective rhetorical tools for manipulating people, this isn’t how reasonable people engage in conversation.

If you’re not a bot then you’re a narcissist or have picked up narcissistic behaviors from reading social media arguments and should, as the kids say, touch grass.

I’m not sure about Cloudflare but it might be as well.

Cloudflare was a chain of unfortunate events.

The TLDR is, a permission change caused a poorly written SQL query (without a properly filtering ‘where’ clause) to return a lot more data than normal. That data is used by an automated script to generate configuration files for the proxy services, because of the large return the configuration files were larger than normal (roughly 2x the size).

The service that uses these configuration files has pre-allocated memory to hold the configuration files and the larger config file exceeded that size. This case, of having a file too large for the memory space, was improperly handled (ironically but not literally ironically, it was written in Rust) resulting in a thread panic which terminated the service and resulted in the 5xx errors.

So, it’s more similar to the Crowdstrike crash (bad config file and poor error handling in a critical component).

I thought you were setting up the “He’s famously from Ypsilanti Michigan, everyone knows this!” line, lol

Yes, exactly. They use a lot of bad faith arguments, logical fallacies, etc. They’re all just literary devices to amp up the messaging and they all follow roughly the same kind of bad faith argument templates.

Of course, these could be people. The problem is that they seem to simultaneously be completely ignorant of how reality works while also being an encyclopedia of pop culture political references and grievances. If it’s a person then it fits the ‘chronically online, touch grass please’ stereotype, but there are so many others that argue exactly how it does on the exact same topics across all of social media so I just assume it is a bot or a bot-affected person.

We know there are information operations that exist to shape public perception that are happening on all social media. This is what it looks like, very hard to tell apart from real people (made even harder by real people mimicking them).

BabyTron meme unless my Internet sense is failing me.

Pretty obvious the person you’re arguing with isn’t a trump voter

Is that really obvious though? It reads like a troll trying to create pointless division among progressives.

I always read these people as either political bots or children (or adults, similarly intelligent) spreading a meme created by a political bot.

That is 100% the cat’s chair, and also a warning as to what will happen if it is ever disposed of.

If they were the learnin’ type, they wouldn’t be here

It gets weirder the longer you look at it.

Sure, let’s just say the guy was overzealous with the (silicone caulk? lol) adhesive compound. Maybe the white cord is DC power, replacing the battery… but the red wire that’s right beside the ‘power’ wire is a USB cable plugged into the phone’s USB port.

What is plugged into the other end? It’s Zalgo isn̴̝̂’̶̯̾ṭ̷̆ ̶̫̈i̷̹̚t̴̩̉?̶͊͜