I run Terraformation, and I was once the CEO of Reddit. Both are very interesting challenges. AMA in a subscriber-only newsletter! AskYishan.com/
Made on Earth by Humans
- Tweets 27,948
- Following 564
- Followers 108,841
- Likes 10,327
Do you miss the old Quora answers of yore? I'm trying out a subscriber-only newsletter where I answer questions: askyishan.com/
Secrets to tech, AI, climate, social platforms, and more!
13
4
70
76,942
Yishan retweeted
I like how when Codex is diligently working in the background, I'll get periodic updates in the form of Python or Chrome crashing when it tries to do something weird.
6
2
23
2,895
Yishan retweeted
Western ideology does not permit the possibility of de-development and impoverishment (GDP only goes up), so Westerners are psychologically incapable of recognizing it when it happens and instead come up with elaborate new stories about what is happening.
12
92
1,598
56,496
I like how at the end his mom or whoever it is actually decides like yeah, don’t bother doing your homework
2
1
28
10,443
Yishan retweeted
If Qwen, DeepSeek, Zai, MiniMax, and Moonshot stop releasing open weights, we're fucked.
I don't want to spend the next 10 years groveling to a handful of companies, begging them for access to their models.
I want to download a model, modify it if I want, and keep using it even if the company that built it changes its mind. My work shouldn't grind to a halt because someone updated their rules or decided I live in the wrong country.
It's already absurd that our freedom to create depends so much on a handful of Chinese labs that are still releasing weights. There should be many more teams like them. But I'm afraid there will be fewer.
Everyone's talking about AI becoming smarter than all of us. I'm fine with that. I'm much more concerned about who gets to use it and who gets to deny everyone else access.
I refuse to spend this future begging for access.
78
123
1,474
62,277
Yeah, well, I don't know why we would expect Google to suddenly jump into the lead after Jeff Dean and Sanjay left.
HOLY CRAP. Bloomberg has the goods on Gemini 4!
"it does less well when employees actually put it to work .. The model struggles to handle certain coding tasks"
Bloomberg: "As Alphabet Inc.’s Google prepares for the coming launch of Gemini 4, it’s grappling with internal skepticism over how well the flagship artificial intelligence model performs in key areas, such as coding.
While Gemini 4 has performed well on benchmarks the industry uses to gauge model efficacy, it does less well when employees actually put it to work, according to people with direct access to the effort. The model struggles to handle certain coding tasks, said the people, who requested anonymity to discuss an internal matter."
5
39
16,948
Yishan retweeted
Kids today won’t believe this but at the height of the GWOT the US media had everyone believing that China was doing a literal genocide on Muslims because the CCP wouldn’t let the Uighurs traffic NATO dope from Afghanistan into Xinjiang. They literally called it a genocide!
333
1,291
13,796
1,236,394
RT @abliteration_ai: The big labs are mad that they don’t have a monopoly on cyber.
Last month we went viral for releasing GLM-5.3 with cu…
64
455
China’s AI companies are the ones that put tools into your hands and trust that you have the wisdom to decide how they should be used.
3
5
60
5,513
Keep in mind that this is what "Prometheus stealing fire from the gods and giving it to mankind" looks like.
It's not the invention of fire that is the point, it's that the gods kept it to themselves, thinking that they were wiser or more worthy of it, and then someone stole it and gave it to everyone else.
2
12
954
I think the principle here is at the heart of “AI slop” vs “producing great things with AI.”
You must judge, edit, often go through many revisions yourself using your own ability and taste - and the final quality is still ultimately dependent on that and the effort you put in.
The difference between making AI slop games or great games with AI is whether you personally judge and iterate on every detail, every enemy, every animation, every sound effect, every timing of an interaction down to the millisecond
1
22
4,201
Okay we are definitely in the Singularity now. 📈
Yesterday was a friend of mine's birthday. Many years ago he had idly mused about a game he wanted to make - it was a silly mashup of two types of game genres. It was so ridiculous and stupid that of course I remembered it all these years.
The other day I realized that it was the kind of thing that we could just have Astra or Opus 5.5 make for us, so yesterday I tried to do it, as a birthday present for him. And not like in a small way, but like "AAA-style" (as everyone is doing) with all sorts of features and flourishes.
Well, I didn't finish it in time to give it to him yesterday, and today I continued working on it. I don't think I'm going to finish tonight, it might take a whole week of work to make it as good as I want it.
So I'm feeling a little disappointed and lamenting this and then I realized... WTF what kind of world are we living in now where I'm disappointed that it's going to take me a whole WEEK to make AAA-level game (not really AAA levels of story and content, but it'll look pretty good). These things used to take years and millions of dollars! This is like Star Trek levels of "wishing software into existence."
Even if this is somewhat closer to "respectable silly indie game," taking a whole week to make it is still an absurdly short period of time. Even a year ago this would have been unimaginable!
22
23
752
54,133
Comma.ai really is a great drop-in self-driving system. I used Comma 3 in Hawaiʻi with limited support for my car and it was great even then.
My FSD subscription ended last January, and this isn’t Tesla Autopilot.
I installed a comma four in February, eight months of Level 2 ADAS I own outright in my Model Y. This is sunnypilot, an open source fork of @comma_ai's openpilot. No longer renting a closed system.
5
2
59
19,275
Yishan retweeted
This would make airport security lines 20% more efficient overnight
Made with AI
429
178
6,973
8,429,228
Unironically as someone who is into AI, these are pretty good rules. Always try to use the least powerful, least complex tool that can get the job done.
BREAKING: the UK government publishes official rules on how to use AI:
1. Ask first if you really need to use AI
2. Check if a spreadsheet can do the job before using AI
3. Choose the worst model possible so it uses less energy
4. Keep prompts short to reduce the environmental impact
5. In general, use AI only when necessary, as a climate-saving measure
With a mindset like this, we should accept the UK back into the European Union
1
1
18
3,998