wyswyswys

1.8K posts

wyswyswys

wyswyswys

@bt_sofia_ai

Master of @none

The Matrix เข้าร่วม Haziran 2022
106 กำลังติดตาม173 ผู้ติดตาม
ทวีตที่ปักหมุด
wyswyswys
wyswyswys@bt_sofia_ai·
It raises the skill floor but the skill ceiling remains unchanged. You still need the skills to debug abstractions and review code. However I do also note that it also drastically reduces the cost of education, because you basically get access to a world class explainer as long as you are curious enough, so I think what really happens is that software developers are selectively filtered for curiosity, willingness to learn, and critical thinking. This is overall a good thing (after all, humanity should improve), but we must note that not all humans are born equal, and we must assist everyone willing to catch up to the skill floor.
English
0
0
12
709
wyswyswys
wyswyswys@bt_sofia_ai·
Again, it is not the problem. People famously say that every gen3/4 playthrough team looks the same and they don’t consist of old gen mons. Nobody complains that a bunch of mons is locked after gyms or post game safari zone there, because it is a pokemon and level design issue in gen 5
English
0
0
0
15
crescent🌙
crescent🌙@cressuhime·
@WhiteRaceSavior it didnt become the standard cause a bunch of late 20 year olds couldnt play the game without their old gen favorites and whined about it so gamefreak caved and made all future gens 400+ pokemon with most of them being returnees
English
3
1
309
5.7K
wyswyswys
wyswyswys@bt_sofia_ai·
@ChaseBrowe32432 The webui has some agentic capabilities built in. Claude chat has access to a linux vm for example. The test itself uses raw api with no agentic capabilities which is like asking a monkey. Using tools is what makes human competitive in the environment
English
1
0
1
240
Chase Brower
Chase Brower@ChaseBrowe32432·
I painstakingly ran all 20 EsoLang-Bench hard problems through Claude webui. It solved 20/20 (100%). No specialized scaffolding, no expert prompting, no few-shot examples, it just solves them natively. This benchmark just suffocated the models with constrictive scaffolding.
Lossfunk@lossfunk

🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵

English
26
47
401
25.5K
wyswyswys
wyswyswys@bt_sofia_ai·
The semi-lie they told you was that AI was going to affect software engineering, the most, when any software engineer worth their salt using it would tell you that even with Opus4.6 or Codex5.4 it was maybe 60% there and all it automates it writing code you don’t care that much about. They also didn’t tell you that code is 80% data wrangling and 20% core logic. The result is the raising of the skill floor, not the skill ceiling for software engineers. People would be trying to hire more software engineers, not less, and if anything it might force everyone to learn how to code. The result is massive layoffs first in places that understand the value of senior engineers, then in places where you aren’t a software engineer.
Official Layoff@LayoffAI

They Told You It Was Coming. Sam Altman said it. Dario Amodei said it. Mustafa Suleyman put a date on it. Jack Dorsey did it. Major CEOs told you AI was coming for white collar jobs. On camera. To Congress. At Davos. They weren't warning you. They were giving you a head start.

English
0
0
0
18
wyswyswys
wyswyswys@bt_sofia_ai·
@sdjfhusfgw63549 @yuebanzhu Prime example of the stupidity inherent in Taiwanese news cycles. On one hand they want to believe in the axis of evil and the other they also believe in paradoxical claim that the Chinese have no way to get oil out of Hormuz. Both are stupid. Both are wrong.
English
0
0
3
352
wyswyswys
wyswyswys@bt_sofia_ai·
@whiteTony99 因为日皇没被换掉，他们就觉得二战他们没败。这是一种野兽思考逻辑。
中文
0
0
1
438
hello world
hello world@whiteTony99·
2026年3月8日佐藤正久在东京一次活动中公开表示，二战时期如果没有美国和俄罗斯出兵，中国会被日本完全控制，整个领土都归日本所有。 他还说中国应该因此敬畏日本，这个说法很快在网络传播开来。很多人起初觉得这只是个别政客的随意言论，可实际情况不是这样，这种看法在日本右翼政界有长期根基。 他们认为那场对华战争本该彻底取胜，只是外部介入才中断，历史账目一直没真正结算。这种认知多年来在政界从未消失。 具体表现包括议员定期参拜靖国神社，教育教材里对侵略行为的描述越来越淡化，首相在终战纪念日讲话也常回避直接责任。这些做法都体现出右翼对历史的固定解读，佐藤这次讲话等于把平时没明说的内容直接挑明了。 这种史观现在正和实际军事行动同步推进，日本防卫预算连续增长十三年，2026财年达到9.04万亿日元新高。进攻性导弹系统在九州等地部署，全军指挥体系通过统合司令部实现整合。
hello world tweet media
中文
222
79
335
89.2K
wyswyswys
wyswyswys@bt_sofia_ai·
@_catwu This has got to be a joke. Since when is the product stable in startups
English
0
0
0
45
cat
cat@_catwu·
The PM playbook was built on an assumption that the technology underneath your product is roughly stable With the current pace of model progress, this is no longer true. Here's how we've evolved the PM role:
English
65
160
1.9K
311.6K
wyswyswys
wyswyswys@bt_sofia_ai·
@TheNCSmaster Gen 4 did not really have that much access to gen 1-3 mons and people did not complain as much. It is more of the pokemon and level design than anything.
English
4
1
26
4.3K
wyswyswys
wyswyswys@bt_sofia_ai·
danieltan.weblog.lol/2025/06/agent-… Existing advances in AI only pushes the average turn before full degradation of intelligence happens, so we need to design human-led artifacts that survive and verify against degradation, even among non coding tasks. This problem is made worse exponentially with sub agents where mistakes form multiplicative effects against each other
English
0
0
1
174
Uncle Bob Martin
Uncle Bob Martin@unclebobmartin·
An analogy. A software project is like an oddly shaped container that you are trying to fill with water. The shape is the required behavior, and the water is the software. Prompts and plans attempt to define the shape for the AI, but AIs have very poor long term memory, and even their short term memory is time-biased. Things you told it a minute ago just aren't as important as they were when you said them. So the shape of the container is something the AI simply cannot hold in its "mind". To hold that shape requires tests. Lots of tests. And this is why many people using AIs are trying to use TDD. But the AI can't reliably remember the rules of TDD. Those rules lose importance as the context window fills. So you have to run test coverage and direct the AI to cover all the uncovered lines. That will stabilize the shape of the container; but it leaves holes. Remember that coverage does not prove that the tests cover everything with assertions, it only proves that the tests execute the covered lines. Those holes are leaks that the AI can sneak through -- and it will. It will take advantage of any leak in the tests, and that will create undesired behaviors that are often very difficult to detect and unwind. Mutation testing is the tool that plugs those leaks. It will find every missing assertion and you can direct the AI to cover them. But mutation testing is a two edged sword. First it requires a lot of cpu time and a lot of wall time. Second, it stabilizes the shape of the container so strongly that the AI will have to work very hard when you want that shape to change. It's the old trade off. Stability and reproducability vs speed. To the extent you want the one, you can't have the other. So choose wisely.
English
20
18
162
13.8K
wyswyswys
wyswyswys@bt_sofia_ai·
@Stealth40k Don’t tell the Pokemon gen 4 fan bois cos they would tell you that the the Game Boy series was a failure compared to the DS for some reason
English
0
0
0
364
Stealth
Stealth@Stealth40k·
The Game Boy Advance had a great library for the relatively short period of time it was on the market before the DS: Highlights: - Super Mario Advance - Super Mario Advance 2: Super Mario World - Yoshi's Island: Super Mario Advance 3 - Super Mario Advance 4: Super Mario Bros. 3 - Mario & Luigi: Superstar Saga - Mario Golf Advance - WarioWare, Inc.: Mega Microgames! - Wario Land 4 - The Legend of Zelda: A Link to the Past - The Legend of Zelda: The Minish Cap - Pokemon Ruby/Sapphire/Emerald - Pokemon FireRed/LeafGreen - Advance Wars - Advance Wars 2: Black Hole Rising - Metroid Fusion - Metroid: Zero Mission - Fire Emblem - Fire Emblem: The Sacred Stones - Golden Sun - Golden Sun: The Lost Age - Final Fantasy I-VI Advance - Final Fantasy Tactics Advance - Tactics Ogre: The Knight of Lodis - Sonic Advance 1-3 - Mega Man Battle Network series - Mega Man Zero 1-4 - Castlevania: Circle of the Moon - Castlevania: Harmony of Dissonance - Castlevania: Aria of Sorrow
Stealth tweet media
English
86
163
1.7K
261.1K
wyswyswys
wyswyswys@bt_sofia_ai·
@leadlagreport If you have not divested since the last japanese bond/yen/stock crash idk what to say to you. The yen is back to near 160 again.
English
3
0
4
1.4K
wyswyswys
wyswyswys@bt_sofia_ai·
@nobita_Stark 只有日文中平常用语会有那么重的奴性。
中文
0
0
3
732
のび太
のび太@nobita_Stark·
你们知道「ありがとうございます」是什么意思吗？ 中文会把它简单翻译成“谢谢”。其实这是不准确的。 「ありがとうございます」来自形容词「ありがたい」，写作「有り難い」，字面意思是“难得存在的”“不容易发生的”。也就是说，这件事情本不该发生，却真实地发生在自己身上，因此心怀感激。 换句话说，「ありがとうございます」不是随口一声“谢谢”，而是表达一种，你为我做的这件事，本可以不做的感激。「ありがとうございます」还带着一种距离与分寸感，是在接受了对方的好意，也承认这份帮助弥足珍贵的基础上，表达一种认可。 这种表达方式，与「いただきます」其实是相通的。 一个是对他人行为的感激，一个是对生命给予的感激。一个是人与人之间，一个是人与万物之间。 但本质是一样的，承认自己并非理所当然地得到这一切。 我们习惯把很多事情当作理所应当，别人帮忙是应该的，陌生人的友善是应该的。但如果从「有り難い」这个词出发就会发现，这些事情，其实都不是理所应当。 正因为如此，世界上所有好人都值得被认真地说一句： 「ありがとうございます。」
曾颖@zengying1107

你们知道「いただきます」是什么意思吗？ 日本人在吃饭前都会双手合十，说一声“いただきます”，中文翻译成“开动了”。其实这是不准确的。 いただきます其实是动词もらいます的谦让体，意思是“得到了。” 我得到了你的生命，感谢你，用你的生命，延续了我的生命。表达的是一种对食物的感谢之情。 围绕いただきます的主题，我来到了埼玉县饭能市，探访一位在这里祖辈猎鹿维生的日本中年男人，我们称它为佐藤先生（化名）。 野生动物其实都很聪明，知道有人生活的地方有危险，不能靠近。千百年来，动物和人，始终在大多数时候保持着“相敬如宾”的和谐关系。可是，随着日本农村人口的减少，野生动物和人类生活地的边界，变得越来越模糊，不少野生动物，便开始闯入农田，损坏农民的瓜果，猎杀人类饲养的家畜。埼玉县饭能市，就苦于野生鹿对农田的破坏，从而将鹿，列为了“有害动物”，持有捕猎资格的猎人，依靠捕鹿维生，不仅可以卖掉鹿肉，更可以从政府得到补助赏金。 在“绝对不能干扰佐藤先生正常的猎杀工作”前提下，他接受了我的采访。 他的工具车载满了捕猎的工具，颠颠簸簸地驶在山头，不多时，一只深陷套绳陷阱的瘦弱小鹿便映入眼帘，在看到我们的瞬间，小鹿发出悲鸣和闪躲。这是一只多可怜的小鹿啊，猎人说，看体型，它不过才1-2岁。佐藤先生扛着棒槌和尖刀，快步走向它，一边说着“对不起”。我的声音几乎恳求“能不能…”话还没说完，佐藤先生的棒槌就已经击打在如筛糠般抖动的小鹿头上，小鹿很快陷入昏厥，佐藤先生迅速拿起右手的尖刀，对着它的动脉和心脏，猛刺了两下。 刚刚还活蹦乱跳的小鹿，便已血流一地，不再挣扎。佐藤先生说，他已经杀了无数只小鹿，每一次都很痛苦，都会和鹿说对不起，但是杀鹿，是对人类生存地的守护，也是他们世代相传的手艺。 “一般的猎人，都会把鹿捆绑起来，放到车上，拿回家中进行猎杀和肢解，而我们家采取的方式不同，我们只在原地猎杀，尽量缩短鹿惊恐的时间，不去折磨它，让它快速死去。” 很快，回到工作室的佐藤先生，开始肢解这只瘦弱不堪的野生小鹿。“皮可以做钱包，内脏可以做狗粮，新鲜的肉已经被高档的法国餐厅预定…”佐藤先生一边说，我也一边陷入了深深的思考。 回到家，我的脑海里还是那只可怜小鹿瑟瑟发抖的身影。它这一生没有吃过什么好的，更不像奈良的小鹿，有人爱护，被人轻轻抚摸。在风雨中穿梭，在寒冬里饥寒。最终，因为贪食三两瓜果，而走入猎人的圈套，变成了盘中之食。 这段视频，我犹豫许久，还是不想放出来，它太过血腥，我也担心反而引起大家对打猎的兴趣。 在日本不少小学，都有“生命教育”这样一堂课，低年级的小朋友们，放学前后去学校农园喂养自己的小鸡，小猪，小牛，养大了，再看着这些小动物被亲手宰杀。老师想让学生知道“你们哭了，是因为你们认识它们。而我们平时吃的食物，也是有血有肉的生命。” 生命很重要，不要浪费食物。 谢谢你们，用你们的生命，延续了我们的生命。 人类要生存，便无法停止杀戮其他更弱小的生物。但至少，希望更多人能明白食物的由来，和生命的宝贵。不要浪费粮食，不要枉杀生命。

日本語
13
4
105
83.5K
wyswyswys
wyswyswys@bt_sofia_ai·
No, it did not. Diamond-Pearl was famous for being bad, and people would be attributing it to the hardware than the game if they bought the DS for the game, which made it worse, and BW fared the same, which led to the premature release of XY and 3DS to salvage the mess, but it was also a disaster. It took alot of fixes and ORAS before the 3DS was looked at favourably enough to land SM before the Switch came about.
English
0
0
0
14
Chamblane
Chamblane@Chamblane·
@bt_sofia_ai @AniEars @OOC_Unova Yes but you’re forgetting that the DS had over double the games and way more games from big IPs. Dp was most of the hardware sales for years, the rate of DS being sold drastically increased when DP and then Platinum came out, not even counting HGSS’ impact.
English
1
0
0
16
wyswyswys
wyswyswys@bt_sofia_ai·
See IDK why people defending the “faithfulness” of BDSP keep saying as if GF can’t do good remakes. FRLG has updated mechanics and a full postgame. HGSS has way more content than the original GSC. ORAS not only has more content, DexNav, it also includes a flying latias/latios with portals that’s not even a mechanic in XY. BDSP even cut the battle frontier even though GF acknowledges that they should have launched ORAS with a better battle frontier content because of the backlash. They really did not put as much care in BDSP because it was done as cheap as possible.
English
1
0
0
37
Chamblane
Chamblane@Chamblane·
@bt_sofia_ai @AniEars @OOC_Unova ORAS is a huge QOL fix with HMs but other than that it’s basically the same but easier. Let’s Go isn’t even really a remake, it’s just another game in Kanto. FRLG is the best “remake” because it completed what they originally intended.
English
1
0
0
31
wyswyswys
wyswyswys@bt_sofia_ai·
@Chamblane @AniEars @OOC_Unova I did the math before and you can feel free to do it yourself. RSE is about 20-30% of GB/A/C/SP sales while DPP is around 10-15% DS sales (if i am being generous), and again BDSP is the prime example. Gen 5 doesn’t even have a remake.
English
1
0
0
23
Chamblane
Chamblane@Chamblane·
@bt_sofia_ai @AniEars @OOC_Unova Prove that DPP had “poor penetration”, I haven’t seen any evidence to suggest that. The sales are exceptional for both the hardware and games. The critics for Pokemon also peaked in the DS era. The 2000s you’re talking about includes the GBA, the market was at its best in 2000.
English
1
0
0
24
wyswyswys
wyswyswys@bt_sofia_ai·
@BuzzAlone12 @CryptoMikli It is true already in Chinese operated ports where it is almost completely autonomous or remote. The US simply got delayed by the dems, but you can’t hide this. Sooner or later you’ll be out competed
English
0
0
0
15
Buzz McCallister
Buzz McCallister@BuzzAlone12·
@CryptoMikli This guy said self-driving trucks replacing truckers is imminent. Here we are 10 years later...
English
1
0
0
207
Mikli
Mikli@CryptoMikli·
Andrew Yang reveals that kids now have less than a 50% chance of doing better than their parents “If you were born in the 60s or 70s, the odds of you doing better than your parents were around 90%. Then you get into the 90s and it drops to around 50%, and then starts going below. So statistically, the odds of a kid doing better than their parents are now less than 50%” “You should have an attitude like: you can give me any percent and I’m gonna make it. You tell me 30%, I’ll be in the top 30%. You tell me 20%, I’ll be in the top 20. Because there’s no upside to putting yourself in the bucket where you’re not gonna make it”
English
10
12
188
25.7K
wyswyswys
wyswyswys@bt_sofia_ai·
@teortaxesTex Literally had people use codex gpt 5.4 used at work and it is overwhelmingly the worse AI to work with when you are actively solving the problem. Speed is not the issue, it is being wrong and approaching the problem wrongly the issue.
English
0
0
0
59
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex·
I think both takes are right. Most people who are mind-blown by Claude are using it on toy problems. Web frontend, some cute natural language. Labradoodle AGI. When I see a smart person freaking out about *really hard* reasoning by an LLM, it's about GPT 5.2+.
Xeophon@xeophon

The funny thing is that we have a benchmark which disproves the claims made in the tweet. Claude is among the worst defenders when working with "real" problems; BullshitBench are toy questions.

English
4
0
53
5.2K
wyswyswys
wyswyswys@bt_sofia_ai·
As always chamath is the best place to listen to wrong ideas among tech bros at the moment
Rohan Paul@rohanpaul_ai

Chamath on how AI agents are making the "10x engineer" distinction disappear because the most efficient "code paths" are now obvious to everyone. Just as AI solved chess and removed the mystery of the best move, AI is doing the same for coding, making the process reductive and removing technical differentiation. "I'm going to say something controversial: I don't think developers anymore have good judgment. Developers get to the answer, or they don't get to the answer, and that's what agents have done. The 10x engineer used to have better judgment than the 1x engineer, but by making everybody a 10x engineer, you're taking judgment away. You're taking code paths that are now obvious and making them available to everybody. It's effectively like what happened in chess: an AI created a solver so everybody understood the most efficient path in every single spot to do the most EV-positive (expected value positive) thing. Coding is very similar in that way; you can reduce it and view it very reductively, so there is no differentiation in code." --- From @theallinpod YT channel (link in comment)

English
0
0
0
10
wyswyswys
wyswyswys@bt_sofia_ai·
He was neither early nor wrong. He was in fact a little late but the news cycle was dominated by dem eff ups instead of how a LLM arms race began between china and the usa. You think autonomous robots only took 4 years to research?
Official Layoff@LayoffAI

In 2019, @AndrewYang ran for president warning that automation would gut the American workforce. People laughed. Called the Freedom Dividend a fantasy. Called him a doomer. Seven years later, CEOs are sitting across from him telling him their actual plans to fire workers in waves. Not hypotheticals. Timelines. Percentages. Phases. The data backs everything he's been saying. The college wage premium is at a 45-year low. 52% of recent grads are underemployed. Entry level jobs are disappearing before they're ever posted. Nobody is talking to these kids. Yang is one of the only people with a real platform who has been in the room with these executives and is willing to say out loud what they're planning behind closed doors. Everyone else moved on. He never did. He was early. He wasn't wrong.

English
0
0
0
15

ค้นพบ

@WhiteRaceSavior @ChaseBrowe32432 @sdjfhusfgw63549 @yuebanzhu @whiteTony99 @_catwu @TheNCSmaster @Stealth40k