Anshu

857 posts

Anshu

Anshu

@anshuc

CEO and token human at @DaysailAI. Formerly: YC founder, Apple

SF Katılım Haziran 2009
908 Takip Edilen465 Takipçiler
Sabitlenmiş Tweet
Anshu
Anshu@anshuc·
Super excited to share something I've been cooking for a while: @DaysailAI makes it extremely easy for small teams to scale themselves with teams of AI agents. Your AI team integrates with all your tools, and continuously improves. You can literally watch it build itself!
Daysail@DaysailAI

Introducing Daysail: The operating system for the one-person unicorn. The most innovative companies of tomorrow won’t be built by massive organizations. They’ll be built by someone with a laptop, a vision, and a team of AI employees. Daysail is where they’ll build that team.

English
4
2
16
6.1K
Anshu
Anshu@anshuc·
@kimmonismus Where’d you get the info that it’s Q3? Have not seen that anywhere else
English
1
0
12
1.2K
Chubby♨️
Chubby♨️@kimmonismus·
Claude Mythos is scheduled for Q3 2026. We're currently still in Q1 2026. Anthropic will certainly factor in compute limitations and project how good the model will be. However, if it's already this good *now*, and NVIDIA's Vera Rubin GPUs aren't even live yet, what will we see by the end of the year? Or in 2027? I'm increasingly convinced that Dario is right and superintelligence will become a reality. But there's also a caveat: Anthropic is also planning its IPO for Q3 2026. I certainly see a connection there.
Chubby♨️ tweet media
Chubby♨️@kimmonismus

Confirmed by Fortune: Opus 5: so good that it poses a danger. It's a massive leap from version 4.6, and it's so advanced that it poses a huge security risk. They'll roll it out slowly because they're concerned about unforeseen consequences, especially security risks. Besides Anthropic, OpenAI has also made similar announcements. As The Information recently reported, their upcoming model, "Spud," is an equally massive leap that will greatly accelerate the economy. In short, 2026 will be the year everything changes. OpenAI has already renamed its department "AGI Deployment."

English
88
69
1K
190.2K
Anshu
Anshu@anshuc·
@karpathy It’d be cool if we trained models to respect an emphasis level, like “this block of the prompt should matter to you 0.25x as much as the rest of the prompt.” I’ve done this with prompting but I imagine it’d be a lot better if the model understood it natively
English
0
0
0
21
Andrej Karpathy
Andrej Karpathy@karpathy·
One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.
English
1.7K
1.1K
21K
2.6M
Anshu
Anshu@anshuc·
@ianzelbo I wasn’t even following you but I can reply! How cool am I
English
0
0
0
50
Ian Zelbo
Ian Zelbo@ianzelbo·
If you can reply you’re cool and i wanna know you
English
383
4
467
46.7K
gabriel
gabriel@gabriel1·
@levelsio Absolutely incredible post king 👑🔥 You’ve done it again 😭💯 this is the kind of content the world needs right now 🙏✨ genuinely changed my perspective and elevated my consciousness 🚀🧠 never stop posting legend 🫡💖
English
9
2
328
40.1K
@levelsio
@levelsio@levelsio·
Okay let's see who can reply to this
English
2.5K
17
2.2K
1M
Anshu
Anshu@anshuc·
@levelsio oh pieter? yeah, he and I connect on twitter, yeah, we’re in the same circles. not a big deal or anything. I can reply to his posts and stuff
English
0
0
0
94
Anshu
Anshu@anshuc·
had a ton of flaky e2e tests. replaced them with plain md files that explain how to test each feature CI runs my dev stack + chrome + codex CLI on each file - deleted 30k LOC - zero flakiness - took 20 min to set up - i can even watch it from my phone markdown is code now
Anshu tweet media
English
1
0
1
45
Anshu
Anshu@anshuc·
maybe gpt-5.4-mini is a good model
Anshu tweet media
English
0
0
0
45
Anshu
Anshu@anshuc·
I'm convinced Codex subagents are not respecting the model param. I tell GPT-5.4 to spawn Spark subagents to save my 5.4 quota, and it does, but when I look at their threads, they're just running 5.4. They're definitely not generating tokens at Spark speed.
Anshu tweet mediaAnshu tweet media
English
1
0
0
144
Anshu
Anshu@anshuc·
@thsottiaux 90% of the time when GPT-5.4 tries to spawn a Spark subagent, I'm seeing it get kicked back to GPT-5.4. Can you please investigate? Is this a visual bug or a real one that affects quota?
English
0
0
0
44
Tenobrus
Tenobrus@tenobrus·
@gdb @basedjensen bro it's a good model but why have you been reduced to elon-tweeting
English
8
1
199
6.4K
Anshu
Anshu@anshuc·
@chatgpt21 there’s always the tried and true Elon strat of just posting things and assuming your employees will figure it out
English
0
0
0
200
Chris
Chris@chatgpt21·
Elon is betting Xai will be so far ahead of Anthropic, OpenAI, & Google within the next 9 months that you will need a “James Webb telescope” to see who’s in second place Keep in mind the first two labs have entered takeoff and or limited RSI. I’m thinking Elons best way of catching up is on the compute side, but even then passing OpenAI and Anthropic seems like a far cry at this point. If he were to pull this off it would be a true marvel …
Elon Musk@elonmusk

@peterwildeford xAI will catch up this year and then exceed them all by such a long distance in 3 years that you will need the James Webb telescope to see who is in second place

English
64
10
326
43.9K
Anshu
Anshu@anshuc·
@Dimillian yep, i can trivially consume a pro sub's weekly quota in a day by casually doing my job as a software engineer but don't worry, the team thorougly investigated it and it's totally working as intended
English
0
0
1
497
Thomas Ricouard
Thomas Ricouard@Dimillian·
Two more weeks of living like that
Thomas Ricouard tweet media
English
15
2
173
13.3K
Anshu
Anshu@anshuc·
@jxnlco i'd love to try out codex this weekend but my pro subscription ran out of quota like 12 hours after tibo reset it
English
0
0
2
101

Keşfet

@kimmonismus @karpathy @ianzelbo @gabriel1 @levelsio @embirico @OpenAIDevs @thsottiaux