Earlier quoted context omitted.
"Attention Is All You Need" was a paper by a bunch of Google researchers
Still, attributing that progress to "years of research at Google" alone is simplifying the facts to the point of being just plain wrong. That kind of research was always very much in the open and cooperative, with deep levels of standing-on-shoulders. Attention e.g. was developed by Dzmitry Bahdanau et al. (those being Kyunghyun Cho and Yoshua Bengio) in 2014 while interning at the University of Montreal. The insight…
Our eighth generation TPUs: two chips for the agentic era
131–140 of 240 posts
Re: Our eighth generation TPUs: two chips for the agentic era
#132Earlier quoted context omitted.
Gemini CLI isn't a great product unfortunately. While it's unfortunately tied to a GUI, antigravity is a far superior agent harness. I suggest comparing that to Claude code instead.
Bad software kills good hardware. And the converse is true also. I mean, look at NVIDIA. For the longest time they were just a gaming card company, competing with AMD. I remember alternating between the two companies for my custom builds in the 90s and it basically came down to rendering speed and frame rate. But Jensen bet on the "compute engine" horse and pushed CUDA out, which became the defacto standard for doing…
I make the mistake of thinking its 2020 as well. CUDA was announced 2006 and released Feb 2007. So its actually 20 years that AMD/RADEON hasn't caught on that they need a good software stack.
Re: Our eighth generation TPUs: two chips for the agentic era
#133In recent discussions about Tim Apple [sic] moving on there was a discussion about whether Apple flopped on AI, which is my opinion. Of course you had the false dichotomy of doing nothing or burning money faster than the US military like OpenAI does. IMHO that happy medium is Google. Not having to pay the NVidia tax will likely be a huge competitive advantage. And nobody builds data centers as cost-effectively as Goo…
Apple has not flopped on AI as you say. They are just focused on privacy and are likely waiting for the time when local models become efficient enough to run on iPhones (which is quickly becoming a reality). Google could probably train models for orders of magnitude less money as you say, but they aren't. They are not capable of creating high quality models like OpenAI and Anthropic are. Their company is just too dis…
>They are just focused on privacy
Is that why they rushed out introducing AI summaries etc in order to play catch-up and then backpedaled when they exploded in customers' faces/individuals concerned in false headlines threatened to sue?
I use Gemini on purpose all the time. It can start timers for me, add calendar entries without having to type it out, convert email to calendar or reminders etc. I'd use it even more if it had more access to other bits of my phone.
Re: Our eighth generation TPUs: two chips for the agentic era
#134Re: Our eighth generation TPUs: two chips for the agentic era
#135I already felt that gemini 3 proved what is possible if you train a model for efficiency. If I had to guess the pro and flash variants are 5x to 10x smaller than opus and gpt-5 class models. They produce drastically lower amount of tokens to solve a problem, but they haven't seem to have put enough effort into refinining their reasoning and execution as they produce broken toolcalls and generally struggle with 'agent…
> They produce drastically lower amount of tokens to solve a problem, but they haven't seem to have put enough effort into refinining their reasoning and execution as they produce broken toolcalls and generally struggle with 'agentic' tasks, but for raw problem solving without tools or search they match opus and gpt while presumably being a fraction of the size. Agreed, Gemini-cli is terrible compared to CC and even…
I thought the same for a long time, borderline unusable with loops/bizarre decisions compared to Claude Code and later Codex.
But I picked it up again about a month ago and I have been quite impressed. Haven’t hit any of those frustrating QoL issues yet it was famous for and I’ve been using it a few hours a day.
Maybe it will let me down sooner or later but so far it has been working really well for me and is pretty snappy with the auto model selection.
After cancelling my Claude Pro plan months ago due to Anthropic enshittification I’ve been nervous relying solely on Codex in case they do the same, so I’ve been glad to have it available on my Google One plan.
Re: Our eighth generation TPUs: two chips for the agentic era
#136I already felt that gemini 3 proved what is possible if you train a model for efficiency. If I had to guess the pro and flash variants are 5x to 10x smaller than opus and gpt-5 class models. They produce drastically lower amount of tokens to solve a problem, but they haven't seem to have put enough effort into refinining their reasoning and execution as they produce broken toolcalls and generally struggle with 'agent…
Interesting mix of words: "I felt" -> "proved" -> "guess". One of those is not like the others!
Re: Our eighth generation TPUs: two chips for the agentic era
#137Earlier quoted context omitted.
> If the whole AI bubble spectularly collapes Every other news for the past month has been about lacking capacity. Everyone is having scaling issues with more demand than they can cover. Anthropic has been struggling for a few months, especially visible when EU tz is still up and US east coast comes online. Everything grinds to a halt. MS has been pausing new subscriptions for gh Copilot, also because a lack of capac…
Is the internet bigger or smaller than it was in 1998 compared to today? Demand for internet and web services is significantly higher today than in 2000 but a bubble still popped. Heck a regular old recession or depression, completely unrelated to AI could happen next year and could collapse the industry. I mean housing is more expensive than ever nearly 20 years after collapsing in the Great Recession.
The problem that I have with dotcom comparisons is that people miss what popped and what remained after that bubble. Catsdotcom and Dogsdotcom popped. But the tech remained, and now we have FAANG++.
If we apply the same logic, any of oAI, xAI, Anthropic might pop, but realistically they won't, and even if they do, some other players will take their spots, and the tech will survive, and more importantly the demand will still be there. This cat isn't going back into the bag. People want this now. More than all the providers can give them. Today. The demand won't suddenly disappear now that "we got a hit" like someone put it recently.
Re: Our eighth generation TPUs: two chips for the agentic era
#138Whats interesting to note, as someone who uses Gemini, ChatGPT, and Claude, is that Gemini consistently uses drastically fewer tokens than the other two. It seems like gemini is where it is because it has a much smaller thinking budget. It's hard to reconcile this because Google likely has the most compute and at the lowest cost, so why aren't they gassing the hell out of inference compute like the other two? Maybe a…
I'm 50% convinced that the main lift in GLM-5 over GLM-4.7 was that it was much more willing to use tokens. I had the hardest time getting 4.7 to read enough source code to actually know what it was doing, but once I convinced it to read, it was pretty capable. Being thrifty can be good! But it also can mean your system is not reflecting sufficiently, is not considering enough factors, isn't reading enough source cod…
I’m pretty convinced that this is what Kimmi 2.6 does, mostly just thinks more.
Re: Our eighth generation TPUs: two chips for the agentic era
#139I already felt that gemini 3 proved what is possible if you train a model for efficiency. If I had to guess the pro and flash variants are 5x to 10x smaller than opus and gpt-5 class models. They produce drastically lower amount of tokens to solve a problem, but they haven't seem to have put enough effort into refinining their reasoning and execution as they produce broken toolcalls and generally struggle with 'agent…
Am I tripping or is this an AI reply? Like it barely has anything to do with the article other than both are related to AI
Re: Our eighth generation TPUs: two chips for the agentic era
#140At this point, when you are doing big AI you basically have to buy it from NVidia or rent it from Google. And Google can design their chips and engine and systems in a whole-datacenter context, centralizing some aspects that are impossible for chip vendors to centralize, so I suspect that when things get really big, Google's systems will always be more cost-efficient. (disclosure: I am long GOOG, for this and a few o…
I'd bet that too if their management wasn't so incredibly uninspiring. Like, Apple under Cook was also pretty mild and a huge step down from Jobs, but Google feels like it fell off a cliff. If it wasn't for OpenAI releasing ChatGPT, they might still be sitting on that tech while only testing it internally. Now it drives their entire chip R&D.
Google was calling itself an "AI-first" company beginning in 2016 or 2017. They designed and built TPUs nearly a decade ago and were using transformer models in products like Google Translate but didn't make a big fuss about it, it just made the product way better. People should at least credit Sundar somewhat for this, it turned out to be quite prescient, especially the advantage of having your own chips that are specifically designed for ML.