Dominik Lukes (@techczech) profile picture

Exploring schemas and propositions about language models of all kinds on https://xitter.twitterapi.xyz/t.co/GU07uzaA4F and on https://xitter.twitterapi.xyz/t.co/VDN6KwY9h8.

UK
Joined April 2009
Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)Photo by Dominik Lukes (@techczech)
Dominik Lukes retweeted
The bitter lesson in 26 words: Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
139
997
152
7,551
627,856
Dominik Lukes retweeted
Copy pasting between Claude and Codex is the 2026 equivalent of copying pasting from Stack Overflow
280
45
35
2,421
110,166
Dominik Lukes retweeted
I think literally the opposite will be true. Academic journals that impose stringent restrictions on AI use will simply stop attracting the best papers. The pursuit of truth and knowledge is not a sport.
My prediction: Academic presses and journals will hold prestige to the extent that they are known for stringent policies against using Generative AI in composing publications. Conversely, they will lose prestige to the extent they are caught publishing works composed by GenAI.
11
8
4
82
9,690
How good are the interviews? Trained sociologists rated AI-led interviews at 2.93/5—about the level of an average human expert under the same online text-chat constraints. AI is a complement to, not a substitute for, in-person interviews. 3/6
1
1
1
980
Dominik Lukes retweeted
Science is not some hobby that serves the self actualization of people in a publicly sponsored job placement program. It is the systematic search for truth.
Wow, over 1,000 mathematicians have now joined the Association for Human Mathematics (AHM), committing to: > NOT publish AI generated math papers > NOT work with frontier labs > 654 of them went even further to NOT even use AI in their research, at all
140
338
47
3,376
315,880
Replying to @GosuCoder
The meetings are rarely the real bottleneck. It's that nobody owns the decision, so every document becomes a way to avoid making it. In B2B integration work the fastest projects I've seen had one person who could say yes and a working prototype in the first week.
2
2
12
1,267
Dominik Lukes retweeted
Now is a good time to change passwords that you've reused or that have appeared in data breaches, get a password manager, turn on 2FA with an authenticator app, and turn on automatic updates. This weekend. Carve out some time and do it today.
17
47
10
876
66,822
Dominik Lukes retweeted
It's actually insane that there are papers coming out about AI capabilities evaluations using two-year-old models. We have to reduce the length of time to get studies published or academic publishing will have to become completely irrelevant for AI.
46
76
9
2,628
60,932
It would be fun to figure out when “too many books” post-Gutenberg turned into “not enough books!”
Replying to @tamarawinter
Nothing new under the sun, including laments about the death of the book:
1
4
11
3,166
Dominik Lukes retweeted
Every eval that is sufficiently diverse but still unsaturated by current frontier models is suspicious. People aren't looking at the transcripts nearly enough. tbench is commendable for how open & accessible their data is, so everyone can contribute to improving it.
2
5
1
110
7,515
Dominik Lukes retweeted
LLMs can seamlessly navigate any kind of language. You can say in Chinese whatever you can say in English. Your code skills are no longer constrained by your knowledge of C++. Your math skills are unbound from your knowledge of mathematical notation and definitions
8
8
118
4,909
Dominik Lukes retweeted
Tips for new teachers: 1. Go at 1/3 the pace you naturally want to. 2. Repetition repetition repetition. 3. When you ask for questions, allow an uncomfortable silence to give them time to respond. 4. Ask them many questions to see what they already understand.
9
29
4
484
17,803
Dominik Lukes retweeted
What used to pass as a 'good' academic book review can nowadays be done much better by AI. There's no need to employ a junior academic to say something nice about an elder, just to prove their own competence. Whatever age, you should review a book only to challenge it.
2
2
2
6
1,451
Dominik Lukes retweeted
I think people believe that OpenAI and Anthropic, since they are so good at AI, are also somehow not normal companies that suffer from the normal problems of hypergrowth, but they are not immune from the inevitable politics, disfunction and confusion that haunt fast-growing firms
103
24
11
686
35,138
Dominik Lukes retweeted
If you are into AI (you should be) and PhD students, join us for the Summer School in Prague! (deadline is the end of November) @ninabegus
1
2
4
182
Dominik Lukes retweeted
we should just scan all the books and burn them all. a book is just a hilariously inefficient .txt file
21
6
2
57
7,050
Dominik Lukes retweeted
I asked Opus and Suno to turn my article on the "China shock" narrative into a song. It's a brave new world.
6
24
18
121
16,999
Dominik Lukes retweeted
Replying to @techczech
Oh for sure, there are a lot of terrible professional developers. Right now I find they're the ones who say AI is useless at coding because they tried it once and it made a mistake
1
41
Dominik Lukes retweeted
I’ve said it before, I’ll say it again: If someone claims to be an AI expert but spends their time typing prompts into their computer, they aren’t an ai expert They are behind Cutting edge AI use has been hands off for about a year now and it gets easier every week
86
1
7
132
72,455
Dominik Lukes retweeted
Anyone still using an IDE now to write code by hand is doing it to be contrarian.
Replying to @signulll
You sound like @Steve_Yegge . Almost exactly 12 months ago he got up at @aiDotEngineer in front of of everyone and said if your still using an IDE in 2026 your a bad engineer. Granted he’s a contrarian and funny dude but I though that was nuts at the time. Crazy how things change @swyx
25
11
6
235
20,949