dehrmann

It's more like Buck and Tupperware.

How do models approach SVG generation? In one version, I imagine them actually trying to reason about them as an LLM. In another, I imagine something closer to a GAN.

A nit, but fusefs is a well-known thing, so you'll face some naming confusion.

I enrolled in 2002. The Nasdaq bottomed in September 2002, there was talk everywhere about outsourcing, but I went into it because that's where my passion was. I graduated into the 2006 job market which wasn't booming, but recovered enough. I remember chatting with younger coworkers about how we wish we saw the dot-com bubble. Then I saw the 2010-now tech industry.

Turns out I bought what's probably the largest investment in my lifetime at the bottom.

The industry will change in the next four years. What I didn't realize in 2002 was widespread broadband internet would arrive and change everything. Revision to mean is a thing, and I wouldn't expect the 2010's boom where tech moved to take over every corner of our lives, but it will still be there, and the systems take proper care and feeding. Remote work is equalizing global tech wages, but the talent's gotten better and the pie's gotten bigger.

If you don't have a passion for tech, pick a different major because you'll spend the next few years questioning your choice. If you stay with it, look for other areas in your life you enjoy so you can plan out a minor or double major. If you plan it right, you'll be able to reassess in three years and possibly pivot to math, actuarial science, electrical engineering, or mechanical engineering with only an extra year.

It's like a basic Kindle. Or Ryanair.

Between this, the proprietary nut patent, and the heated seat subscription, I'm probably not going to look at BMW for my next car. They've been going back on things, but all of this adds up to they don't get it.

It's differently interesting. It's interesting that they didn't get an important detail right with a partner, so more on Anthropic's attention to important details rather than the power of their models.

He changed his tone after Meta no longer lead neither open-weight nor SOTA models.

Sibling has a good comment about silicon. Another reason you're printing something like 100B transistors. So that's at least 100B raster steps. It would just take too long.

This is picking the wrong tool for the job twice over. Half of it is seeing if text-to-image mechanics exist in general purpose LLMs, the other half is seeing if they can reason their way about generating the 2d array. Imagine asking this of yourself and doing it through just a train of thought. Yes, I could probably draw a basic smiley face in the color of your choice in a 16x16 without antialiasing in 15 minutes.

People are actually angry about inflation and a sense of a stagnant labor market. Data centers and AI play into both those fears.

> dopamine hits from irregular and uncertain outcomes

Like the LLM getting the solution right?

> Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare

AI subscription pricing is so goofy. You get some amount of usage that varies by models, is measured by opaque token usage, driven by how many tokens the (usually) vendor-provided interface (or model itself) wants to use. Then your usage is limited by time opaque time windows.

The headache I recently had was it somehow started interpreting mouse clicks in the terminal to mean I clicked an option when I was really just trying to get/confirm window focus.

What you didn't elaborate on is that it's a poor investment thesis, so while the association is Nasdaq == tech, it's not entirely true, and it missing things if what you really want is tech. It also penalizes small floats less than S&P 500, enabling these shenanigans.

Compilers only really use them for things like break and the end if blocks. There isn't much to gain by being able to execute anything arbitrarily, and then go...where? There might be places here and there within a function where compilers already take advantage of it, but I doubt it's useful in the way you think.

> They don't have Claude write assembly because there is no training corpus on people making CRUD apps in assembly.

I suspect that despite its translation abilities, this is true, but I'd like to see it do things in languages that are more or less appropriate for tasks to see how much the training corpus matters vs. its ability to translate. Assembly is a bit of an extreme example because you're either writing it as close to C as possible (C is essentially portable assembly) or you're writing complex, unreviewable code that happens to work. And who know if it's been trained on register allocation, or resorts to doing everything on the stack because it works.

Isn't Proxmox a management layover over Debian and KVM? I doubt the kernel is even smaller.

There's a lot of Gell Mann amnesia going on with LLMs.

We can't rewind; we've gone too far.

Milk, not cream?

> a perpetual bond with declining coupon payments

Most things with royalties (oil fields, songs) work like this.

Consolidating stagnant or dying SaaS offerings makes sense, but it'd be nice if there were a version of this that's a better steward of the companies.

This makes a case for engineering margins, maybe even running the numbers assuming a worse grade of steel or bolts than specified. Also worth remembering this building wasn't special. If this was a design or construction flaw that surfaced with added load, a lot of other buildings from that era probably have a similar issue.

Dismissing people like this is part of what fuels the antivax movement. Vaccines are generally effective, but they're not perfect and have side effects, and failing to acknowledge that when someone is asking in in good faith polarizes people and makes it look like someone's trying to hide something.

There are multiple FDA-approved lab-grown meats on the market. You can literally go to a handful of restaurants and order lab-grown meat today. The production process is just expensive and it's getting scaled out.

I used to work for a Pricerunner competitor. Not Nextag, but they were the most well-known in the US. They're called "comparison shopping engines," and most were names people would barely remember.

These companies got steamrolled by Google because their service wasn't very good. It's all an affiliate marketing play, where they get traffic for keywords through Google organic search (these sides made heavy use of SEO) and Adwords. When you land on the site, everything is an ad, either a merchant product link the merchant pays per click for or less commonly, an affiliate link. Result ranking is tuned between revenue and relevancy.

The problems are it's not really a comparison as much as search results, you're clicking in from a Google SERP to another list of search results, and the results used factors other than relevancy, and the side was designed to encourage click-outs.

Anthropic must think that their moat isn't very large if they're this worried about distillation.

> but solid state storage similar to Nintendo Switch cards makes more sense to me

These likely degrade in 5-10 years, and have you seen the price of NAND lately? AAA gaming is going to get to be out-of-reach because of storage costs.

I haven't used Kalshi, but I expected it's actually easy because it's technically a futures exchange, so it'll be regulated more like a stock brokerage than off-shore gambling.

It's not even the Nasdaq Composite, It's the Nasdaq 100. And yes, the Nasdaq 100 is a shitty index because Nasdaq vs. NYSE is a bit of an arbitrary, tradition-based way to get tech exposure. It's missing NYSE tech companies (Oracle, Dell, Spotify, Uber) but gets some non-tech like Costco and Pepsi. It'd be like only dating iPhone users. It's sort of a rough proxy for something, but you're better off choosing what you actually care about.