DeepSeek跌落神坛？多个机构质疑其安全性，多个国家禁用

I love @realDonaldTrump as much as a straight man can love another man

国内楼市真的没有底部。未来想买房子，不需要很急，而且房价会越来越便宜。 汽车的价格会越来越便宜。现在国内燃油车拼命降价，燃油车未来会越来越便宜。新能源车虽然现在很火，但未来的降价空间很大。 汽车今后一定是白菜价，房子汽车今后是不值钱的东西。

The more I think about it, the more I think 2013-era morality is pretty much correct: * free speech good * starting companies and making good products good * monopolies bad, vendor lock-in bad * democracy good * greed bad * trying to achieve national security through oppressing people bad * cosmopolitan humanitarian values, caring about faraway people, etc good Things we've "learned" since then have been harmful more than helpful

🇨🇳95.5% of Chinese people are satisfied with their gov’t. 🇺🇸Only 38% of their American counterparts are. SOURCE: Harvard University ⚠️Please save and use ⬇️

This BYD EV ad is so funny.😂

None of this would be possible without President @realDonaldTrump

The risk of politician coins comes from the fact that they are such a perfect bribery vehicle. If a politician issues a coin, you do not even need to send *them* any coins to give them money. Instead, you just buy and hold the coin, and this increases the value of their holdings passively. Furthermore, there is deniability: holding the coin is, in terms of financial effect, a linear combination of donating to the issuer and gambling. Hence you can have the intention of doing the former but when challenged claim that you are doing the latter. You can even hold the coin privately, and show that you are holding it to whoever you need to show; you do not need any zero knowledge proofs, you just send a test transaction. This is all risky to democracy, for reasons very similar to what I wrote in , and elsewhere. TLDR: the economic arguments for why markets are so great for "regular" goods and services do not extend to "markets for political influence". I recommend politicians do not go down this path.

We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely. DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets. They are perhaps the first OSS project that shows major, sustained growth of an RL flywheel. Impact can be done by "ASI achieved internally" or mythical names like "Project Strawberry". Impact can also be done by simply dumping the raw algorithms and matplotlib learning curves. I'm reading the paper: > Purely driven by RL, no SFT at all ("cold start"). Reminiscent of AlphaZero - master Go, Shogi, and Chess from scratch, without imitating human grandmaster moves first. This is the most significant takeaway from the paper. > Use groundtruth rewards computed by hardcoded rules. Avoid any learned reward models that RL can easily hack against. > Thinking time of the model steadily increases as training proceeds - this is not pre-programmed, but an emergent property! > Emergence of self-reflection and exploration behaviors. > GRPO instead of PPO: it removes the critic net from PPO and uses the average reward of multiple samples instead. Simple method to reduce memory use. Note that GRPO was also invented by DeepSeek in Feb 2024 ... what a cracked team.