Richard Zhang

More than a magician.

[Richard Zhang @commentban]

@MaxForAI @tann_kyuu 108A12B

[Richard Zhang @commentban]

@tualatrix 10刀一个月，这么贵？

[Richard Zhang @commentban]

@haas39483312 首先感谢大哥一直分享这么多。但是我想补充几点：
1. 我是业内人士，所谓世界模型的数据蒸馏和文本模型差别不大，总有一万种蒸馏方式，所以未来跟美国顶尖模型也只会是三个月差距
2. 反而所有的物理部件基本上都是我们掐老美脖子
3. 芯片是唯一比较落后有代差的

[Richard Zhang @commentban]

@haas39483312 @realPureNomad 1. 龙虾带来的个人token需求正在疯狂扩大
2. 会有很多中转公司套壳中国模型
3. 谁说我们不能用中国电力卖美国的开源模型（还可以拿到美国数据）

[Richard Zhang @commentban]

@emollick Wait for AI agents that can make tools themselves in a chatbot session

[Richard Zhang @commentban]

@AnnaRMatson @threadreaderapp unroll

[Richard Zhang @commentban]

not exactly, partly because of the training data, training data are highly filtered and selective and even have many QA data in the pretraining stage. Novels may be largely filtered out cause they are not useful to improve benchmarks.

[Riley Goodside @goodside]

It’s crazy in hindsight that the stylistic mimicry of the GPT-2/3 era was such a local maximum for AI-written texts as cultural artifacts. For all their gains in making sense, aligned LLMs still can’t write with the eerie naturalism GPT-3 exhibited four years ago:

[Richard Zhang @commentban]

@VictorTaelin @karpathy The most basic problem is hallucinations. LLMs do not know what/how much they know. Only if you know there's a problem, you find solution to it.

[Richard Zhang @commentban]

@luoleiorg 用红薯粉擦一擦就行了

[Richard Zhang @commentban]

@svpino No, groovy is the best. No any other can even see its rear lights.

[Richard Zhang @commentban]

@haas39483312 @LuoSeaking 找不到人可以投，是因为我们所有的投资都是不能有风险的。只许成功不许失败。那还创新个屁呀。

[Richard Zhang @commentban]

@RetirementUnity @elonmusk If one of them burns all of them burn.

[Richard Zhang @commentban]

@haas39483312 历史会重复，也不足为奇了。

[Richard Zhang @commentban]

@chenshaoju @haas39483312 谢谢，我研究下。

[Richard Zhang @commentban]

@chenshaoju @haas39483312 请教一下估值怎么看？市盈率？还是有什么评估的方法？

[Richard Zhang @commentban]

@haas39483312 这个确实是拼芯片

[Richard Zhang @commentban]

@marktenenholtz I think you are right, just as what they do for 3.5

[Richard Zhang @commentban]

@wsgelu @haas39483312 成功几率很小很小

[Richard Zhang @commentban]

@haas39483312 但是GPT完全不同于这些人工智障——从根本上不同

[Richard Zhang @commentban]

@haas39483312 你关掉这些人工智障是因为1. 他们蠢 2.跟他们交互太费劲 3. 没啥用