Bao Bui
1.6K posts
the whole "cannot use our models to develop a competing product" is getting really confusing now that model labs are launching regular products like Claude Design
Be Mark Zuckerberg: > learn Mandarin > give speeches in Mandarin in China > put Xi Jinping’s book on your desk > get employees to read Xi Jinping’s book > ask Xi Jinping to name your baby > get rejected > host Chinese internet officials at Facebook > try to bring Facebook back to China > get rejected > explore China-friendly censorship tools > get rejected > quietly test a China-only app > get rejected > buy an AI company with Chinese roots > get rejected by China again
LisanBench results for DeepSeek-V4 Both Pro and Flash score a bit higher than V3.2 but this is mostly due to them thinking longer There have been no noticeable gains in terms of reasoning efficiency The non-thinking version of V4 Pro scored almost exactly the same as V3.2 Due to the extremely low cost of they both lie on the cost frontier
ぶっちゃけ左翼さえどうにかすれば大体の問題解決できんだよな
Former Japanese television presenter and YouTuber Kanon Aoki has warned that X’s automatic translation tool is damaging Japan’s international reputation as a polite and harmonious society. She explains that the feature now exposes blunt or negative online comments from some Japanese users to global audiences, revealing opinions on topics like tourism and cultural norms that were previously hidden by language barriers. Via @FujiNews_
Meta’s strategy of making their crappy model so hard to use that no one realised it was crappy worked surprisingly well Goal was clearly to build the perception of momentum and it worked well in that regard
