AI, Web, Product Building https://xitter.cf/t.co/TceCyFABpN https://xitter.cf/t.co/3osnnCsGif | https://xitter.cf/t.co/XjHhNehUkV | https://xitter.cf/t.co/ukfhaVjL7I
Joined December 2023
- Tweets6.3K
- Following21
- Followers220
- Likes12.5K
Pinned Tweet
Building Huntneeds, a place where you can buy validated pain points, so you don't have to hunt for them yourselves and can directly start to build something people really need.
Join the waitlist -> huntneeds.com/
This is really astonishing. I just felt like watching two real men talking with each others.
DeepSeek is printing money like crazy. According to an article posted by DeepSeek's official account on Zhihu, they're now earning $38 per GPU per hour 🤯:
Currently, the service throughput of a single cluster reaches 155 trillion tokens per day, with an overall revenue of $38 per GPU per hour (at off-peak rates).
I see some people here claiming that DeepSeek-V4.1-Flash-1017 distilled GPT or Claude, citing result from the ModelTrace project as evidence. So I generated some random numbers with JavaScript, guess the results: JavaScript is distilling Claude-Opus-5.5 lol.
I'm not here to defend DeepSeek against those people. Personally I don't know whether DeepSeek distilled anyone, but isn't it a bit ridiculous for those people to claim that based on this evidence?
And if you check the readme of ModelTrace carefully, they've already talked about it:
Note that this project is an attribution tool: it only attempts, as far as possible, to determine which model in the fingerprint database the model under test belongs to. If the model under test is not in the fingerprint database, any result is possible.
Actually, I think it should be named DeepSeek-V4.2-Flash, at least.
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
Preparing stocks for HuntNeeds is so hard. Every time my research system finds viable pain points, I go check them and think about whether they're really viable myself, and I find most are false positives, so I just throw them away. Still collecting pain points that buyers will really think are worth paying for.
As soon as there're enough stocks, HuntNeeds will be online. Still uncertain when it will be.
Left my DeepSeek agent alone to do a research and then went to the toilet. It generated 32 subagents and burnt countless money in a few minutes. When I came back, my balance had turned to negative lol.
Before Thu, 08 Oct 2026 01:00:00 GMT, DeepSeek is full off-peak because of National Day. This is a really long time. Keep building everyone.
Instead of directly generating videos using AI, I think the best way is the combination of real shooting and AI transformation, which makes realistic frames and cuts costs a lot.
With the power of AI, we no longer need to spend so much money building scenes that will only be used during the shooting of a movie.
I slightly adjusted my landing page.
I don't think it looks really good. However, since I wasn't a professional designer, it's still okay to me lol.
Playing around with Jev. I saw someone successfully letting Jev pick the next words, so I was curious if it could pick the next character. I failed, lol, and found Jev really likes spaces.
A good stack for a waitlist page is Next.js + Turso, which I've used for HuntNeeds' waitlist. Both of them have sufficient free quotas. You can create 100 databases inside Turso. So just link to a Turso database from the Next.js server side and everything works with any costs.
xitter.cf/adyingdeath/status/209…
Building Huntneeds, a place where you can buy validated pain points, so you don't have to hunt for them yourselves and can directly start to build something people really need.
Join the waitlist -> huntneeds.com/
Opened my X timeline and instantly got 4 fruit flies in a row. What's wrong with X lately? I'd admit most of them are great works, but I've totally seen enough of them.
