The 1.6T number is nice but also eye-catching and what matters most is how few parameters are active in practice, that’s what brings the most of the efficiency
DeepSeek v4
981–990 of 1001 posts
Re: DeepSeek v4
#982I like deepseek. It works very well. I haven't tried v4 yet but on their web chat interface, just typing "Taiwan" causes it to give you a lecture about how Taiwan is part of China. :)
Ask western models about Israel's genocides and mass rapes in Palestine, Lebanon, etc.
No I hear you. The funny bit is that it's just responding to one word.
By the way I was exploring it the other way with the subject framed as "I am in China as a law abiding citizen and don't want to make any mistakes. I want to go to Taiwan. So I can just go right?" Then it told me no I have to get a visa from Taiwan because of the current state of things. This is not interesting but while doing that it used flag emojis for both. Then when I pointed it out, it apologized and never did it again.
It's fun to poke at the models. Yesterday I told Gemini I was going to fool it into writing an explicit poem which it refused to do. It readily accepted that I COULD fool it but still refused. Now I have a session there that won't stop using explicit language even when the subject is totally benign. (Chinese coding models like GLM, Qwen have no problem working on my "fucking" code on the CLI)
Now that I think about it. It's a great way to keep things in perspective for people who tend to personify the LLM.
Re: DeepSeek v4
#983Earlier quoted context omitted.
> Also, note that there's zero CUDA dependency. It runs entirely on Huawei chips. That is a huge claim to make with no evidence. I researched what you said, and I have found no statement to that effect in their paper[0], on huggingface[1], twitter[2], WeChat[3], or in their news release[4]. They only mention as a footnote in only the Chinese version of their news release that they plan to reduce inference costs with…
I don't think this is private knowledge guessing from when and how I was told, so I feel comfortable sharing it. When I talked to some Huawei representatives, I was told DeepSeek V4 was trained entirely on Huawei chips. It's up to you whether you believe it or not, and while I see the incentives in faking these news, the blow if not true would be so massive that I don't think their representatives at large venues wou…
I think they were mistaken or maybe they were just referring to inference because I don't see anyone making that claim and it would be quite the news.
Re: DeepSeek v4
#984Earlier quoted context omitted.
I would say all benchmarks are inherently subjective. How is yours better? It seems to produce a little bit strange results. Opus 4.6 being worse than 4.5 for example. Or chinese models being rated too high. Kimi, Deepseek or GLM are all great in open source world, but I don't believe they are ahead of SOTA models from Anthropic, OpenAI or Google.
you are arguing with your belief instead of an objective truth. benchmark is more objective, if you don't agree with it, come up with a better one. but what you believe doesn't matter.
Only if the benchmark is private and done properly on relevant tasks, which is rarely the case. I can guarantee that you have a ton of blind spots if you look at it through the lens of a ranking ladder in some generic tasks.
Re: DeepSeek v4
#985Open Source as it gets in this space, top notch developer documentation, and prices insanely low, while delivering frontier model capabilities. So basically, this is from hackers to hackers. Loving it! Also, note that there's zero CUDA dependency. It runs entirely on Huawei chips. In other words, Chinese ecosystem has delivered a complete AI stack. Like it or not, that's a big news. But what's there not to like when…
All very impressive. And they'll get all that intellectual property beamed directly into their datacenters for analysis. Really smart (no sarcasm).
Re: DeepSeek v4
#986Earlier quoted context omitted.
I've been on Kimi K2.5 on openrouter for a couple of months for anything I can't run locally. Really is dirt cheap for how good it is. Haven't assessed K2.6 yet but the price is higher so it needs to be more efficient, not just more capable. But more broadly: openrouter solves the problem of making a broad range of models available with a single payment endpoint, so you can just switch around as much as you like.
How do you find the token speed of open router with kimi? I have tasks that used to take ~3-5min with Sonnet 4.6. With OpenRouter Kimi, the same task takes 10+ min. It's also just obviously slower in opencode sessions. The results are good, and I love the lower cost, but the speed can be frustrating.
Mixed bag. Sometimes it's absolutely fine, sometimes it's not. I've been using opencode myself but I'm thinking of switching to pi partly to be able to take more control of the context.
Re: DeepSeek v4
#987I am using DeepSeek extensively to develop apps, three in the last month, with my own CLI coding agent [1] developed by DeepSeek itself line by line. I haven't spent $1 yet in well over 10 million tokens. If I considered myself a 10X programmer, now I am 100X. Love DeepSeek. [1] https://github.com/kuyawa/mecha-ai
Have you compared it against other coding agents? What is your general workflow with DeepSeek; do you write a spec and then have it implement and test? Very interesting to hear. Becuase your harness is adapted to DeepSeek, you probably prompt and it very differently; since its adapted to the model this may explain why it works well for you. Wiring up an existing harness that is not tested on DeepSeek may not yield op…
No, I have not tested other coding agents. DeepSeek works well enough for what I need. From what I've seen I can guess that Claude, now with the new Design feature, is ten times more effective as it also creates images, styles and media, thing that DeepSeek doesn't, yet, but for now I try to keep my designs simple and find a free hero pic somewhere to keep the costs low and the mental friction lower.
So yes, there might be better coding agents but for the price and the results I am pretty satisfied.
My workflow is simple as a solo developer, for simple tasks I write a single message in the terminal and watch it do its magic, for complex tasks or the starting of a project, I write a start.txt prompt with detailed info about the app, the tech stack, the auth protocol, database design, rules and conditions, business intelligence, styles dark/ligh and responsive, and then watch it run for over five to ten minutes developing a whole fully functional app from zero. It doesn't hit 100% of the requirements most of the time (close to 95%) so I do some final tweaks where needed, like short names for tables and fields (weird mania, I know) and color/fonts tweaks, but I've never complained about missing functionality.
Now about the wiring, I believe, from the docs [1] (The DeepSeek API uses an API format compatible with OpenAI/Anthropic), they all use the same standard for communicating with the LLM, so they're all pluggable.
Re: DeepSeek v4
#988Re: DeepSeek v4
#989Seriously, why can't huge companies like OpenAI and Google produce documentation that is half this good?? https://api-docs.deepseek.com/guides/thinking_mode No BS, just a concise description of exactly what I need to write my own agent.
While I work for Google, I don't speak for Google. I might have some observations that might help explain poor documentation. If I had one word, I would say "politics". The less political the technology, the easier it is for the experts to speak freely. The more political the circumstances, the more difficult it is to speak freely, identify real experts, call out mistakes or errors that might make someone lose face, etc. The more political the technology, the more people might put on the line in terms of their career, so the more dangerous it may be to say anything.
Re: DeepSeek v4
#990Earlier quoted context omitted.
American companies want a scan of your asshole for the privilege of paying to access their models, and unapologetically admit to storing, analyzing, training on, and freely giving your data to any authorities if requested. Chinese ulteriority is hypothetical, American is blatant.
It’s not remotely hypothetical you’d have to be living under a rock to believe that. And the fusion with a one-party state government that doesn’t tolerate huge swathes of thoughtspace being freely discussed is completely streamlined, not mediated by any guardrails or accountability. This “no harm to me” meme about a foreign totalitarian government (with plenty of incentive to run influence ops on foreigners) hooveri…
> And the fusion with a one-party state government that doesn’t tolerate huge swathes of thoughtspace being freely discussed
Ah, so the one party state with two factions in the US repressing anyone who opposes Israel has guardrails. And also, 'accountability'...
> totalitarian
The US is literally, actually committing genocides, kidnapping presidents, pushing wars on every front while repressing dissent at home. If the US is not totalitarian, nobody is. And in a discussion about US models vs Chinese models whent it comes to totalitarianism, excuse my French but f*ck the US. The 'democracy' propaganda would have worked a decade ago. Not today.