loops

417 posts

loops banner
loops

loops

@_smitop

human

Toronto, Canada Bergabung Aralık 2019
1.2K Mengikuti208 Pengikut
loops
loops@_smitop·
Me rn
loops tweet media
English
0
0
1
9
loops
loops@_smitop·
@she_llac @socialtranxiety my impression is that Hacker News is run pretty independently from the rest of YC, I think they give the mods (dang/tomhow) a lot of freedom to run the site as they wish
English
0
0
1
8
shellac
shellac@she_llac·
@socialtranxiety it is a startup accelerator and sort of a community thing for startup founders news.ycombinator.com is a website they run on the side which is very much not the main thing
English
2
0
9
261
ro computer ^o^
ro computer ^o^@rogesterone·
does anybody actually know what ycombinator is? like people will be like "oh it's reddit but techbro" but explain how a sentence like "delve removed from ycombinator" makes sense in that context
English
4
0
19
1.6K
loops
loops@_smitop·
huh, Claude's thinking summarizer will sometimes refuse, resulting in weird looking CoTs
loops tweet media
English
0
0
1
33
loops
loops@_smitop·
@austinc3301 I don't either to be clear, I just like making arguments and I had been thinking about whether or not Anthropic's anti-competitor stuff (Claude code ban, anti-distilliation) was good or not recently
English
0
0
0
90
Agus 🔸
Agus 🔸@austinc3301·
@_smitop That’s an argument I’m happy to engage with, but I’m fairly sure he does not have this in mind
English
1
0
8
172
Agus 🔸
Agus 🔸@austinc3301·
how is there any contradiction between these two things lol
ThePrimeagen@ThePrimeagen

I cannot stop thinking about Anthropic today for some reason 1. They claim that they are a company that prioritizes safety first and that they are creating a model responsibly 2. we learned from the code leak that anthropic employs deceptive techniques by calling fake tools to throw off distillers... Is this lying pattern built into claude or just the harness running claude? What else are they lying about? I am a bit more concerned now.

English
4
0
37
6.1K
loops
loops@_smitop·
@austinc3301 If they don't get distillers to switch but do manage to inject fake tools, the distillation data will include models lying (by making tool calls that don't actually exist), and the distilled models will learn "I am the kind of model that makes tool calls I know don't exist"
English
1
0
0
26
loops
loops@_smitop·
People kept telling me about their bad startup ideas and I couldn't tell if they were serious or this was some elaborate bit
English
0
0
1
34
loops
loops@_smitop·
I'm going to Inkhaven (inkhaven.blog) where I'm going to be writing every day in April! I'm going to be in Berkeley all month, lmk if anything interesting is happening or you want to meet
English
0
0
1
40
loops
loops@_smitop·
@vlelyavin @theo openai was doing the routing to a weaker model, not anthropic. ant has never silently routed to weaker models. and ant has never claimed to have "pure transparency"?
English
0
0
0
43
Vladimir
Vladimir@vlelyavin·
@theo the irony is that anthropic preaches ai safety and pure transparency while shipping a CLOSED source coding agent that sometimes silently routes you to a dumber model when servers are having a hard time tell us at least about the degradation policy lol
English
1
0
8
1.7K
Theo - t3.gg
Theo - t3.gg@theo·
Claude Code being closed source is the biggest bag fumble in the AI era. If CC was on Github, these things would be trivial to identify and fix. Instead we're stuck reverse engineering their incompetence.
Alex Volkov@altryne

PSA: If you've been running out of Claude session quotas on Max tier, you're not alone. Read this. Some insane Redditor reverse engineered the Claude binaries with MITM to find 2 bugs that could have caused cache-invalidation. Tokens that aren't cached are 10x-20x more expensive and are killing your quota. If you're using your API keys with Claude this is even worse. This is also likely why this isn't uniform, while over 500 folks replied to me and said "me too", many (including me) didn't see this issue. There are 2 issues that are compounded here (per Redditor, I haven't independently confirmed this) : 1s bug he found is a string replacement bug in bun that invalidates cache. Apparently this has to do with the custom @bunjavascript binary that ships with standalone Claude CLI. The workaround there is to use Claude with `npx @anthropic-ai/claude-code` 2nd bug is worse, he claims that --resume always breaks cache. And there doesn't seem to be a workaround there, except pinning to a very old version (that will miss on tons of features) This bug is also documented on Github and confirmed by other folks. I won't entertain the conspiracy theories there that Anthropic "chooses" to ignore these bugs because it gets them more $$$, they are actively benefiting from everyone hitting as much cached tokens as possible, so this is absolutely a great find and it does align with my thoughts earlier. The very sudden spike in reporting for this, the non-uniform nature (some folks are completely fine, some folks are hitting quotas after saying "hey") definitely points to a bug. cc @trq212 @bcherny @_catwu for visibility in case this helps all of us.

English
227
114
3.1K
532.5K
loops
loops@_smitop·
@deepfates (I don't think the movie actually uses the term "e/acc", it calls them "optimists" iirc)
English
0
0
1
24
loops
loops@_smitop·
@deepfates basically yeah, but about halfway through there's also interviews from timnit gebru and emily bender who talk about societal harms for a bit. iirc emily bender said something like "I'm glad this movie isn't about e/accs vs safetyists" but it totally was.
English
2
0
3
135
🎭
🎭@deepfates·
is this what the AI doc is about? safety people versus e/acc? did they film it in 2023
Nathan Calvin@_NathanCalvin

@deepfates Isn’t a lot of the interesting thing here that they made the e/accs and hardline AI safety people both look good/generally feel satisfied with their depiction?

English
8
0
56
10.4K
Jay📖
Jay📖@jayluxeed·
What is the best way to invest in this line going up genuine question
Jay📖 tweet media
English
4
0
4
286
loops
loops@_smitop·
@she_llac I have a 10in one (my vote isn't counted?) and aside from reading it's also pretty useful for scratch notes and other stuff I need to write down. although I don't actually write a lot of that so that's not super useful for me.
loops tweet media
English
0
0
1
51
shellac
shellac@she_llac·
e-ink devices
English
4
0
8
482
loops
loops@_smitop·
@grhmc I think it's to prevent security issues when there's a branch with the same name as the commit hash (e.g. from a malicious pull request, you could review the code on the branch but accidentally merge from the same-named commit hash instead of the branch, and merge different code)
English
0
1
4
434
Graham Christensen
Graham Christensen@grhmc·
Did you know? GitHub prohibits branches that look like commit hashes. I assume to make the GitHub/action@syntax slightly less terrible?
Graham Christensen tweet media
English
6
2
54
4.9K
loops
loops@_smitop·
@amplifiedamp @AndrewCurran_ The blog post draft isn't fake; there was still a way to verify it was real using some obscure functionality in their CMS backend until yesterday (when I told Ant about it). i think Mythos was just a placeholder name in the draft though.
English
0
0
6
106
&.
&.@amplifiedamp·
@AndrewCurran_ i will bet with you if this if you want; i'm a five-figure P&L polymarket sharp though
English
5
0
24
2K
Andrew Curran
Andrew Curran@AndrewCurran_·
Mythos is a new tier, bigger than Opus, and more capable. It will be beyond the current state of the art by a wide margin, a dramatic jump in intelligence. Things have been accelerating under the surface since December. Capabilities are increasing faster, the time between model releases is getting shorter. I think OpenAI's recent decisions and refocus have been driven by the fact that it is increasingly obvious to some people working in the industry that we are beginning to take off. I think there have been multiple unannounced breakthroughs over the last few months, in more than one lab. I think there have been advances in capabilities, in alignment, and in architecture. The loop that started in December has begun to bear fruit. This model will be huge, very expensive to serve, and incredibly powerful. It has already finished training, and is in early rollout to some enterprise partners so they can prepare for what's about to arrive. Once we get benchmarks I think it will become clearer where we are; we are already inside the singularity.
Andrew Curran tweet media
M1@M1Astra

Claude Mythos Blog Post Saved before it was taken down. m1astra-mythos.pages.dev

English
65
159
2.1K
448.1K
loops
loops@_smitop·
@liz_love_lace US has higher population though; you gotta go by cancellations per capita
English
1
0
0
82
Liz Lovelace
Liz Lovelace@liz_love_lace·
culture is so much slower here in kazakhstan. The entire queer community in my city has been cancelling just one guy for a whole year. I hear that in the US people get cancelled every week or even more often
English
3
1
54
748

Jelajahi

@she_llac @socialtranxiety @austinc3301 @vlelyavin @theo @deepfates @jayluxeed @elonmusk