dai

[dai @yufanfei]

@cHHillee would it be recorded?

[dai @yufanfei]

@andrewc44104127 @grok @grok how long does it usually take for a company like microsoft or google to restore a global service like outlook? is this incident unusual?

[dai @yufanfei]

@nekofrize 世界就是不完美也是无序的所以才有熵这个概念 所以也没必要过分强求内在秩序 100%都是完美的 大体有序就不错了

[dai @yufanfei]

Perhaps I'm missing something, but fusing everything into one kernel isn't always the optimal solution as you'd have to use the same level of parallelism across all layers, forcing that kernel to use more regs than what it should use. There are ways to overlap kernel launch as well

[dai @yufanfei]

@RepYassAnsari 中国人民的好朋友🫡

[dai @yufanfei]

@SnowyWar36965 好奇 我怎么感觉opus说人话能力还是差一截呢 尤其是解释东西经常漏context

[dai @yufanfei]

I’ve tested GPT Astra and Sol extensively for over three weeks on my perf and data heavy project till Opus 5.5 came out. The GPT family just can’t generalize and often wrote inefficient and code that’d panic if data/load pattern changes.

I’m leaning on this is a benchmark issue by seeing Sol being ranked #1

[dai @yufanfei]

@NobelPrize @grok why Tu Youyou was mentioned for this years laureate? Based on this post what’s your prediction?

[dai @yufanfei]

@trq212 I still use plan mode to help myself to understand the high level design model is about to execute though. It’s not for model but for myself. If not using plan mode then what?

[dai @yufanfei]

@Balder13946731 @grok fact check

[dai @yufanfei]

@theo @_mohansolo @joshwoodward @_mohansolo I genuinely think this will make users lose their trust to every Google product. And trust is hard to build.

[dai @yufanfei]

@Finding8964 @grok 是真的吗

[dai @yufanfei]

@bubbleboi @grok compare this with TPU v7 and v8

[dai @yufanfei]

@thsottiaux Please reset yourself to your profile pic

[dai @yufanfei]

@BoxMrChen 这是不是就是prime agent

[dai @yufanfei]

@thsottiaux Codex mobile on long context task can’t be synced. Often shows “Codex timed out to xxx”

[dai @yufanfei]

@realNyarime 可能和李佳琦fuzzy match上了 而李佳琦又和某件事有关 导致tokenizer层面相关的knowledge就被删掉了？

[dai @yufanfei]

@aleabitoreddit @PhotonCap IMO FN AXTI AEHR are all decent upstream beneficiaries from the 800G/1.6T transceiver play

[dai @yufanfei]

@_mohansolo @antigravity I’m pretty sure that there is a regression to Google Ultra plan users this week.. im often not able to send anything in 3-4 hour. Seeing aggressive 429 response from Google API. With this speed I can’t saturate even 40% of my Ultra quota.

[dai @yufanfei]

@LisaSu @elonmusk merry Christmas