Symlink

DeepSeek Maximalist.

Interests: Energy, High-Density Housing, Infrastructure, Mass Transit, ML, Robotics, Software

[Symlink @SymlinkX]

@BoringBiz_ I mean the goal of rational actors within a capitalistic society is to become a rentier. This makes complete sense to me. If anything this is our society reaching a more mature form.

[Symlink @SymlinkX]

@daniel_mac8 I'd rather use GPT-5.6 because OpenAI doesn't seem to hate their subscribers.

[Symlink @SymlinkX]

@matt503ea5sf9z5 When it swaps to using Opus 4.8, it charges you from the Fable meter too, which is wild.

[Symlink @SymlinkX]

@jmbollenbacher OpenAI has been better value for quite a while now. 5.5 is genuinely great and they frequently reset usage limits (which can be stashed now).

[Symlink @SymlinkX]

@redtachyon None of the tweets that claim Sonnet 5 is significantly smaller than Sonnet 4.6 have backed it up with evidence. With that said, it does have a small model smell simply by how token inefficient it is to complete tasks. Smaller models in my experience need longer reasoning chains.

[Symlink @SymlinkX]

@jmbollenbacher They should have released it as the new Haiku then…

[Symlink @SymlinkX]

@ahall_research It’s a feature and not a bug. Neo-feudalism is the intent.

[Symlink @SymlinkX]

@pequityresearch Using Gemini instead of Claude or GPT 5.5? Even GLM 5.2 is better for agentic coding. No wonder Meta is cooked.

[Symlink @SymlinkX]

@AndrewCurran_ It's nice of closed labs to kneecap themselves by calling for their own self-regulation so that they can't release their own frontier models, giving space for open labs to catch up to what's less capable but generally available to the public.

[Symlink @SymlinkX]

@Mascobot Is that true though? You need a lot of compute to generate synthetic data to train on.

[Symlink @SymlinkX]

@himanshustwts What explains how bad Gemini is though?

[Symlink @SymlinkX]

@gdesouza Yes, you either hire people who are irrational, or people who are good at hiding their dishonesty. Both seem bad.

[Symlink @SymlinkX]

@teortaxesTex I think the Chinese labs are aware of where some of these gaps are. What I'm looking at is the rate of improvement not the pure capability gap at a point in time. As long as Chinese frontier labs remain open, I'm bullish.

[Symlink @SymlinkX]

@segyges Very bold of you to assume rationalists at Anthropic are actually sane?

[Symlink @SymlinkX]

@teortaxesTex Idealist? Yes. True and achievable vision? I don't think so. Fundamentally, the EA framework he believes in is broken.

[Symlink @SymlinkX]

@teortaxesTex Is this benchmark fr though? MiniMax M3 below Qwen3.6 27B? Is M3 really that bad?

[Symlink @SymlinkX]

@kyleichan He’s totally cooked isn’t he?

[Symlink @SymlinkX]

@PeterBerezinBCA The median surprises me more than the top 1%

[Symlink @SymlinkX]

@chi_t_williams I think the reason why this isn’t controversial is that Gemini isn’t even worth distilling CoT from.

[Symlink @SymlinkX]

@petergostev Just use https://t.co/4iO2f3ADGW. Given the fact that Fable is extra usage or API only, there isn't much downside either.