iamwil

what do you use to run Qwen3? and do you use it for coding? I'm still on an M1, and local models are often just too slow.

I would feel more comfortable if Ollama / Llama.cpp showed the checksums of the models I was downloading. I mean, I want to know I got the original.

I think I've been doing it wrong. I should set an alarm for when to go to sleep, rather than when to wake up.

I kinda want one of those PCs that can barely fit a GPU in it to run some OSS language model as an API. But it'd look like a friendly computer. Then I can carry it to coffee shops and still have access to it on the go.

Google Gemini can summarize Youtube videos with linkbait titles and thumbnails. Now I can find out "What everyone gets wrong about X" and I don't have to waste my time. All the metrics for creators are going to tank.

Tha launch of the probe and Picard's realization that he was the one it'll find. Star Trek: The Next Generation - S5E25 The Inner Light (1992)

The next intergalactic probe we send out should have an LLM on it, and not just a golden record. Heck, make it multimodal that can output VR with nerf splats.

m.youtube.com/watch?v=RQK...

YouTube video by Ammiraglio GL Star Trek: The Next Generation - S5E25 The Inner Light (1992)

The next intergalactic probe we send out should have an LLM on it, and not just a golden record. Heck, make it multimodal that can output VR with nerf splats.

m.youtube.com/watch?v=RQKp...

Should have trained on some Eastern European training data.

i.redd.it/72e03vux9aj1...

The moats are in the GPT-wrappers Despite the disdain for "GPT-wrappers," it's currently where the value is accruing in the supply chain of value to the end-user with LLM-driven apps. Unless something changes with the foundational mod...

I also write more about it here

interjectedfuture.com/the-moats-ar...

And getting whipped. And no access to air conditioning. And occasionally burned at the stake for being accused of being a witch.

The rich have an entire professional class that helps them minimize taxes legally. I suppose you can outlaw that, but I don’t get on what grounds or constitutionality that would be.

So I dunno. Capping at $10 mil sounds easy on paper, but just thinking through 2nd order effects, there are lots of down stream issues the article gives no thought or solution to—even if it was easy to implement, which I don’t think it is.

So you’d have to come up with another peaceful lever of national power, lest you want to lean on military power, but that throws us back to continental powers where might makes right, rather than the current maritime power where trade makes right.

That would likely slow the pace of the economy. The US’s diplomats has a lot of leverage over keeping the maritime trade and order due to the strength of its economy and the dominance as the world currency. A slower economy would probably reduce that leverage significantly.

You’d want assets that has a doesn’t fluctuate and also keeps pace with inflation. So maybe treasury bills and bonds?

That would also disincentivize public companies from making too much money in one quarter, lest they be seen as too volatile to attract investors.

What if the stock then goes down? If the surplus value >10 mil is taken away in units of shares, then you lose shares proportional to its increase in value. It’s kinda like leveraged losses in a way.

So that basically disincentives one to own assets that fluctuate at all.

Considering the limited case of equity: If you own public stock, and its value increases so your total assets > $10 mil, where does the surplus value go?

Does it go back to the company? Is it randomly issued to another citizen? Is it paid as tax to the gov’t?

There will be a shift in ecommerce site and platforms. MCP for major shopping destinations, like home depot, with a fallback website for human oversight. Boutiques should get on this, before Shopify wakes up to it. Amazon will likely sleep this one out.

Someone should fine-tune an LLM focused on restarting civilization, should things go awry.

The Kobayashi Maru Scenario, but for AI system evals.

People shouldn't be able to delete tweets. I want to be in on the joke too, retroactively.

How did you decide what math problems to try to tackle and prove with lean? Did you pick an area just cuz it was interesting? Or you picked up a textbook and started doing the problems in it using lean?

I hadn’t! Emojis changed how the young speak. But he should write one about how prompting will change language.

We're not there yet, but it's not hard to see that as prompting and prompt eng become more prevalent as a skill, we'll write for LLMs to read first, and humans second.

We change the medium, and the medium changes us. Medium is the message, I guess.

I suspect part of the cause was the product choice to truncate long posts into the first line or two, or an infinite feed, etc. so then everyone turned into a copywriter for headlines to get attention.

Worse, sounds like everyone read the same tips on how to write for Linkedin.

Every social media network is a culture onto itself, and Linkedin has this particular voice that I can't put my finger on. It's a voice written for the feed algorithm, where every first sentence is a hook. It's off-putting because I know you, and you don't talk like this!

Turns out Amazon Kindle was in the NFT business all along.

“Sorry for the long letter for I had not the time to make it shorter.”

"Hold my beer"

Intelligence might be on tap, but taste, is not. Gotta go get your own.