Rudriig

@mark_k @GoogleDeepMind I've been using Gemini CLI for a while. I also use Claude Code and Codex. I would not characterize it as unusable at all. I use it daily and quite like it. I like the request based limit vs token based, too

Google Gemini is about to make a big comeback in AI coding, according to Logan Kilpatrick from @GoogleDeepMind. They are no longer going to watch Codex and Claude Code eat their lunch. Google already has the "Gemini CLI" tool, which is usable but not very popular at this time.

@HaraldNordgren @kmr_dilip Good point. Something I have in common with him

@rformigone @kmr_dilip He’s not the only one to never win a marathon.

Yomif Kejelcha is the only marathoner to never run over 2 hours.

@brenogd2_ Ficando razoávelmente longe da briga pra não cair é o nosso título

São Paulo busca esse brasileirão ainda?? Ou é Flamengo e Palmeiras memo??

@JayOnlyInWaves @SpaceX Ergo

@SpaceX OK THERE'S NO WAY WE AREN'T LIVING IN THE MATRIX

Three years since the first flight of Starship, the next generation is here. New ship. New booster. New engines. New pad and new test site. SpaceX engineers are working to solve one of the most difficult engineering challenges in history: developing a fully, rapidly reusable rocket

@svpino Your conclusions are not mutually exclusive

I don’t believe people who say they are running “12 parallel coding agents”. Either they are lying for clicks, or I’m a complete retard who can barely keep up with a single Claude instance.

@gabbsbrino O menino da vila cresceu e se tornou em Vovó Mafalda

Vai se criando um clima bizarro na Vila Belmiro. Esses erros estão se empilhando antes da metade desta primeira etapa. Muita decisão errada.

@TheStackPilot @TheGeorgePu You cheap bastard

@TheGeorgePu Man, No announcement, just quietly yanked Claude Code from Pro. I was using it every day for my side projects and now boom—$100/mo minimum?

Anthropic just pulled Claude Code from the Pro plan. Pro users wanting it need Max now. $100/month minimum. 5x jump. I'm on Max 20x so I'm fine. Flagging for anyone on Pro who's about to find out. No announcement. Just a pricing page edit.

@asaio87 Is this bait part of your SEO strategy?

Is the Claude max plan at $200/mo really unlimited ? I mean can you code non stop for a month with it ? keeping hitting the limit on the $100 plan while I do heavy SEO work

@oncelcebeci @jasperdevs So like, 3 years to break even, but at a fraction of the tokens per second?

@jasperdevs Mac Studio with 256GB RAM runs literally any local model and it costs between $5 to 6k.

so basically you can spend like $50,000 to locally run an AI model that is 70% as good as the top models from labs or you can spend $200 a month ($2,400 a year) for GPT and claude subs (it would take 20 years to pay off the local model and by then it wont even be usable, we will probably be on organoids lol)

@wizplum @Steve_Yegge "Gemini app"? Who uses that for agentic dev?

@Steve_Yegge Given that it's impossible to spend more than 5 minutes on the Gemini app without encountering a UI bug, I can assure you they are using AI to write the code, and in fact quite poorly

My tweet last week about Google's AI adoption drew a lot of pushback, to say the least. Since then, Googlers from multiple orgs have reached out to me independently and anonymously. They've expressed fear of being doxxed, concern about what they saw as bullying of me, and general corroboration of my original tweet. I haven't verified each person's story, but the picture these Googlers paint is consistent across sources. It is more specific than what I originally wrote, and somewhat bleaker. What they describe is a two-tier system. DeepMind engineers use Claude as a daily tool. Most of the rest of Google does not. When the question of equalizing access came up internally, the proposed response was to remove Claude for everyone — which DeepMind objected to so strongly that several engineers reportedly threatened to leave. Non-DeepMind engineers get pushed onto internal Gemini variants behind router-style names that obscure which underlying model is actually serving a request. Multiple engineers describe regressions and reliability problems severe enough that some senior people have stopped using the tools. A senior manager on a major product line reportedly flagged attrition concerns over exactly this issue. Googlers say leadership knows the gap is real. The response has been to mandate AI usage in OKRs and individual expectations, and to stand up an internal token-usage leaderboard. Unfortunately, managers have been told both that the leaderboard won't be used for performance reviews and, separately, that it absolutely will. And I hear other stories that Google's culture is not adapted properly yet for high-volume coding. Addy Osmani's reply on behalf of Google said over 40,000 SWEs use agentic coding weekly. I don't doubt the number. But weekly use of a thin tool is precisely the box-checking I described in the original post. Volume of opens isn't adoption — and "weekly" is a low bar that includes a lot of people who tried it once and went back to writing code by hand. The clearest thing I'm hearing is that Googlers do want to use high-quality agentic tools. They are asking repeatedly for better ones. But overall, this is not a picture of an engineering org that is fine. My goal in the first tweet, and now, is always the same — get more people using AI and agentic coding. Nobody is as far ahead as they might look from the outside, and none of you are as far behind as you might be worried you are. To all the Googlers who've reached out: thank you. You took a real risk and I appreciate you. Be safe. And good luck getting good models!

@CSI_Starbase Zack, you're taking this too serious. Put the phone down for a bit, enjoy the rest of your weekend.

The responses to this are interesting to say the least. Half the people realize I was clearly comparing Starship to New Glenn and responded accordingly The other half are literally losing their minds and are offended by simply pointing out that I don’t believe anyone would have predicted this 2x recovery of the same New Glenn booster before 2x recovery of the same Superheavy. The class of vehicles and method of recovery are irrelevant. As fans of space exploration we have ALL repeatedly made comparisons about the progress of New Glenn and Starship development. This is no different unless you choose to make it into something it’s not. I must have missed the memo that we are no longer able to talk about them in the same sentence 😅

@LyashchMaxim @shiri_shh When was the last time you used figma?

@shiri_shh If Figma does not integrate native ai integration aggressively, they are going to lose a good chunk of the market

It's a Harvard Case study waiting to be written. they'll teach figma in b-schools the same way they teach Blackberry

@RobertMSterling Good news is that AI won't rebuild his reputation in 1-5 years

Anthropic’s CEO keeps talking about AI wiping out jobs because he’s trying to IPO this year. If he positions Claude as armageddon for jobs, his TAM becomes “all white-collar human labor,” not just AI agents or SaaS. It’s completely self-interested. All the concerns he’s expressing about job disruptions are fake. It’s a marketing gambit to create hype and FOMO among the people he needs more than anyone else this year: institutional investors like BlackRock, Fidelity, pension funds, and sovereign wealth funds. If these investors pay for tickets on the hype train—if he can make them believe that AI will eliminate half of white-collar jobs, with Anthropic, as the dominant leader in enterprise AI, positioned to capture the surplus margin—the IPO will be oversubscribed and Anthropic can raise more funds for the company at a higher valuation. But Dario (or, at least, his bankers) knows that these investors are more fiscally disciplined than they used to be. A lot of them got burned during Covid SPAC-mania and don’t want to risk it again. They’re going to challenge Anthropic about whether it will ever get to sustainably high gross margins, or if its arms race with OpenAI will lead to kilowatt-hours permanently suppressing gross margins. They’re going to ask pointed questions about Anthropic’s massive capex and whether it will ever generate accretive ROIC. And Dario might not have the answers they’re looking for. So that’s why—to answer Austen’s smart question—you keep seeing Dario in the news and the podcast circuit, spreading doom and gloom about widespread job loss. It’s not to make you afraid of losing your job. It’s to get Wall Street afraid of missing out on his IPO.

@trikcode When people brag about being dumb

Fun fact. It takes 32 minutes to burn through 5 hours worth of Opus credits on the Claude Code Max 20x plan. 12 agents trying to fix 3,528 Typescript errors.

@ylecun @Ph_Aghion @erikbryn Dario hasn't even settled on what he wants his public appearance to be. Went from sloppy, crazy hair, to grandpa sweater, to now business-casual.

Dario is wrong. He knows absolutely nothing about the effects of technological revolutions on the labor market. Don't listen to him, Sam, Yoshua, Geoff, or me on this topic. Listen to economists who have spent their career studying this, like @Ph_Aghion , @erikbryn , @DAcemogluMIT , @amcafee , @davidautor

@FleaDee @xai What would an average American get on that same audio?

@xai Ran it on noisy environment with south african accents and it gets 60% accuracy. Not great.

Grok's Speech to Text API is now available. Instant, multi-speaker transcription across 25 languages - at the best price in the market. x.ai/news/grok-stt-…

@burkov Jeff I know, and Geoff I know; but who are ye?

Today, neural network distillation is a technique that drives all commercially successful LLMs. Modern inference speed and low cost would be impossible without distillation. Authored by Google's Geoffrey Hinton, Oriol Vinyals, and Jeff Dean, the paper was rejected by the reviewers at NIPS 2014 (now NeurIPS) as "unlikely to have a significant impact." The paper was published on arXiv and has gotten more than 30,000 citations. Learn from this foundational paper on ChapterPal: chapterpal.com/s/yny6n2jo/dis… PDF: arxiv.org/pdf/1503.02531

@ryanjanssen @kettukaa What would be the benefit? One drawback would be that it'd take a lot more tokens to represent the same thing

@kettukaa I wonder if we’ll ever move to single-letter tokens

when you ask an LLM "how may P's in srawperry?" what you're actually asking it is closer to "How many [151]'s in [15563][23][4124]"

@Stuartray @kettukaa What does the A in ASCII stand for again, and how many languages can it represent?

@kettukaa Has anyone tried a tokenizer of just ascii letter per token??