gipp

I don't think the GP had anything to do with the language change itself. More that it marks a clear point of Anthropic taking ownership and governance of the project in a less open direction.

https://fred.stlouisfed.org/series/BOGZ1FL594090005Q

Pension funds hold $30 trillion, my guy. Even restricting to only defined -benefit funds it's 7T.

As I and several other people pointed out last time this was posted, "A is B," in natural language, does not imply "B is A." "Is" can denote any of many different shades of relationship weaker than logical identity.

This has very quickly become an uninteresting, and often even unconvincing critique. Especially on this site where it is levelled at essentially every blog post submitted.

Maybe true, maybe not. If it actually says something, which this one does, I just don't care. And I'm hardly an AI cheerleader

As has been mentioned the goal was a port so they "could" eventually rewrite most of it to be idiomatic rust.

They may have said that, but quite clearly the value they actually get out of it is getting the headline "AI reimplements complex, broadly used software in 2 weeks, but makes it way better because it's rust now" in front of a million people's eyes, only 1% of whom will ever find out it was mostly fluff

When we're talking lab revenue, we're taking what companies are spending on AI.

The question is not whether companies are investing in AI, it's whether they're getting anything in return. Or, whether execs are just as anxious and confused about the story being sold as everyone else, taking the ludicrous amount of capital being put behind it as evidence that there's a "there" there, and hopping on the train out of pure FOMO and hedging, whether they're actually getting anything out of it at all.

For everyone but the individual account holders (who have ~no voice in this market) the opacity of the trusts is a selling point, isn't it?

I'm not sure what's different about my setup (just a Vaultwarden deployment hosted behind Tailscale, connected using the official Chrome extension and Android app), but I've never once encountered the long unlock delays due to sync attempts. It's always unlocked instantly. And the app is frequently unable to connect since I'm not always on the tablet.

Buddy... The whole point of the post is that he wants his students to question whether "succeeding in this market" is really the right choice.

The set of tasks for which "correctness" is formally verifiable (in a way that doesn't put Goodharts Law in hyperdrive) is vanishingly small.

The problem with these attempts always seems to be that you can see in dimensions 1-3, but never in dimension 4, so any movement or exploration along that axis is always just blind fumbling. The extra dimension is not equivalent to the others

The only answer would seem to be an extra axis of rotation, but (a) doesnt work well with existing input methods, and (b) would be even more of a brain-breaker

He was sarcastically paraphrasing earlier deflections from the administration

"Supply chain risk" is a specific designation that forbids companies that work with the DOD from working with that company. It would not be applied in your scenario.

Something I've certainly witnessed on this site in particular more than once

Maybe read that quote again. The figure is 1000 per day

It would be easier to judge this if the jokes weren't 90% about AI and silicon valley, understandable only to people who subscribe to astralcodexten

If choosing the "wrong" model, or not wording your prompt in just the right way, is sufficient to not just degrade your output but make it actively misleading and worse than useless, then what does that say about the narrative that all this sort of work is about to be replaced?

I see a lot of the same (well thought out) pushback on here whenever these kinds of blind hype articles pop up.

But my biggest objection to this "engineering is over" take is one that I don't see much. Maybe this is just my Big Tech glasses, but I feel like for a large, mature product, if you break down the time and effort required to bring a change to production, the actual writing of code is like... ten, maybe twenty percent of it?

Sure, you can bring "agents" to bear on other parts of the process to some degree or another. But their value to the design and specification process, or to live experiment, analysis, and iteration, is just dramatically less than in the coding process (which is already overstated). And that's without even getting into communication and coordination across the company, which is typically the real limiting factor, and in which heavy LLM usage almost exclusively makes things worse.

Takes like this seem to just have a completely different understanding of what "software development" even means than I do, and I'm not sure how to reconcile it.

To be clear, I think these tools absolutely have a place, and I use them where appropriate and often get value out of them. They're part of the field for good, no question. But this take that it's a replacement for engineering, rather than an engineering power tool, consistently feels like it's coming from a perspective that has never worked on supporting a real product with real users.

Well, let's start by confronting and acknowledging the very strong case that we -- "we" here being the tech world in general, and the audience of this site -- bear a heavy burden of responsibility for it.

It could be argued that it was all inevitable given the development of the Internet: development of social media, the movement online of commerce and other activities that used to heavily involve "incidental" socialization, etc. And maybe it was. But "we" are still the ones who built it. So are "we" really the right ones to solve it, through the same old silicon valley playbook?

The usual thought process of trying to push local "community groups," hobby-based organizations etc is not bad, but I think it misses an important piece of the puzzle, which is that we've started a kind of death spiral, a positive feedback loop suppressing IRL interaction. People started to move online because it was easier, and more immediate than "IRL." But as more people, and a greater fraction of our social interaction moves online, "IRL" in turn becomes even more featureless. There are fewer community groups, fewer friends at the bar or the movies, fewer people open to spontaneous interaction. This, then, drives even more of culture online.

What use is trying to get "back out into the real world," when everyone else has left it too, while you were gone?

People were absolutely giving attitude towards people in Teslas in general, and Cybertrucks in particular, around the peak of all the DOGE nonsense.

Still are, for Cybertrucks

And NYC. Just saw one this morning

that the author offers no charitable reasons for why the experiment took place.

Right, and neither did the GP. They both offered the exact same two reasons, the GP just apparently doesn't find them as repugnant as the author

This feels forced, there are obvious and good reasons for running that experiment. Namely, learning how it fails and to generate some potentially viral content for investor relationship. The second one seems like an extremely good business move. It is also a great business move from WSJ, get access to some of that investor money in an obviously sponsored content bit that could go viral.

That's... exactly what the author said in the post. But with the argument that those are cynical and terrible reasons. I think it's pretty clear the "you" in "why would you want an AI" vending machine is supposed to be "an actual user of a vending machine."

I think we're using very different senses of "deterministic," and I'm not sure the one you're using is relevant to the discussion.

Those proprietary blobs are either correct or not. If there are bugs, they fail in the same way for the same input every time. There's still no sense in which ongoing human verification of routine usage is a requirement for operating the thing.

Those are completely deterministic systems, of bounded scope. They can be ~completely solved, in the sense that all possible inputs fall within the understood and always correctly handled bounds of the system's specifications.

There's no need for ongoing, consistent human verification at runtime. Any problems with the implementation can wait for a skilled human to do whatever research is necessary to develop the specific system understanding needed to fix it. This is really not a valid comparison.

Sure, but how many LLM streaming clients are out there?

Namespacing, sure. But is "We use gh:someguy/openai/llm-streaming-client to talk to the backend" (x50 similarly cumbersome names in any architecture discussion) really better than "We use Pegasus as our LLM streaming client"?

Engineers at Google are much less likely to be doing green-field generation of large amounts of code . It's much more incremental, carefully measured changes to mature, complex software stacks, and done within the Google ecosystem, which is heavily divergent from the OSS-focused world of startups, where most training data comes from

I feel like none of these discussions can ever go anywhere, if they don't start from a place of recognizing that "AI is a massive bubble" and "AI is a very interesting and useful technology that will continue to increase its impact" are not mutually exclusive statements

I think the OP meant something far simpler (and perhaps less interesting), which is that you simply cannot encounter key errors due to missing fields, since all fields are always initialized with a default value when deserializing. That's distinct from what a "required" field is in protobuf

Yes, at about 1% of this scale. OpenAI's obligations are not something they can just run to daddy VC to pay for; he can't afford it either