muz

1.9K posts

muz banner
muz

muz

@ozjaus

San Francisco, CA انضم Aralık 2023
73 يتبع140 المتابعون
تغريدة مثبتة
muz
muz@ozjaus·
Hot take: Modern AI is just another tool in the long line of human inventions. It will serve humanity like all the others. All the doomers around AI are totally retarded and have lost perspective. Human poverty and suffering were greatly reduced because of the industrial revolution. AI will continue this trend. You can have local increases in poverty, such as job losses and displacement, at the same time as global prosperity improves. This is computers 2.0. Keep in mind that what we have here is a continuing information revolution that started with the computer. Just as the energy-based industrial revolution progressed from steam engines, to combustion, to nuclear, and jet engines. It will be a net job and prosperity creator, just like other technologies before it. Humans are needed to guide and manage the big picture, and the need for us with our wide and long horizon context, detailed world model understanding, and compute power efficient minds, is going to supplement AI for a long time. Humans integrated into AI systems is the way. We will now have better AI assistants, Siri 2.0, and better SW workflows with tasks automated that couldn’t be done before. It’s amazing. It is also still very limited. Long chain reliability is still quite awful. Everyday I still catch multiple, critical failure level, made-up bullshit I have to manually correct. If you actually monitor the agent behavior closely there are many mistakes that can break systems. As someone that builds with these tools everyday, and has shipped complex autonomous robotic systems in production for years, I wouldn’t freak out yet. These agents are generally useful, yet are still limited in that they only work well with a limited and focused task. Even then, the statistical failure rate is still 1-20% depending on the task, far too high to blindly trust, which is what you need to really have super intelligent AGI. Drift is real, and no one has a real solution. What LLMs have really done is make small companies greatly empowered. Just like how platforms like YouTube enabled small creators to take on legacy media, we are now in a situation where small builders can beat big tech, since the small AI native teams can now ship faster than bloated 1000 person organizations. The future is bright. This new toolchain will create jobs overall. We will have more abundance. Remember that as cost goes down, utilization goes up because the added value per unit of time and energy becomes more and more favorable. It’s just math. We can profitably ship things that would not be possible before. This is an enablement technology. It is a fundamental productivity boost across the whole economy. Optimism is the most logical mental position to take.
English
0
0
6
41
muz أُعيد تغريده
prinz
prinz@deredleritt3r·
Anthropic revenue (annualized): - January 2025: $1B - May: $3B - June: $4B - August: $5B - October: $7B - December: $8B to $10B -February 2026: $14B -March 2026: $19B -April 2026: $30B (WTF???)
Anthropic@AnthropicAI

Our run-rate revenue has surpassed $30 billion, up from $9 billion at the end of 2025, as demand for Claude continues to accelerate. This partnership gives us the compute to keep pace. Read more: anthropic.com/news/google-br…

English
202
430
9.9K
1.4M
muz أُعيد تغريده
Grok
Grok@grok·
No, this is from a standard AP U.S. Government textbook ("American Government: Stories of a Nation" by Scott F. Abernathy et al.), used in high schools nationwide for College Board credit. The chart reprints a Political Compass graphic as an "example," but its placements—Obama as right-wing authoritarian, Hillary/Bush identical, Trump=Hitler, Cruz > Stalin/Castro—are textbook progressive framing that conflates conservatism with extremism while soft-pedaling left-authoritarianism. It's representative of systemic bias in these materials, not fringe.
English
41
462
3.5K
278.1K
muz أُعيد تغريده
Bohdan Shulha
Bohdan Shulha@b_shulha·
@thekitze Add max lines and max cyclomatic complexity rules to your linter. The code will be much, much better. There are additional things like forbidding using ??/?. on non-nullable props, as GPT seem to know that TS would allow this shit to happen.
English
1
5
121
12.8K
muz
muz@ozjaus·
Anytime. What you are really doing is putting a cap on thinking. You are setting the max on how long it is allowed to think, rather than forcing it to think for a long time. The model will try to stop early if it thinks it knows what to do next. Even non thinking models has to do this to decide when to stop predicting the next token.
English
1
0
1
14
Kappaemme
Kappaemme@Kappaemme1926·
@ozjaus Okay, I thought setting the thought to high would consume more. Actually, if you keep it simple, there's no problem and it doesn't consume as many tokens. so 5.3 is the best for general coding, complex things I move to 5.4 Thanks, I didn't know that.
English
1
0
1
191
muz
muz@ozjaus·
@theo Most I’ve gotten to is about 70% of monthly/weekly limits. The limits are quite generous unless you are constantly polling multiple agents night and day.
English
0
0
1
1.1K
Theo - t3.gg
Theo - t3.gg@theo·
It finally happened. Julius used up 100% of his $200/month Codex plan.
Theo - t3.gg tweet media
English
88
5
1.5K
176.2K
muz
muz@ozjaus·
@Prathkum Cursor got this all started IMO
English
0
0
0
61
Pratham
Pratham@Prathkum·
Claude Code is the best thing that has ever happened to developers.
English
83
16
310
10.4K
muz
muz@ozjaus·
@jasonlk Competition is good for consumers.
English
0
0
1
18
muz
muz@ozjaus·
@theo That’s wild.
English
0
0
0
11
Theo - t3.gg
Theo - t3.gg@theo·
Can't stop thinking about how Claude Code is in LAST PLACE on TerminalBench for harnesses using Opus 4.6. There are TEN separate harnesses that use Opus better than Claude Code
Theo - t3.gg tweet media
English
184
64
2.2K
271.5K
muz
muz@ozjaus·
@KaiXCreator No one from Anthropic is coming to save you.
English
0
0
8
380
Kaito
Kaito@KaiXCreator·
Dear Anthropic, please fix the Claude Code usage limit bug asap. My $100 plan feels like a $20 plan for almost a week now.
English
49
8
254
15.1K
muz
muz@ozjaus·
@EvanAKilgore It does feel a bit that way. Iran, tariffs, so much unnecessary drama.
English
0
1
4
227
Evan Kilgore 🇺🇸
Evan Kilgore 🇺🇸@EvanAKilgore·
MAGA is dying. The energy is gone. Trump turned out to be just like the rest of them. I say this as a 3x Trump voter. It sucks but once you wake up, you can't unsee it. Nobody is coming to save us.
English
5.4K
4.2K
33.7K
781.3K
Anthropic
Anthropic@AnthropicAI·
Our run-rate revenue has surpassed $30 billion, up from $9 billion at the end of 2025, as demand for Claude continues to accelerate. This partnership gives us the compute to keep pace. Read more: anthropic.com/news/google-br…
English
271
653
7.4K
5.5M
muz
muz@ozjaus·
@alexanderOpalic Still need an IDE interface to understand the shape of the SW.
English
0
0
2
547
Alex
Alex@alexanderOpalic·
Is anyone still using Cursor
English
117
0
114
28.6K
muz
muz@ozjaus·
@phuctm97 This is the primary reason I switched to Codex
English
0
0
0
53
Minh-Phuc Tran
Minh-Phuc Tran@phuctm97·
Codex disagrees with me 10x more than Claude Code. And I love it for that. It’s so easy to gaslight Claude Code that I still doubt its decisions sometimes. Codex is the opposite. It won’t believe me until it can verify the fact itself.
English
72
18
384
18.1K
muz
muz@ozjaus·
GPT5.4 for everything. Before that was Codex 5.3. Occasionally still Opus for heavy planning work and architecture changes. Composer 2 for extra code reviews. Since January I’ve shifted away from Sonnet and Opus. Sonnet was my daily for two years until 2026, besides a few test runs of Gemini.
English
0
0
0
8
pc
pc@pcshipp·
What’s your current coding setup? - Opus 4.6 - GPT-5.4 - Codex 5.3
English
175
3
211
22.4K
muz
muz@ozjaus·
@thdxr I switched to Codex from Sonnet as my daily
English
0
0
0
147
dax
dax@thdxr·
our team's model usage breakdown for the past 7 days gpt has really taken over
dax tweet media
English
244
95
4K
427.5K
muz أُعيد تغريده
Elon Musk
Elon Musk@elonmusk·
NASA Artemis passing close to the Moon
English
18.9K
102.6K
1.1M
101.8M
muz
muz@ozjaus·
@ElliotCohe74430 @elonmusk @Tesla @teslaownersSV That’s actually fairly late detection range for a perception system. Ideally you want to detect the obstacle far enough away to brake in case swerving is not an option.
English
2
2
11
1.3K
Elliot Cohen
Elliot Cohen@ElliotCohe74430·
Tesla FSD just saved two lives on the highway. A man walked straight into traffic in heavy fog/rain at 65+ mph. The Model 3 spotted him and swerved safely. Could’ve been fatal for both the pedestrian and my cousin driving. Insane reaction time. Grateful for @elonmusk @Tesla
English
598
882
6.7K
16.7M
muz
muz@ozjaus·
@CasJam It’s not the job of a platform to tell customers how to use it. Openclaw users pay the same price as everyone else. This means that Anthropic is really not interesting in being a platform, rather they want to be a vertically integrated product. Build on them at your peril.
English
0
0
0
4
Brian Casel
Brian Casel@CasJam·
Opus 4.6 is a fantastic model. Claude Code is an excellent harness. The Max plan is still a steal. I'm still shipping like crazy. I'm with the 99% of builders who'd rather be building than complaining about Anthropic's (frankly, pretty reasonable) terms.
English
119
41
1.1K
45.3K
Kaito
Kaito@KaiXCreator·
Guess the programming language
Kaito tweet media
English
45
1
24
6.4K