Twitxy Saved

timtimtim

Verified account This account is verified. Learn more

@timtimtim_eth

London, England Joined November 2013
4.8K Following 8.3K Followers 9.7K Posts

Latest posts 5

Pinned post

timtimtim Verified account @timtimtim_eth
I wrote this article for two aims

- You should never forget what you dream of. I spent 10+ years finding the right timing. It is never too late to work on your dream.
- I wrote this by hand. I still love writing raw; it is a way for me to express my philosophy and my emotions. I know, many times I write shit grammar, but writing is still a fun activity to me.
timtimtim Verified account @timtimtim_eth Feb 24
View quoted post on X
44 replies 2 reposts 118 likes 17K views View on X
timtimtim Verified account @timtimtim_eth
Day 20 of 30.

I've written 19 of these posts and not one of them has a price in it.

That isn't an accident. I've been avoiding the pricing conversation the whole sprint. Users, cold emails, intros, a paid character build: all of it is easier to talk about than what @Creader_io costs.

Monday I have a call with a publisher. It's the only live shot at a paying customer before Nov 20. I still haven't decided what to show or what to ask for.

The plan so far is to skip the dashboard. I'd show one manuscript with one broken fact: the character who died in chapter 6 and is at dinner in chapter 22, flagged with the line where it breaks. The editor still makes the call. Creader just makes sure nobody has to catch it from a reader's email.

Then I ask for a number for a defined set of books, with a start date. Asking whether it would be useful only gets me a polite yes.

If I leave Monday with interest and no price on the table, the sprint ends where it started.

So the question is smaller again. What's the one number I'm willing to say out loud on Monday?

1,500 cold emails, about 10 replies. About 1,000 users, zero paying subscribers. 8 days left.
timtimtim Verified account @timtimtim_eth
这次很荣幸参加了上海西岸的艺博会, 也是人生第一次和艺术沾边并展出了作品. 有种艺术中最ai, ai中最艺术的感觉.

主题是表达的未来, 探索AI, 人类认同和表达

我们基于Kevin Nguyen做了一个实验性的尝试, 从meme去构建一个角色, 然后让他更加的具有人性. 感谢艺术家koa的指导. 让我和 @Shorpenleo 一起去完成这个作品.

在做这个作品的时候正值 @ryolu_ 老师在AIE做分享. 我很能共鸣ryo老师说的. 当技术不再是壁垒, 不在是blocker的时候, 我们能够很快的去做出. 那问题是否应该回归到怎么做, 为什么做.

我在最初设计的时候, 并没有去做类chatbot的思考, 我更多的从几个维度去思考了表达这件事. 1. 以人为本, 思考人类是怎么思考的, 2. 观察我们是怎么交流和沟通, 3. 人性究竟代表了什么. 基于这三点我们去量化然后数值化最后通过代码去实现. 目前在几个版本迭代之后, 整体体验好了很多. 引发了很多有意思的观察.

表达和对话的复杂, 我个人觉得很大程度在于对话中的个体很难感知且复现对话中的思考. 我们大部分时候对话的反应在0.2s, 所以对话中的思维是瞬间的. 但是当我们设计kevin的时候受到技术限制最低压到了0.8s, 同时字幕打字会提高观众的思考时间从而产生新的想法. 还有就是对话pattern也是一个很有趣的事, 人真的是很有趣的生物, 大家的对话pattern会有一个内核, 但是路径会随机应变. 这个是很难在设定+模型上去实现的. 最后我简化成了两种模式, 主动和被动. 我通过观察会发现, 大家在主动问问题这件事上会非常的排斥, 150+ 观众中, 只有差不多10-15%会倾向去主动问, 大部分可能更多的是想被问.

还有一个有趣的发现是, 大家对于ai的态度其实挺糟糕的. 我们对于人性的判断和实现是基于网络对于Kevin的刻板印象, 因此人格和本性会偏邪恶混沌. 他所说的所有内容不会超脱他的认知和性格, 也就是他是谁, 他知, 他不知, 他想知和他相信. 不过当观众得知他是ai之后反而不会沟通了, 会问一些奇怪的问题, “单身吗”, “家庭情况”, 当然kevin会义正严辞的拒绝回答. 我通常会反问观众, 如果你第一次见陌生人会问这些吗 哈哈.

最后就是我们通过知识图谱构建了kevin的大脑, 通过聊天他对于事物的认知也会提高, 我觉得真正的AGI, 不是全知, 而是受限的去学习认知同时具有理解因果关系的能力. 我们虽尚未能理解神经元是如何存储记忆且激活记忆的,但是理解因果, 控制逻辑, 逐渐提高认知是可以达成的

还是得感慨, 越研究,越感觉人, 很有趣.

最后感谢
@Alibaba_Qwen的asr和tts, 两个1.7b效果太好了
@deepseek_ai 的v4 flash 作为底层, 下次尝试本地部署
@MiniMax_AI 的H3做的视频太棒了

如果你来西岸看展欢迎过来和kevin 一聊
9 replies 1 reposts 26 likes 1.4K views View on X
timtimtim Verified account @timtimtim_eth
Day 19 of 30.

First money of the sprint came in this week. Cash, invoiced, paid.

It wasn't for @Creader_io.

My team built an interactive digital character based on the Kevin Nguyen meme. You talk to it, and people come away thinking they talked to a real person. It was a contracted build for a client. We shipped it, sent the invoice, got paid.

I'm happy about it. I'm also clear about what it isn't. It isn't a subscription. It doesn't tell me a single writer will pay to have a manuscript checked. Creader has about 1,000 registered users and still zero paying subscribers.

The goal for this sprint was a paying customer by Nov 20. Read loosely, I hit it early. Read honestly, I got paid for building something people believe. What I'm actually trying to sell checks whether a story holds up: whether the character who died in chapter 6 is really at dinner in chapter 22. Those two skills sit close together. They are not the same product, and only one of them is the business.

Monday I have a publisher call. That's the conversation that tests the real question.

So the question is smaller again. What does a publisher need to hear on Monday that a paid character build can't say for me?

1,500 cold emails, about 10 replies. One invoice paid. Still no subscribers. 11 days left.