B-Con

Google's style guides tend to reduce namespace redundancy.

Everyone remembers finding some hideous name like foo.widget.widgetBuilder.widgetFactory in Java or something. It's nice to have naming rules, or at least guidelines, focused on trimming redundancy and keeping the signal high.

It's almost as if we're building something that... mimics intelligence.

The US has proved it can do horizontal scaling, as the majority of adults gets a college degree.

The horizontal capacity is ineffectively used by having many degree mills and low-value degrees, but if we re-tooled the knowledge sets we could raise the bar.

Not OP but I've tried using all the major frontier models to find niche items in a field that I collect. The models get confused and hallucinate items that sounds complete reasonable, but don't actually exist. Usually it's an amalgamation of several real products.

My prompt is akin to "recommend <item type> with <niche criteria>". The first 3-ish results are about right, and then 7 of the next 10 are hallucinations and the LLM clearly can't throw up its hands and say "I got nothing".

I'm sure this is a hard problem because of a) how many items there are, b) how much overlap there is between product names, descriptions, manufacturers, different versions of the same product, etc, so keeping them distinct in the model's memory is probably hard, and even worse if it is dynamically fetching and summarizing content then it will be very easy to conflate different items, and c) LLMs are known for not working well on the edge cases with few examples.

I read Google's business incentive as "in it to not lose". AI is a new consumer endpoint and Google currently captures a lot of the endpoints.

> I think HN skews coding agent focused

100% this. Coding agents are an interesting test ground, but the people who care about them are a fairly small bubble.

I think they carry some of the overall AI weight because of the "what if you can vibe code your entire business" moonshot, but that's still orders of magnitude away and who knows if today's coding agents will actually be a stepping stone to that. If we ever get there it will likely be with entirely new domain languages.

I've learned this applies to a lot of life. Being a good manager, tech lead, consultant, etc, advisor, parent, friend, etc, is sometimes just half being a good therapist and helping them regulate.

I've watched a lot of "not officially financial advice" finance videos on YouTube (the solid people, not grifters), and while the financial theory side is interesting, when they talk about pragmatic investing and patterns of client behavior they have dealt with professionally, a large part of it is emotion management. Convincing clients to stick with a solid plan even when this month is abnormally bad, or avoid going all-in on the latest hotness, etc.

Go seems like a pragmatic fit, even though it was obviously not designed for LLMs and nor is it purely functional.

I review a lot of Go code, which means I review a lot of LLM Go code. Even though the human vs LLM authorship distinction has strong signals, Go's simple nature seems like a useful constraint on how LLMs can express themselves.

I think one of the key differences is that math is abstract whereas CS is relatively concrete.

CS examples are often easy to picture and understand the motivation for. You can use tools to visualize or play around with them and test them.

Math gets abstract so fast you have to spend a week of research to even understand the problem statement. The the motivations themselves can be completely unclear until you have a lot of context.

I majored in math (B.S.) and upper level math is completely foreign to me.

A human reader for a popular product is generally a paid actor who reads the script nearly verbatim, not a domain expert, so it has no bearing on the technical quality of the script.

I can see them as a shared fate for a couple reasons:

* the obvious one is Elon - both valuations are largely propped up on belief in Elon. Whenever he falters, his companies that are speculation-based (all of them) will take a hit

* Elon pitched SpaceX as an AI company. Tesla needs better AI because they keep sending signals that they won't be at L5 anytime soon, and Tesla's valuation is still very speculative[0]at least in part due to the race to L5 autonomy. i.e. Tesla will need better AI , and SpaceX is that natural fit (on paper, at least, I'm not sure SpaceX has any useful AI for any use case, let alone self-driving).

[0] Tesla's PE ratio of is still 30x massively out of line with it's actual earnings and ~30x the American automotive industry.

Same.

And since I won with an age, company, and title that is literally me right now... Well, I'm not exactly sure what to do with it information but some internalization is in order.

I'm simply impressed they're releasing at all. This has to be literally the worst 6 month time window in the last 20+ years to launch a new computing device at scale and have to build the vendor contracts and inventory from zero.

Yeah, I imagine Linux support would be more like a supported Linux distro rather than generic Linux support. Something like SteamOS but with kernel anti-cheat and secure boot from the start.

Big question is whether they can make craching the anti-cheat it hard/unpredictable enough that the publishers will trust it. If the publishers release such a platform and someone releases a live distro that can crack it with 3 mouse clicks, that's a lot of wasted effort.

I have no idea how effective the Windows anti-cheat is, but I imagine that Linux tooling in general is going to make it harder to lock a user out of controlling their own machine.

Even though I balked at the Steam Deck prices on the recent inventory restock, as they were up ~30% presumably due to the same hardware shortages, I got one anyway. Prices won't drop anytime soon and if any for-profit organization has earned my loyalty, its Valve.

When I used it I was somewhat incredulous that I could simply exit Steam mode have an actual Linux desktop environment, where I could literally do what I wanted. It was my computer, a proper general purpose computing machine, and it was (willingly* in my control. No sneaky root needed.

People want to do X, so the metric is how much X can be done.

Everyone is over-complicating the explanation. The answer for "why are we fixating on this bad metric" is almost always the same pattern.

Broad audiences need simple metrics to talk about. If the metric itself requires nuance, it's hard to communicate and hard to reason about. It's easier to push the need for nuance from understanding the metric itself down the road to where the metric is applied, which allows everyone to ignore it in immediate conversation.