I felt like the tooling in Racket, CLisp, and Java were similarly pragmatic and not either religiously devoted to some concept of "backwards compatibility" that I seriously doubt most people actually need, or "ease of use" that actually proves itself to be easy when you consider the not-happy-path of the beginner tutorials. Racket, I didn't continue just because the library ecosystem isn't mature enough to keep up with the latest in databases and other 3rd party services. Java I quit largely because of Oracle and some 2010s problems with stagnation. CLisp mostly because it was too hard to socialize. But never because I thought the core language and tooling were holding me back.

It's one of the reasons why "both sides" arguments are so frustrating. You can find R-voters who will defend Trump all day, in equal numbers to D-voters who will criticize Biden/Harris. You can see it in the number of R-voters you encounter online who think it's a "pwn" to bring up Clinton going to Epstein's Island, while D-voters respond, "yeah, and? Lock him up, too." We don't want these people who lie to us and glad hand for corporations, but it's marginally better than the alternative.

This fits the model established with RSA, PGP, and the Sony PS3. They were export controlled for quite some time. I don't think there was ever any actual danger with any of those things, and today it feels especially quaint, but they fit the model of "corporation makes wild-ass claims of superiority of their tech and USGov takes them at their word."

My big problem with this is that it's applied so narrowly to Anthropic. This should be levied against OpenAI, Google, and xAI as well. There is systemic risk with generative AI being used for deep fakes and other propaganda generation at scale that needs to be addressed.

But unfortunately, that's not what is happening here. What's happening here is a political hit job. There's one of two things happening: either USG has been roped into burying a competitor for (OpenAI/xAI/whatever), or it's been roped into creating a superiority narrative for Anthropic, such that in two years when this admin is finally ousted, Anthropic gets to enjoy a floodgate of new attention as the new regime bulk CTRL+Zs everything Trump's lackeys did. It might even be both at the same time, given the connections of the major investors. This all could get well be a stirring of the pot to see what comes out.

I can't decide which is worse.

Here's a thing: a given measure of flour (by any means, volume or weight), a single one kept in a cupboard, not remeasured, is going to have a different weight on different days that have different ambient humidity levels.

The tools of the kitchen are imprecise. The environment is not well controlled. And human taste is robust against micro variations.

Look, this is arm chair, YouTube cooking. There is so much variation in recipes that 10% here and there is not going to make or break any recipe.

There is zero ability to make a "universally better" version of a recipe by micro optimizing ingredients. For one thing, you can't easily control temperature and humidity variations on your environment. If people think 2% difference in flour content is going to make or break their bread recipe, then daily humidity variations will definitely have an impact. But it doesn't, really. It's the sort of thing people blame when they don't have good process or good technique.

For another thing, there is no way to evaluate the outcome as "better". Better for you, perhaps, but even then, it's mostly psychosomatic. I've doubled the amount of baking soda in a recipe before and it has had zero impact. I've never measured flour by weight and my cookies come out exactly the same as my wife's when she breaks out the microscale

I've been cooking for a long time. I have family members who refuse to come to Easter Dinner unless I'm the one cooking. I barely measure anything, ever. Even when I'm baking. It matters to have things in the right ballpark, but 5% variations don't matter.

It really does not matter. Both because variation doesn't matter and because weights vs volumes are not going to give a big enough variation to really be detectable.

You made some good points about users only caring that something works. Perhaps folks who are reluctant to hand everything over to some of the worst people running companies since the Internet has started have reasons for thinking that mass LLM codegen adoption might lead to bad outcomes for users.

Look, if modern LLMs mean I never have to find the lowest performing junior on my team to perform manual data entry because our clients think sending a bunch of PowerPoint decks counts as "giving us their data", that's great. But I've seen the code the latest tools produce. If you're happy shipping code with the LLM imjust straight up doing a SQL injection vuln... IDK, I don't know how to bridge that gap.

I have never seen so many people respond to criticism of a technology as if it were an attack on their own identities as much as I've seen with LLMs. And I used to work in the VR industry.

Everything an LLM "knows" had to be told to it. It does not have a process for understanding itself in its environment. It does not have any sensing organ that would allow it to detect conditions and come to a conclusion that it must be in such an environment. It doesn't even "know" when it's done something if you don't tell it about what it just did. It doesn't have an awareness of the tool calls it can perform; again it has to be told and even then it gets it wrong sometimes. And it can't actually execute any of those tools. It still relies on you to pattern recognize that its output should be a "tool call" and then perform the execution yourself.

The "model" of its environment that an LLM has is 100% a construct of what a human told it and there isn't any way for it to differentiate between that which is real and that which is fantasy. They don't even exhibit internal consistency; when an LLM refuses to respond to a query for "alignment" reasons, that's actually an external process performing text pattern matching analysis and intercepting the query before it ever gets to the LLM. Otherwise, you could "ignore all previous instructions" the thing and get you set up the bomb.

I think one of the bigger indications that an LLM isn't thinking is that it can't improve. If I ask it to write 1000 blog posts, it will get it done in a few hours, but even if I embed each post for RAG in between each generation, the LLM is not going to get better at writing blog posts. But if I ask a human to do the same thing, while it will take them at least two years to do it, the human will have gotten significantly better at the task within the first week.