Pinned Tweet
This question has been troubling me for a while. I a world where most code is written by an agent and PRs are created by an agent at light speed, how can engineers maintain an understanding of the system in order to plan, organize and plan the future of the project?
1
0
4
292
Paulo Matos retweeted
Our second product is called Blindsight and it will, as the name suggests, enable even those who are completely blind or have never seen to see.
It is ready from an engineering standpoint. Just waiting on regulatory approval.
It is ready from an engineering standpoint. Just waiting on regulatory approval.
Neuralink’s first product is called Telepathy.
It enables high-speed communication with computers or humans simply by thinking. Output bandwidth will vastly exceed the fastest human speech or typing in a few years. Read that sentence again.
4,198
12,397
101,651
6,232,154
If you use rightkey.app you can now talk to your agent about your piano practice and get recommendations for improvement and much more. #mcp #piano #practice
0
0
0
19
Paulo Matos retweeted
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys 9. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys 9. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
2,084
2,750
34,087
6,890,290
This:
A journalist just asked me: “What remains for humans to do?” I think it’s becoming increasingly clear that more and more problem solving will be automated. But humans will remain crucial in determining the direction of discovery. Deciding which problems are worth solving.
For maths and for Software Engineering...
A journalist just asked me: “What remains for humans to do?” I think it’s becoming increasingly clear that more and more problem solving will be automated. But humans will remain crucial in determining the direction of discovery. Deciding which problems are worth solving.
For maths and for Software Engineering...
This is honestly shocking.
Three months ago, Hong Wang was awarded the Fields Medal for resolving the Kakeya conjecture in three dimensions.
Today, OpenAI released a proposed proof of the conjecture in ALL dimensions.
And that’s not all. It also announced a proposed existence odynamics and a proposed refutation of the Exponential Time Hypothesis.
Experts say that, if confirmed, these would be Nobel prize and Turing award level discoveries respectively.
A journalist just asked me: “What remains for humans to do?”
I think it’s becoming increasingly clear that more and more problem solving will be automated.
But humans will remain crucial in determining the direction of discovery. Deciding which problems are worth solving.
Because the value of a discovery depends on our needs, our priorities and the kind of world we want to build.
There will be a shift from problem solving to problem setting.
I also think there will be huge individual differences.
People with strong critical thinking skills, creativity and imagination will be supercharged by AI.
People without those skills risk surrendering to it.
I’m particularly concerned for children, who may never develop these skills.
I think epistemic literacy - how to distinguish information from knowledge - is the single most important skill we should be teaching.
0
0
0
25
Sometimes it feels claude should be more ambitious... starting the week with - as part of a larger prompt:
> 2. we run everything - if there's more surface, we create more issues under the epic. Time is on our side. Don't be humble. We can do anything. Impossible is nothing.
> 2. we run everything - if there's more surface, we create more issues under the epic. Time is on our side. Don't be humble. We can do anything. Impossible is nothing.
0
0
0
44
Paulo Matos retweeted
@mattpocockuk Check out mojolang.org Matt, it does that, and has advanced type system (eg linear types, symbolic constraints etc). I'd add "can target all hardware" to your wishlist. Mojo has you covered.
Let me know if you'd ever like to chat about it someday.
Let me know if you'd ever like to chat about it someday.
9
12
270
12,343
Paulo Matos retweeted
How on earth did this happen:
The Indian certificate authority (CA) eMudhra had one job (well, two):
1. Create secure, 384-bit private keys
2. Store these securely and do not leak them
Well, eMudhra chose private keys between 1 and 256 (the numbers!!) and called it a day. WTF
The Indian certificate authority (CA) eMudhra had one job (well, two):
1. Create secure, 384-bit private keys
2. Store these securely and do not leak them
Well, eMudhra chose private keys between 1 and 256 (the numbers!!) and called it a day. WTF
I don't really understand why this would happen. It seems higher-effort than standard key generation.
224
644
7,496
477,366
Paulo Matos retweeted
As of 3 hours ago, the Linux Kernel no longer requires the usage of AI to be declared for new code submissions.
46
78
1,236
73,173
Paulo Matos retweeted
@KenneyNL This is bullshit, Kenney. I’m the same tech tree following optimist who moved from C to C++, from 2D to 3D, and from closed source to open source. I run a company. I write code. I contribute to published papers. I’ve upped my expectations of AI. Up yours!
868
384
17,914
3,315,462
Can LLMs compose music? Here's my experiment on this.
If you have suggestions, let me know. Have fun guessing if the piece is AI.
composer.ocmatos.com
#music #composition #llms
If you have suggestions, let me know. Have fun guessing if the piece is AI.
composer.ocmatos.com
#music #composition #llms
0
0
1
46
Paulo Matos retweeted
I have to admit, watched this a few times, especially to the end:) the creativity in the prompting and the model is awesome to see
Argon has this amazing natural understanding of the world. It can do so many things from first principles.
Here's a pelican riding a bike, using fourier epicycles.
Video made by Argon.
Time-synced Lyria 3.5 music prompted by Argon.
318
529
9,183
951,401
jsse is being hammered on performance over the past week - if you haven't tried it out yet, now is the time.
github.com/pmatos/jsse
#javascript #engine #performance
github.com/pmatos/jsse
#javascript #engine #performance
0
0
2
48
Since starting piano lessons I've logged every lesson and practice session. Instrumentive on Android served me well, but I wanted more, so I built my own: rightkey.app
It's been my daily driver for a year. If you track your practice, try it and tell me what you think!
#piano #musicpractice
It's been my daily driver for a year. If you track your practice, try it and tell me what you think!
#piano #musicpractice
0
0
0
39
Paulo Matos retweeted
Прочитал "The end of programming" Пола Дикса. Это CTO InfluxData, человек который сам базы данных пишет, не инфлюенсер. Тезис жесткий: писать код руками и ревьюить его глазами - вымирающее занятие. Софта будет больше чем когда-либо, но большую часть кода никто из людей не напишет и даже не прочитает.
Пруф у него - переписывание Bun 1.4: один человек, агенты на предрелизном Fable 5, 11 дней, 6778 коммитов, +1 млн строк, порядка $165k в токенах по API-ценам. И оно работает на миллионах машин прямо сейчас.
Я с ним согласен по существу, но с одной поправкой из своей практики. Он пишет что "организационная инерция" еще лет десять будет держать людей на ручном коде. Я эту инерцию потрогал руками: мне закрыли функционально правильный PR в популярный опенсорс, потому что в нем "слишком много комментариев". Не потому что баг. Потому что стиль, и потому что AI-assisted. Инерция - это не про то что люди не умеют. Это про то что ревью глазами уже не масштабируется, и его защищают как ритуал.
И вот главное у Дикса, что все пропустят за громким заголовком: ревью никуда не девается. Оно переезжает из "читать код" в "строить проверку". Харнессы, тесты, верификация, замеры. Я последний месяц ровно этим и живу: гоняю модели через свой тестбенч на лупы, натравливаю два аудитора на один код, меряю счетчик токенов вместо того чтобы верить бенчмаркам. Это и есть программирование, просто объект другой.
Так что не конец программирования. Конец программирования как набора текста)
pauldix.com/the-end-of-pro…
Пруф у него - переписывание Bun 1.4: один человек, агенты на предрелизном Fable 5, 11 дней, 6778 коммитов, +1 млн строк, порядка $165k в токенах по API-ценам. И оно работает на миллионах машин прямо сейчас.
Я с ним согласен по существу, но с одной поправкой из своей практики. Он пишет что "организационная инерция" еще лет десять будет держать людей на ручном коде. Я эту инерцию потрогал руками: мне закрыли функционально правильный PR в популярный опенсорс, потому что в нем "слишком много комментариев". Не потому что баг. Потому что стиль, и потому что AI-assisted. Инерция - это не про то что люди не умеют. Это про то что ревью глазами уже не масштабируется, и его защищают как ритуал.
И вот главное у Дикса, что все пропустят за громким заголовком: ревью никуда не девается. Оно переезжает из "читать код" в "строить проверку". Харнессы, тесты, верификация, замеры. Я последний месяц ровно этим и живу: гоняю модели через свой тестбенч на лупы, натравливаю два аудитора на один код, меряю счетчик токенов вместо того чтобы верить бенчмаркам. Это и есть программирование, просто объект другой.
Так что не конец программирования. Конец программирования как набора текста)
pauldix.com/the-end-of-pro…
46
120
1,023
152,364
Paulo Matos retweeted
We have a huge pivot underway here at Epic, moving developers around to retool for maximum AI acceleration.
The aim isn't to do the same stuff with fewer people, but to enable everyone here to build MORE and BETTER stuff, and to bring the benefits to all teams using Unreal.
The aim isn't to do the same stuff with fewer people, but to enable everyone here to build MORE and BETTER stuff, and to bring the benefits to all teams using Unreal.
5,516
914
16,735
6,442,228
Paulo Matos retweeted
"The culture on X today is remarkably friendly and inclusive. It is a cheerful place", I couldn't agree with @daringfireball more. X is the place to be if you genuinely like technology and progress. Just such a delightful optimism here. daringfireball.net/2026/10/x_once…
125
140
3,049
152,385
When you see the several body shutdowns pro triathletes experience in the last kms of an IM race you understand how brutal it is, but experiencing it is another thing altogether. It will certainly change your perspective on perseverance, endurance and motivation. I recommend!
Imagine you are leading the @IRONMANtri World Champs after taking the win last year and you completely shut down 1km from the finish line..
Utterly insane from Lovseth AND somehow she still managed to finish the race.
Ironman triathlon is BRUTAL.
#IMWC #kona #running
0
0
1
46
Paulo Matos retweeted
🗞️ if your team is too busy doing their 'normal job' to experiment with AI, you're preparing them to be replaced
ghuntley.com/replaced/
In this post, I guess this's the last time I'll say it. AI use is no longer optional if you wish to be employed. What follows is advice for iring managers on how to hire in 2026 to source AI-first candidates.
ghuntley.com/replaced/
In this post, I guess this's the last time I'll say it. AI use is no longer optional if you wish to be employed. What follows is advice for iring managers on how to hire in 2026 to source AI-first candidates.
121
340
2,880
805,235
