smlacy

What would you suggest instead?

> TPU 8i pairs 288 GB of high-bandwidth memory with 384 MB of on-chip SRAM

Wow. Just Wow. I presume that's for each chip, and there are 1152 chips in a pod so that's 331TB HBM and 442TB SRAM per pod. Just wow.

Is it me or is that aluminum already developing some stress cracking?

Post actual results, make a blog post. Don't just say "this sucks" without tangible evidence.

Otherwise you're doomed to "sample size of one" level of relevance.

The synth engine is an implementation of https://soniccharge.com/microtonic Not sure what "accurate" means in your context though?

Yes, and with virtually zero context, which makes an enormous difference for TTFT on the MoE models.

Total gimmick. I guess we're "making progress", but this is will never lead to any useful application other than "Yes, you're absulotely right" bots. What's needed for real applications is 10000× the input token context and 10× the output token speed, so we're off by a factor of ... 100,000×?

This should be the top comment

And with only like a dozen tokens of context. What happens when this thing gets the ~100k tokens of context needed to actually make it useful?

I like the analogy but which 2 is AI coding?

Fast & Cheap (but not Good?) - I wouldn't really say that AI coding is "cheap"

Cheap & Good (but not Fast) - Again, not really "cheap"

Fast & Good (but not Cheap) - This seems like maybe where we're at? Is this a bad place?

What's your concern?

That background looks like AI for sure though?

Seems to be missing the Etrog? https://en.wikipedia.org/wiki/Citrus_taxonomy#Citrons

Also, the "click to show search results" is cool but fails for "Arizona Citron" in obvious ways.

Been true on most chromeos devices for a while!

There are certainly cases of Greek/Latin without any punctuation at all, typically in a historical context. Chinese & Japanese historically did not have any punctuation whatsoever.

Not all languages have such well-defined and commonly used delimiters. Is this "English only"?

Yes! I'm in the same boat! My guess is that there's something in the backend (image cache/storage size) that presents as a cost/logistical problem instead of as a technical one.

I don't get it. What am I looking at?

Is there a linux equivalent of this setup? I see some mention of RDNA support for linux distros, but it's unclear to me if this is hardware-specific (requires ConnectX or in this case Apple Thunderbolt) or is there something interesting that can be done with "vanilla 10G NIC" hardware?

Still doesn't exactly say what it is? I get that it's glyphs for printable characters, but honestly it could be a PDF, video, collection of PNGs or SVG files, an Adobe Illustrator file, a linux distribution, a web browser, or pretty much any other piece of software or data format. I presume it's a TTF or OTF font file?

You think a fighter jet should run Ruby on rails instead?

What's the pricing like? Taking an Uber/Lyft all the way from Mountain View to SF is outrageously expensive, I presume Waymo is the same?

Isn't this basically fiberglass?

The ultimate "out of sight out of mind" solution to a problem?

I'm surprised that Google has drunken the "Datacenters IN SPACE!!!1!!" kool-aid. Honestly I expected more.

It's so easy to poke a hole in these systems that it's comical. Answer just one question: How/why is this better than an enormous solar-powered datacenter in someplace like the middle of the Mojave Desert?

I presume the motivation is performance optimization? It would be more compelling to include some of the benefits in the announcement?

Just think of the untapped market of fresh 9 year olds who've never seen/played the game before. It's infinite, there will always be more people who have never played Minecraft.

but ... why?

Someone found the literal HCF instruction.

This looks awesome, but there's absolutely no way I'm installing an app for this, sorry.

Oh it's exciting, but not as exciting when sama pumps GPT-5 speculation and the market thinks we're a stones throw away from AGI, which it appears we're not.