davedx

Ironically enough I just started an "Encylopedia of AI" project, and I've decided to generate all of the content with DeepSeek, with me being an editor/reviewer. I'm not really doing it for anyone except myself but I'm quite enjoying it so far. https://enkuklios.com/

It's not just about demand, it's about do you think valuation that prices them into serving 1000x that demand is worth paying for.

Yup.

I am currently throwing a lot of tokens exploring various theories of speeding up matrix multiplication on Metal.

If you won't be ambitious why do you assume somebody else will?

Tell that to legal firms and patent guys

It's not even an application it's a compression library

Honestly it's time to get rid of C as a language used where security is a factor. Enough already. Why the hell is a lib like brotli that processes public general internet traffic, and maintained by google, not in rust?

I think there's a difference between quantization and "gridification" - the latter removes swing too, everything goes to 4:4

Shocking take: you shouldn't really vibecode any part of a serious application!

Definitely use coding agents to write code, but carefully and thoughtfully review any important backend or heavily user-facing components of the software you're developing.

I wonder given its speed how competitive it would be with the current batch of decision models like Jev et al. Seems to me for decision models you really want maximum throughput.

Head of Macro here -- this is a great time series forecasting model, we are migrating to this in our pod!

I use DwarfStar DeekSeek V4 Flash on my Mac Studio with 128gb of RAM and it's pretty good - decent TPS and coding ability. Apparently the V4.1 model can also be run by streaming weights from the SSD - I've not tried this yet because it results in a lower TPS, but eventually this version of 4.1 should also be usable like this.

Such a tone deaf response.

Autodesk tried to trademark a phrase recently invented by an author. They would have had to defend it too. Would they have sued Gibson? Ahistorical my ass

Honestly I don't understand how it was accepted? Obviously Gibson invented it?

God, I only wish

It would be slower :P

Yes, and this version explicitly addresses that

I have started phasing into DeepSeek v4 (Dwarfstar quant from Antirez) as my daily driver and it's going very well. It can one shot iOS apps on my Mac Studio, it's a little slow and sometimes needs retries but can definitely do medium difficulty real work for me.

I didn't try the last flash model because I would need to use SSD streaming (I have 128 GB), but the benches for this release are so compelling (and it using less tokens to get tasks done) it might be worth the throughput hit. I will probably give it a try when I have some downtime.

It feels very freeing to be doing useful work on a local model on my own hardware with my own coding harness. The TCO argument is starting to look compelling too.

That was my first thought. Playing Civ games and reading books like Prisoners of Geography makes it pretty explicit

> First, no one should be asking an AI to rule on life or death (although I fear that is coming, regardless)

"Inside the US Central Command (CENTCOM), which oversaw the attack [on the Iranian school], officials said some personnel relied too much on the AI embedded in the Maven Smart System.

Developed by Palantir Technologies, the AI-enabled system uses more than 150 different data inputs to create a coherent set of information to inform commanders' decisions on complex military operations-- from targeting to command and control.

Since Trump returned to the White House, a Bloomberg probe found that the US Defence Department has heavily relied on Maven, making it a cornerstone of the US military apparatus. Several former senior US military officials now reportedly work for Maven's parent company, and their role is to keep the software functioning inside secure military environments."

Like all the stories about what Trump has done or said today, I find this article also impossible to tell if parody, satire, fiction, or 100% the author's very own direct lived experience. Because so much of it rings very true to my own experience especially with dentists.

Orthos seem much more chill and less terrifying than dentists where I am though, thank god for my kids.

Don't forget that all important whatsapp gateway

I spent a lot of time wandering around the border zones between countries on Street View. It's fascinating, the lengths we go to to keep the "other humans" out of "our country", and yet... Often even the most militarized borders are illusions, with huge swathes of wilderness making border walls or fences impossible.

My Mac Studio arrived yesterday and one of the first things I did was cancel my Claude subscription. Happy to be free of load bearing, price gouging, paternalistic "altruists" Anthropic.

Am keeping my Codex sub while I find the best local model, but my plan is to eventually stop with OpenAI too.

> you have no way of telling apart AI slop vs. good maintainable code

SonarQube

There's a continuum between "vibe coded by someone with no technical knowledge or inclination" and "hand written domain driven design development". You can absolutely use coding agents AND have maintainable code. But yes, the coding agents will not magically make everything maintainable if you don't tell them to.

"Code maintainability and good architecture don’t have good measurements that we can apply"

Who has no wisdom? There are dozens of ways to measure code maintainability. Cyclomatic complexity is just one.

Nothing stops you from wiring up something like SonarQube metrics to your agentic coding workflow.

I'm spreading my bets a lot lately, trying a bunch of stuff:

https://modelrigs.com - a website for running open weights models on various hardware setups, including benchmarks, how to run them, and a TCO economics analysis

https://riverfish.uk - this one's a bit different! It's basically a pSEO experiment in combining fish population surveys and river data across England

https://findclients.org - a B2B client prospecting database of UK companies across different cities and industries, with a market analysis of the different sizes and ages of companies

https://octoloops.com - AI growth app. I've downgraded my time on this as although I could get users to sign up, getting them to actually use the app hasn't worked out at all; new users won't reply to my nice founder emails either. Bit of a bummer because I invested quite a lot of effort but it just doesn't seem to be working. Ironically I still use it to dogfood on my own projects...

https://signalbump.com - another experiment: enrich people's projects in monday.com with data on new (capital) projects and expansions of their clients. It's a monday.com app, the marketplace submission process was very challenging as a solopreneur. We'll see, very early days.

Lots of RLHF - reinforcement learning from human feedback.

Also: "distillation" - seeding or running training sessions on the output of frontier models.

Very nice. Do you have a page comparing it to pi?

Think this is missing a URL?

There's a lot going on in energy, manufacturing, mining, and of course defence at the moment. For example, we're still making progress towards economically positive fusion power; Vehicle-to-Grid for EVs is starting to roll out as trials; hydrogen is still being seriously looked at in different parts of the energy supply chain; the first flight of a mid-sized electric aircraft happened the other day; a "quantum" based navigation system that doesn't need GPS did a couple of trials in the last weeks; laser based mining, new geothermal tech, and lots and lots of autonomous/drone defence tech is being developed.

Then there's also "newspace", an interesting area to watch!

Nope. The projects I'm on where we use it, we're carefully migrating to the newer models. Where we can we test with evals to try and get an understanding of how the models have changed.

It's not all roses -- I've seen some regressions -- but generally the 3.x Flash models are pretty great for our use cases.

The great thing about LLMs though is it's incredibly easy to diversify and have fallbacks. But of course that means additional costs, mostly centered around engineering efforts to test and integrate them.