swalsh

Macrodata Refinery is mysterious but important work

Like I said last week. I think zucks original llama thesis is proving more correct by the week. The rent market on intelligence never materialized, if you have THE frontier model, you can charge rent (for a few weeks, maybe month). But after a certain capability level it's price that drives the decision.

Zuck is doing the thing he's really good at, and building a moat around the network (Muse is VERY sticky), and now that we're all going to standardize on Meta's model, we're going to optimize the heck out of the infra to run the model making it cheaper for him to scale, thus bringing in more participants to his network, and thus making his network more valuable.

Meta is going to win consumer AI.

I've used it in a bunch of places, but my favorite use. I have an app (https://plonkit.app/) i'm the only user... because i built it for myself. In the app I have a shopping list feature. When you add an item it auto assigns it to the right shopping list. It's a very simple feature, but with Jev, it's quick, and it stayed simple.

I disagree actually, I think zucks original llama thesis is proving more correct by the week. The rent market on intelligence never materialized, and will not materialize.

In software, we've all become accustomed to the idea that the way you make money is by building little blocks of intellectual capital, then charging rent on it. That made sense when developing intellectual capital was a rare and valuable skill/asset and the thing you build was a rare and valuable asset. But now both are not rare. You can still charge rent, but the market is highly competitive, so prices match that.

Durable rents will end up going to those who owns the scarce assets, zuck is building those as we speak.

Dropping open weights will build an ecosystem that makes his valuable assets more valuable.

I think it's a joke at this point, but also the visual benchmark is a remarkably dense method for demonstrating how good a model is.

I've been using GPT-6 Luna in some capacity for nearly all my agent workflows. It's just a really good model, and the pricing is cheap. If Haiku 5.5 is better, and the same price (under 100k context... which is a big caveat) i'd probably swap it.

Top of the page in 17 minutes? Now I know what y'all do while your agents are working.

It's a shame OpenAI will never publish the trace that led to the insight.

They're trying to pull digitally what they already pulled physically. The reason we can't manufacture a grill brush for a reasonable price is the result of years of Americans choosing the cheapest price. We gave up our manufacture base. They want us to give up our labs.

I mean, when I upgraded my pipeline from terra to sol saying "it's the same price basically!" I was excited. Probably would not have felt as excited if it was just a version bump.

Not sure that's why they did it. But that was my experience.

If you ever had claude code fan out a task to sub agents it will often use sonnet or haiku for those tasks. Web scraping is an example.

After the cursor aquisition it's become a quite capable coding model. If you take cost into account, it's close to the top. OpenAI is maybe still #1, but I'd put Grok at #2 (again, including cost as a factor).

Codex has become my goto tooling. I used to be a Claude Max subscriber, but I was becoming disappointed with the quality of the output from Opus 5. Fable chewed through my usage too quickly to be practical. Moving to a Pro account w/ Codex was a big improvement. Sol had great output, and the usage was more than sufficient for most of my needs. However astra does tend to chew up usage, so when i've done to much of that, and it's became an issue Grok Build has beocme my second go to account. The output especially after the cursor purhcase has become quite good, and the usage has always been very generous.

So what's the felony benchmark at now?

I was thinking about canceling my claude max sub after a few bad experiences. Kept hitting my usage limit, the quality of code seemed worse than Sol. This just made my decision. I'm moving to Codex Pro.

Meetings, research, look over work done etc.

It's not insulting. Not every problem is a frontier problem.

Now that it's a solved benchmark, can we get the animated version?

I've recently been running these agent sessions on more and more long running tasks because these latest models can do a REALLY good job on big chunks of work, and i've been watching them way less. It's starting to occur to me the importance of alignment is a today problem, it's not a tomorrow problem.

In the past I watched and saw everything the model did, not a lot got past me. Today it does A TON of work while i'm busy on other tasks. It also has extensive access to my computer, other computers on my network, my internet. It's really helpful when you give it a lot of resources, but right now I have very autonomous, very smart agent running around more or less unattended with a lot of resources.

For as long as you're on it, but eventually you run out of body fat.

Calling it now, the downside is the effects are most apparent on higher dosages, but you cant take higher dosages forever. Eventually you run out of body fat, have go reduce the dose, and the benefits dither.

You can probably get a prescription for a low dose, but you'll be self pay.

Yes they run in parallel via LMStudio (250k context)

yeah, i consider that slow.

WOW, my first try running on my 2 3090's, it was a bit slow... but it FEELS like opus 4.5, i gave it an image and a broad overview of what I wanted it to build, and it built the whole thing from beginning to end.

I suspect mythos demonstrated a fully autonomous offensive hack in a similar way Open AI's models performed, and the government is reacting to it the same way we reacted to blackhat.

The threat is real.

Yeah, so I'm going to confess. Ive built production AI systems that have very specific jobs, deployed them, and moved on. The cost is not noticeable, i had the system dialed in. Not worth the effort to reevaluate a newer model to see if its better in some way. Current model works, and I have other projects that are more important.

Lol perfect analogy. I'm still paying for claude because the quality is unmatched.

I think profitability is a matter of accounting. Inference is where money is made, but training is where money is spent. We keep getting new models every few months, but frankly the old models are still quite usable. I suspect labs will soon start specializing in expert models per use case so they can increase the lifespan of individual models, and change the profitability per model.

Its tough to go from max account at home and pay per usage enterprise account at work with heavy usage limits... but the limits are there because pricing is insane. Feel like I'm in the $5 Uber rides phase at home.