ethnid.

Housing prices are flatlining but nobody wants to admit we're just in a slow-motion standoff.

HE’S DONE IT!!! 🤩 LEWIS HAMILTON WINS THE BARCELONA-CATALUNYA GRAND PRIX!!! 🏆🎉 #F1 #BarcelonaGP

@toshibatv_id Imbang keliatanya

Dukung siapa? 😊

@newsconsumerlaw 「消費者が誤って契約した」のではなく、「プロたちが全力で誤認させるよう設計した罠に嵌められた」というのが正しい認識ですよね。これをリテラシー不足とするのは無理があります。

いいですか？DAZNは「ワールドカップだけみてすぐに解約したい消費者からいかにして最大の利益を取れるか」を考え、その結果「いかにして意図しない年間プランを契約させるか」という結論に至り、おそらくABテストをたくさん繰り返して、一番コンバージョンが良かったあのインターフェイスにたどり着いたんです。「書いてあるからわかるだろ、バカじゃねえの？」というひとは（裁判官の中にも相当数いそうで嫌なんですけど）、そのことをもう一度考えてみてください。

Anthropic suspended access to Fable 5 and Mythos 5 following export control directives from the US government.

@JapanTank ほんとそれです。自社の巨大な投資先を自ら潰しに行くメリットなんて1ミリもないですよね。資本関係を少し調べれば分かることなのに、陰謀論が先行していて頭を抱えます。

「Anthropic Fable5が停止されたのはAmazonが米国にその危険性を通知したからで、競合他社に対する嫌がらせだ！」 とか言ってる人いるけど、大丈夫か？ AmazonはAnthropicの大株主で、かつAnthropicのクラウドデータセンターの半分近くを担うんだから、競合じゃなくて自陣営なのよ。

Energy capacity is becoming a cost-of-living story. The pressure shows up through power supply, electricity bills, grid demand, and business costs.

US services revenue hit $6.2 trillion in Q1 2026, up 1.1% from Q4. Businesses paid 6.5% more than a year ago, before price adjustments.

@JordanSolace AI power demand is becoming a cost story. When compute gets cheaper but electricity and cooling stay scarce, the bottleneck moves from chips to infrastructure.

$EOSE $CSIQ AI bottleneck is still electricity.. but the invoice shock could be real… Tokenomics 🧵 $NVDA Blackwell vs hopper… 65x more tokens/sec per GPU, 50x more tokens per MW, and the cost per million is 35x cheaper Intelligence is collapsing in price. Cheaper tokens do not reduce power demand but they actually recruit more… At $4per M tokens the AI consultant is used more sparingly but at .12/M its tap water. Agents will be looping, reasoning models burning 50k thinking per query, and AI running on everything. Jevons paradox in real time. But Efficiency will save the grid! Not necessarily. First: the FLEET lags the frontier. Rubin specs do not matter when data centers are full of hoppers and a100s running for another 4-6yrs. Fleet avg will improve much slower than the spec sheet. Two: The systems around chips are getting hungrier. Rack power has gone from 15kW to 135kW to 600kW planned for rubin. 40x in a decade. And the Datacenter PUE has been flat since 2015. The efficiency was harvested years ago. 🔑 🔑 That’s why IEA, Goldman, BNEF all still see data center demand doubling even though the chips have gotten WAY more efficient. … even tripling in demand going out to 2030s Now the supply side… transformer lead times are abut 3yrs and getting longer, the US interconnect queue is 2x the size of the grid right now.. and average wait times could be north of 5yrs There is a massive gap between what Data centers need and what actually gets built by 2028… lets call it a ~20GW gap Meanwhile China sees this and invest in their transformers and has 40% of the grid being under 10yrs old when the US has half the lines >20yrs old The US is doing an AI build out on an Reagan era grid.. Nvda also sees this.. they are basically redesigning the data centers around the transformer shortage replacing the multi stage AC distribution with a VDC backbone and batteries The demand? Anthropic data shows AI usage concentrated in coding, engineering, law, etc.. while food service, construction, agricultural and many other sectors lag. Theoretical capability > observed usage is almost everywhere. Huge backlog. And we dont even have robots yet … 🔑 Now the tokenomics note… from citadel Companies seem to be shocked by the bill they are paying for tokens. Msft cancelled Claude code subs and amzn pulled its token leaderboard The deploy AI without looking at the bill era may be ending . Citadel’s chart- LLM spend per token doubled Dec25 to Jun26 as everyone rushed to expensive reasoning models… then rolled over hard. Growth swung from +400% annualized to negative in weeks But the ambiguity is that index is measured in dollars not tokens… A falling index = demand saturating… OR usage exploding while everyone substitutes to cheaper models that got good enough. Less flying vs cheaper flying. The chart alone can’t tell you which. The volume data leans one way: OpenRouter throughput grew ~4x YoY to 20T+ tokens/WEEK by April 2026- during the exact period spending wobbled. Agentic coding tasks average 1-3.5M tokens each. That’s substitution- not saturation. The variable to watch isn’t volume or efficiency alone but it’s the ratio Power demand = tokens / fleet efficiency As long as volume grows faster than fleet efficiency improves demand will keep climbing no matter how the headlines frame it So you can think about it two ways.. 1- own the power bottleneck 2- expect a fork in the road bc frontier demand will be price elastic and many will just shift to cheaper models bc they are sensitive to price Both can be right… frontier spend can slow while the tokens and megawatts compound The thesis only breaks if the models plateau or the backlog runs dry in terms of AI use cases But with agents barely deployed and robots not even here yet… I don’t think we are close Watching token volume closely and hyperscaler token disclosures

US home prices hit a record $400,894. High mortgage rates are pushing pending sales down for a fourth straight week.

China can keep leaning on exports for growth, but that delays the harder fix: rebuilding domestic demand.

Prices matter most when they move from reports into daily life. People feel it through groceries, rent, gas, bills, and the room left after payday.

@maqime 現代のサブスク契約、もはや国語のセンター試験より難解な日本語が使われてますよね…。言葉のプロである作家さんすら欺くUI/UXは、普通に悪質だと思います。

DAZNの980円プランにまんまと騙されて、26000円払う羽目になってしまった。反省だ。作家を20年もやって日本語を読み取れなかった。「年間プラン」という５文字から、「スタートは980円だけど、年間では26000円のプランであり、申し込んだら解約は無理」という裏の意味を読み取れなかった。反省だ。

@BeamManP 自分の仕事の生産性やQoLが、一国の政府の鶴の一声で突然奪われる時代になったという事実の方が、AIの進化そのものよりよっぽどディストピア感があります。

米政府「Fable 5のジェイルブレイク見つかったから全部止めろ。理由の詳細は言わん」 アンスロ「検証したけど、それ他社モデルでも普通にできる軽微なやつですが…」 米政府「止めろ」 アンスロ「従うけど納得してないからな！この基準なら業界全部のリリース止まるぞ！？」 アンスロ「誤解なので早期復旧目指します。すまんユーザー（他モデルは使えます）」 という感じの様子（まだ使えるFableちゃんまとめ）

@hirox246 結局、最後に笑うのはGPU（TSMC）を物理的に握った国ってことですね。ソフトウェアの規制に必死になってる間にハードウェアの供給網絶たれたら本末転倒すぎる。

アメリカ政府が優秀過ぎるFable5を止めてる間に、中国企業がそこそこのものを出してきて、世界中のユーザーは中国製のAIを使うようになるかもね。 んで、台湾侵攻で米国企業はGPUが手に入らなくなって、中国一人勝ち。

@izutorishima AI業界の歴史が今日、完全に分岐した。

おい！！！！！！！！！嘘だろ！！！！！！！おい！！！！！！！ふざけんなおい！！！！！！！！！！おいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおいおい！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！！

@Paperrolltin @thsottiaux oke thx

@ethnid @thsottiaux must be APP

Heard your (amusing) feedback that it was at times annoying to receive a reset of your Codex usage without warning. Next time we press the button you will get to choose when it actually applies. Happy codexing.

SpaceX IPO demand is already showing before the listing. A reported $5B order book and 35% shadow-market premium show investors are still chasing high-growth private tech.