It’s interesting the cognitive dissonance this brings out in me. As much as I think Musk carries a fairly reprehensible aura, I would still like to see SpaceX succeed. Space has a funny, transcendent way of inspiring us.
Very cool! FYI I was able to cause all sorts of hell by pinch zooming on mobile safari. This seemed to throw off the click and drag select box and on one occasion crashed the page entirely.
There’s also probably a lot less noise from casual internet conversations. I imagine a nontrivial amount of what LLMs know about certain technologies comes directly from forums like reddit where quality of response isn’t guaranteed.
I think there are some really interesting thought there, but I’d challenge some of this:
> Models are not free. Downloading them is free. Running them is not. This has manufacturing economics, not software economics; the idea that they are "free" is an economic category error as it relates to their actual use
I think a large part of manufacturing economics is illiquid overhead and the cost of expertise to set up and run your manufacturing line. Compute economics don’t have the same illiquidity nor do they require the same expertise or even specialized infra (current temporary chip shortage aside).
The implications of this are small players (e.g. your uncle running an inference server out of his garage) have comparably efficient marginal costs as big players. Compare this to actual manufacturing where small players have essentially no access to the manufacturing facilities of the big players.
Additionally, big players with a lot of compute who are not meaningfully in inference today (e.g. Amazon) have a fairly straightforward glide path to utilizing that compute to compete.
> This is because US labs are leading on cost efficacy of inference ($/task)
It’s possible, but I would need to see better data on this.
>A big part of training now is optimizing token efficiency. It's hard to distill token efficiency; that is perhaps why Chinese LLMs are so inefficient.
I think it’s fair to assume this is true, but also token efficiency is not a meaningful competitive moat. It’s not like these are secrets the Chinese will never figure out, it’s a fairly active research space and the outcomes are quantifiable.
That’s fair. What’s the Apple-recommended way to avoid this? I interact with a lot of software and tend not to hit it but I also don’t really interact with downloaded executables
I’ve been developing on MacOS for years and don’t think I’m familiar with what you’re talking about. I’m familiar with the warnings, but hardly ever see them.
There is a somewhat valid argument to be made that aggressively trying to hack these insecure government portals could lead to a real reprioritization towards competence.
What you’re describing is essentially the concept of civic duty and, at least in America, it’s unfortunately become deconstructed and politicized to the point where it’s impossible to exist as a meaningful cultural phenomenon.
It also relies on people putting effort towards often intangible, uncapitalized, and unextractable shared value. So perhaps it makes sense that this is being diminished over time, as the grip of capitalism squeezes tighter and more efficiently. With more economic stress placed on individuals, people have less available time and resources to devote to things other than staying afloat.
Between polarization/politicization of literally everything and the relentless corporate desire to deconstruct society in the name of quarterly growth, I’m not optimistic this is making a comeback. If we want to teach the children anything, it is that The System has failed and is in dire need of replacement.
I’ll defend batterygate. If you know anything about batteries (especially the tendencies of those in that era), the actions taken by Apple were reasonable, though they should have considered the light in which throttling would be taken. The claim against them was valid but I don’t think the actions were ever malicious.
> Here's a term for what I think is happening: the human reward function problem. In machine learning, a reward function tells an agent what good looks like. Writing code by hand was never easy, but it was full of small rewards. Solving a problem in your head. Understanding a gnarly bit of logic. Watching the code compile. The feeling of control. LLM-assisted programming has automated much of the work that generated those dopamine hits and replaced it with the cognitive load of review and supervision. The satisfying part shrank. The exhausting part grew. And there are no new rewards to fill the gap.
Say what you will about the Claudisms in this piece, this bit certainly rings true for me. With old school coding, there was always a reward at the end, the harder it was, the more satisfying it felt.
With agentic coding, I really doesn’t feel like that, at least not in the same way. It feels more like continually riding a wave of productivity, where small features or huge features have similar levels of interaction required. And that’s exciting in the beginning but quickly becomes very tiring.
I think my silly and unserious naive response would be linear scaling bundled binaries with number of platforms doesn’t seem to be that materially different in terms of total size. But I see your point, didn’t consider the architectures
It seems SQLite could be evolve to solve this by just bundling itself entirely in the data files? After all, the binary is less than 1MB, anywhere you’re putting a database surely has at least that much overhead, for most applications it’s less than a drop in the bucket.
I’d be interested to learn if there are any db implementations that take this approach, or reasons this wouldn’t work.