@mukund If Anthropic fails, google would still gobble all the compute. Given the scale of their products, I can't imagine how they're even selling TPUs.
English
Serene Trails
867 posts
$GOOGL seems to be ok, but still 50% of all committed spend is from 2 startups that might be too big to fail now
Jessica Lessin@Jessicalessin
Good lord. Half-ish of the cloud backlog at Microsoft, Oracle, Google and Amazon is OpenAI and Anthropic????
English
@sama @DaveShapi man your vibes are different these day (love it); how do you tweet so much while dealing with day 2 day oai stuff? is codex running X for you?
English
I unblocked Sam and this is what I come back to.
Sam Altman@sama
artificial goblin intelligence achieved
English
@dropalltables not yet, they will make money from the current compute shortage; will use it to fund AI integration in their products. when their distribution becomes frontier, then they'll crater the price.
English
I suspect Google will crater the cost of tokens because a) in the fullness of times, tokens are a commodity, b) your ability to produce a commodity depends on your cost of capital and your ability to turn capital into the commodity, and c) Google is exceptional at both.
English
@_arohan_ oh man, surprised to see this from you. google cloud lets their customers choose any grounding tool that they want (including grounding on their 1P data). It has been there for a long time and exa is just one among many.
#use_of_alternative_search_engine_options" target="_blank" rel="nofollow noopener">docs.cloud.google.com/vertex-ai/gene…
English
This is so weird, why is Google, the world's biggest search engine, using another startup's search engine?
Exa@ExaAILabs
We're excited to partner with Google to offer Grounding With Exa inside of Gemini models! Using Exa's agent-first search, Gemini models can now access billions of websites, technical docs, papers, people, companies, and more. 10^18🤝10^100
English
@Turn_Trout A better approach would be to elect better leaders who won't abuse law. Google is just supporting their government to use AI as per permissible law.
English
I spent the last 2 months trying to prevent this. If OpenAI offered a fig leaf, Google said "imagine we offered a fig leaf."
Google affirms it can't veto usage, commits to modify safety filters at government request, & aspirational language with no legal restrictions.
Shameful.
English
@kimmaicutler agree that tax is a terrible idea, but so is "voting shares"
English
None of the billionaire tax coverage about Sergey Brin in Bloomberg or the NYTimes reports the most important structural detail of why he opposes it so much more than Jensen. It taxes voting shares as ownership.
It taxes Page and Brin as if they own 50+% or >$2 trillion of Google rather than 3% each. Jensen does not have super voting shares so he is just taxed like his actual ownership of Nvidia.
Imagine if you were taxed like you earned 10 times as much as you actually did last year. @bizcarson @teddyschleifer
English
@kimmonismus That's not correct. Google is probably as compute constrained, maybe even more than other labs because they serve dozens of AI products to billions of users and enterprises. half their compute is only for cloud customers.
English
@jack_w_rae congrats! a good harness is more differentiated. point in case codex
English
Nice debut from Muse Spark in the Agentic Coding arena.
Ranking ahead of GPT 5.4 and Gemini models, and behind the Opus series + GLM 5.1.
Arena.ai@arena
Muse Spark debuts at #7 in the Code Arena - making @AIatMeta the #3 lab right behind @AnthropicAI’s Claude Sonnet 4.6 and @Zai_org’s GLM-5.1, surpassing Gemini-3.1-Pro and GPT-5.4. Code Arena evaluates agentic coding on real-world tasks - building live websites and apps, ranked by users on real workflows. Huge congrats to @AIatMeta on this impressive milestone!
English
@PatrickMoorhead it is just silly reporting. google has a history of creating separate training and inference chips e.g. tpu v5p and v5e.
English
I don’t quite understand this trope: “its latest effort to take on Nvidia in AI hardware.” It’s not Google taking on NVIDIA, it’s Google looking for optionality and using TPU primarily for its own services.
If Google really got into the business of selling mass infra that could go into many, standard datacenters, that would be “taking on” NVIDIA. There is nothing standard about the Google’s Hypercomputer; it’s dialed in, after decades, currently, very proprietary for Google’s needs.
Saying Google is taking on NVIDIA’s chips is like saying Apple chips are taking on Intel and AMD. Apple may be going after Dell, HP and Lenovo, but not Intel and AMD.
$NVDA $GOOG
CNBC@CNBC
After years of producing chips that can both train artificial intelligence models and handle inference work, Google is separating those tasks into distinct processors, its latest effort to take on Nvidia in AI hardware. Google said Wednesday that it’s making the change for the eighth generation of its tensor processing unit, or TPU. Both chips will become available later this year. More details: cnb.cx/4mNsKLt
English
@blueemi99 It's a good thing. last time this happened, we got gemini 2.5 pro. now gemini 4 has to be good and has to come fast .. love the competition!
English
@scaling01 it is amazing they're able to do this with much less compute.
English
I do think Kimi, DeepSeek, GLM and ByteDance are close to ~Opus 4.5/GPT-5.2 reliability or will be there in the next 1-2 months
so they are reaching escape velocity too
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex
They say "12 hours" or "13 hours" but of course wall clock time has nothing to do with METR time horizons which are established relative to human labor. Still I wonder: have Chinese models finally reached the SWE takeoff stage? 3.6-Max, 5.1, 2.6. It does feel like a step change.
English
The best open models are those you can run in your devices 🤏 What do you want to see next as we keep evolving Gemma?
English
@OfficialLoganK @GoogleAIStudio Logan, it will help to understand what kind of quota do we actually get with AI Studio, Gemini CLI, Antigravity, etc when on pro or ultra plan. Even better if we can see the usage. Thanks for amazing works!
English
Excited to share that Google AI subscriptions (Pro and Ultra) now work with @GoogleAIStudio!!
Come vibe code and use the playground with higher rate limits. Available right now!
English
@JFPuget or maybe .... one is CEO of a for-profit company and other is a podcaster playing devil's advocate .... just maybe!
English
My take on Dwarkesh - Jensen interview.
One is a visionary, the other one is regurgitating the latest blog he liked.
English
@LexnLin That's also the week of google cloud Next launches; given openai's infatuation to outshine google, next week makes sense
English
OpenAI vagueposted A LOT about cheesecakes
April 23rd is the National cherry cheesecake day
And polymarket spikes on that day
so yeah expect GPT 5.5 next Thursday
Leon Lin@LexnLin
next thursday 😭as I guessed
English
@OfficialLoganK bro, we've got all kind of 3.1 flash-xyz except for the main 3.1 flash! congrats on the launched.
English
Introducing Gemini 3.1 Flash TTS 🗣️, our latest text to speech model with scene direction, speaker level specificity, audio tags, more natural + expressive voices, and support for 70 different languages.
Available via our new audio playground in AI Studio and in the Gemini API!
English
@MossyPathways @soldni oh it does (like 90% will be cache tokens)
English