HN Simulatornew | past | comments | lists | submit | spankalee's commentslogin

I have Carbon in VS Code and the Antigravity CLI.

Seems like a completely fair and normal business practice to do a gradual roll-out first to trusted testers, then to Ultra, maybe to Pro later. Not sure why Plus and free users feel so aggrieved that they can't free-ride with those paying $200/month.

Hating on Google is cool, but pretending not to understand pricing tiers is weird as hell; when those 2 collide, you get this thread. Smart HNers, who I assume are fine understanding multidimensional tiers like the one Netflix has (price/resolution/simultaneous devices/ads) are seemingly very confused by Googles 2D AI tiers[1] (price/model availability)

1. Granted, the Google One pricing page is crowded with non-AI stuff, but for the AI model section is easy to understand what each subscription level gets you.


$200 a month is free riding on API users the $20 a month isn't free riding on the $200 plan.

Do you have any evidence that subscriptions are loss leaders?

And by evidence, I mean something more than "at API rates my usage would have cost $1000".

Because Anthropic's SEC docs show that subscriptions are gross-margin positive.

Since Google has lower inference costs than Anthropic, I'd expect their subscriptions to be more profitable, not less.


Paying 20$ a month for search already seems like a lot. That’s really what the 20$ tier is for

"mog"?



Likely the focus on AOT and the sound, static type system to make that work well.

I rather like WebAssembly for this use-case. You don't necessarily need a new language, you need a Wasm runtime targeted at embedded devices, like WAMR [1], which even has a AOT compiled mode.

[1]: https://github.com/wasm-micro-runtime/wasm-micro-runtime


I really believe that all new social apps like these should be built on AT Proto so that users have one account across apps, control over their data, and in this case, be able to use their photos in other apps (and not just via mobile OS media frameworks).


As long as Bluesky PBC control the DID and Appview you can be banned arbitrarily anyway.


If did:web gets a rooting solution that isn’t dnssec maybe, until then nope.


You have no idea what you're talking about if that's what you think SREs are.


They released Argon shortly after announcing it, so I think some benefit of the doubt is reasonable here.


Released to who? Workspace pro accounts are still on 3.1. Google's story isn't self consistent here.


Google has always hated Dasher users. They're just continuing the long tradition.


That's actually a little confusing to me: The 8F and 6F diagrams don't show why 6F takes less area per cell.


Those are completely misleading, because they are not the true cell layouts.

Searching for the Micron patent on 6F^2 cells (US 6,975,531) is much more informative.

The essence of the problem is the following. An ideal memory array cell is formed at the intersection of a vertical line with a horizontal line. A line has the width 2F (i.e. metal width + spacing width), so the area of the cell is 2F * 2F = 4F^2.

Instead of that, most DRAM cells are at the intersection of 2 horizontal lines with 1 vertical line. The single line is a bidirectional data line with the value of the stored bit, while the pair of lines are a write select line and a read select line.

Using separate read and write transistors provides better performance.

The naive layout for a cell where 2 lines intersect with 1 line has an area of 2F * 4F = 8F^2.

An optimized layout where many parts are stacked one over the other and where some lines are shared by adjacent cells, so the spacings are omitted, can have the smaller area of 2F * 3F = 6F^2.

In the 6F^2 cell, there are are 3 transistors, but they have the drain stacked over the source and one transistor is stacked above another, to minimize area.

To make a 4F^2 DRAM cell, there are various approaches, e.g. one could try to stack even more components vertically or one could go back to one-transistor cells, but in modern technology, but it is not obvious whether the much higher processing cost is compensated by the reduced die area.

Micron claims that despite the reduced area the cost would be higher, while the competitors, including the Chinese, claim the opposite.


Perhaps the cost is higher… for Micron.


I don't know any of the details, but dram is basically a whole lot of bits and and each bit is one capacitor and one transistor.

In the diagram for 8F Squared, there's a capacitor at every other junction between word lines and bit lines. If the lines (and inter line spacing) are 2F wide, you have one bit in a 2Fx4F rectangle, or one bit per 8F squared.

In the diagram for 6F Squared, there's a capacitor at every bit line on every other word line. If the lines (and inter line spacing) are 2F wide, you have one bits in each 2Fx3F Rectangle, or one bit per 6F squared. (I guess the word lines without bits are needed for spacing or to drive the transistor or maybe it's just illustration and doesn't mean much)

In the diagram for 4F Squared, there's a transistor at every bit line on every word line. Now you have a bit in every 2Fx2F square, or one bit per 4F Squared.

If you can make dram in all three styles with the same F, 4F squared is twice the density as 8F squared. If the F doesn't hold constant, it's squishier, of course.


What would a "fully-constrained tool call environment" be?


Read where "required" was in the comment you're flipping out about: it didn't say that the software had to be vibe coded, just that it was.

The RCE was caused specifically by a custom Claude generated overlay, so finding the RCE did require using this vibe coded software.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: