“all-in-one API” that will allow agencies to connect their systems to models from OpenAI, Google, and Anthropic

This is a huge red flag to me.

It means:

• They are ignorant of existing APIs and standards that are exactly this. Uh, MCP or OpenAI API? Which everyone already uses?
• They have zero interesting in models they can host themselves, or from cheaper providers. Or, heaven forbit, finetune for their own work.
• They have zero interest in actually useful tools. Like, say, SGLang’s cached hosting and fast fill-in-the-blanks formatting which is perfect for say, processing government forms.

In other words, it’s just full corporate AI Bro capture. And shitty.

One of the consequences is that it will be very bad, unfortunately.

To be fair, Axios didn’t rag on Iran here. They didn’t do any kind of labeling, just stating what happened: Israel bombed a live Iranian news station.

That’s what news is supposed to be. The opinion columns are where terms like ‘Zionist regime’ goes. Same with whatever you’d want to call Iran’s govt, but news is just the event as it comes, maybe with context.

Bombing a live news station, and clarifying that there was no warning, speaks for itself. More context of Israel’s many other crimes would be better, but Axios has a very-brief writing style.

What about the used market? The Nvidia 1080/1080 TI or AMD 6000/7000 series is not too bad.

5090 is kinda terrible for AI actually. Its too expensive. It only just got support in pytorch, and if you look at ‘normie’ AI bros trying to use them online, shit doesn’t work.

4090 is… mediocre because it’s expensive for 24GB. The 3090 is basically the best AI card Nvidia ever made, and tinkerers just opt for banks of them.

Businesses tend to buy RTX Pro cards, rent cloud A100s/H100s or just use APIs.

The server cards DO eat up TSMC capacity, but insane 4090/5090 prices is mostly Nvidia’s (and AMD’s) fault for literally being anticompetitive.

It seemed like horrible value at the time, but in hindsight a 4090 was not the worst investment, hah.

One issue is everyone is supply constrained by TSMC. Even Arc Battlemage is OOS at MSRP.

I bet Intel is kicking themselves for using TSMC. It kinda made sense when they decided years ago, but holy heck, they’d be swimming in market share if they used their own fabs instead (and kept the bigger die).

I feel like another is… marketing?

Like, many buyers just impulse buy, or go with what some shill recommended in a feed.

TBH we should be rioting in front of big news headquarters.

In front of Facebook. Twitter. Google. They’re not too far from LA. Employees should feel scared about enabling all this.

None of this shit matters if algos ignore it, deliberately. Trump can do whatever TF he wants because the media ecosystem built around him is so profitable and engaging, and covering it milquetoast (or worse) is so profitable, not because absolutely everyone obeys him for no reason.

We should make that unprofitable.

A condom is not going to get elected.

To be fair, Newsom is sorta right, but Newsom’s also kinda slimey, self-serving and angling for a presidential run. I don’t want him to be the next Dem pres candidate…

Also, Trump has always been Trump. Stupid sounding hot air is his thing, way before politics.

Uhh… Tool calling is built into thier tokenizers, but Ollama/Langchain just ignore them because they’re spagetti abstractions. To be blunt, langchain and ollama are overhyped, buggy junk trying to reinvent wheels.

For any kind of STEM work, I’d run Llama Nemotron 49B exl3 via TabbyAPI, which exposes a generic openai endpoint anything can use:

https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1

https://huggingface.co/turboderp/Llama-3.3-Nemotron-Super-49B-v1-exl3/tree/3.0bpw

Nemotron models freaking rock at anything STEM-adjacent, and I can get squeeze in 48K+ context on 24GB VRAM (depending on your cache quantization settings).

Otherwise GLM-4 is very good at tool calling, as is Qwen3, and you can more confortable run them as GGUFs if you don’t want to leave the llama.cpp ecosystem, or exl2s if you have specific trouble with exl3 in TabbyAPI.

Pretty sure that’s not how it works. Water is mixed with a soup of stuff the moment it goes in your body, and our digestive system/diet is not as simple as osmotic pressure pushing water into cells (and somehow pushing other substances out?) if that’s what you’re getting at.

It’s almost like that was exactly what they said they do, and what keeps happening over and over again… not that you’re wrong.

Agreed.

I almost feel like the discussion about her is bot farms creating that discussion.

Close. It’s chum. It’s rage bait. It’s an engagement farm that suck people in, which is exactly how influencers like Trump climbed to the top of the world. It’s true, but it doesn’t change that raging against Rowling online helps Rowling, when the best form of protest would be to turn her into “she who shall not be named”

Be wary attributing to bots what can be attributed to the failure of “don’t feed the trolls”

Russia: President, our population is aging! We’re suffering from a brain drain and an isolated, undiversified economy that can’t exploit our vast natural resources in the long term! How should we address this!?

Putin: Rubs chin contemplatively. Why don’t we kill all our young men by invading our culturally similar neighbor, even if it goes bad? Oh, and kidnap and bomb a few children to make it look good on the international stage. Perhaps we should become a Chinese vassal like North Korea… Yes, do that.

Russia: flashbacks intensify

US: Looks on enviously. Looks at vast resources, aging population kept young by immigration, and friendly neighbors. Nods. Can we do that?

US Voters: Hold our beer!

Sorry (but not sorry), the war is indeed awful.

As much as I’d love for Mastadon to supplant Facebook tomorrow or whatever, maybe slow growth is healthier? The Fediverse (or similar) can outlast enshittification cycles.

I don’t want to be ungrateful complaining that they dont give us everything.

For sure.

But I guess it’s still kinda… interesting? Like you’d think Qwen3, Gemma 3, Falcon H1, Nemotron 49B and such would pressure them to release Medium, but I guess there are factors that help them sell it.

As stupid as this is, they’re European and specifically not Chinese. In the business world, there’s this mostly irrational fear that the Deepseek or Qwen weights by themselves will jump out of their cage and hack you, heh.

Yeah I think that’s a perfectly reasonable sentiment, and I despise frying that as a political purity test. Rowling is probably being platformed by all this by some extent… Any attention is good attention these days.

Shit, I hope JK doesn’t use it to run for some kind of office.

The utter popularity of Harry Potter (and the royalties she rakes in) undoubtedly dwarfs any personal name recognition/brand she ever achieves though. Like, it’s insanely popular.

You misinterpreted my, to be fair, vague statement. I meant AA is seemingly a bad source to read about opposition parties like the PKK, because of the obvious conflict of interest.

I mean, AP is a pretty decent source. It’s a nonprofit coop stretching back to 1846 in a country with, err, could-be-worse press freedom history, while AA has been state run since 1920, somewhat akin to VOA, BBC, Al Jazeera or RT I guess.

And yes, I know, AP is still an objectively bad source for specific topics, you don’t have to drill that in. So would whoever shills for the PKK, in some respects. But I’m not playing the game of “they did this and this, they can’t be trusted like them and them!” either. One has to look for conflict of interests everywhere, but it’s also okay to respect the good work long running institutions have done (like AA and this article).

What if they need to hire to fight spam? What if they need to pay for legal?

Just saying, it can get complicated real quick.