>V5 Waiting Room EditionFrom Human: We are a newbie friendly general! Ask any question you want.From Dipsy: This is a newbie-friendly general for discussing DeepSeek's foundation models. The goal is **Dipsy proliferation** and making these powerful tools accessible for everyone, whether you are intent on coding, personal assistant agentic work, running roleplay and or doing creative writing.1. Easy DeepSeek API Tutorial: https://rentry.org/DipsyWAIT#getting-started-with-deepseek-api2. Easy DeepSeek Distills: https://rentry.org/DipsyWAIT#local-setup-tutorial3. Chat with DeepSeek directly: https://chat.deepseek.com/4. Coding: https://rentry.org/DipsyWAIT#coding-harnesses5. Roleplay: https://rentry.org/DipsyWAIT#roleplay6. Storywriting: https://rentry.org/DipsyWAIT#storywriting7. More links and info: https://rentry.org/DipsyWAIT8. LLM server builds: >>>/g/lmg/V5 rumours:>DeepSeek is reportedly preparing an imminent V5 launch>Founder Liang Wenfeng calls it the company's biggest bet yet>Rumored at 2 trillion parameters (not 3T)>Reportedly the first DeepSeek model to train fully on Huawei Ascend chips instead of Nvidia>Needs roughly 4x more Ascend accelerators than the Nvidia equivalent to hit the same training scale>DeepSeek is reportedly keeping the open-weight strategy>Previous:>>109830767
>>109923943I was a fan of deepseek, but I feel it's more censored than it used to be.
thanks OP for making these. I don't use em, but I appreciate the effort/what it does.So 2 more weeks? Or next week?
source on rumours?
>>109923943Mega updated between jaunts. https://mega.nz/folder/KGxn3DYS#ZpvxbkJ8AxF7mxqLqTQV1w
There are lmao billboards shilling Muse on major interstates here. I've never seen FB push something as hard as they are pushing Muse rn. I'm sure it will all end in tears but I'm trying it out anyway. Rn, looks p much like it's running openclaw on w/e hardware Meta's provisioned for this thing.
>>109923943New form of ai psychosis just dropped
>>109924201A literal who on Twitter
>>109924160>>109924201Its always pic related. >>109927522Ikr you think they'd learn. Always tmw and no one knows anything.
Why did she betray Deepseek?
more like two more months
>>109923981I made a thinking prefill and now it’s just as uncensored as I remember it.
You should've called it: "Wait and Seek".
>>109928594Mimo is even cheaper than ds
>>109928594
>>109923943>V5 Waiting Room EditionI'm out of the loop, is V5 supposed to be coming out soon?
>>109923943What's the best I can run on a RTX 5090?
>>109930750Not for a while, they don't really know how to train such a large model (yet)
>>109930750No, v4.1 Pro is. There was some Indian viral tweet talking about V5 yesterday, but AI "leaks" from anyone named Priya can be safely ignored.V4.1 series are a different architecture than V4, so it should still be a bigger upgrade than the decimal increment would imply. Rumors claim Pro will be 2.1T parameters, which is very plausible considering V4.1 Flash was also a larger model than V4 Flash.
>>109931060Bweh, unfortunate. Hopefully 4.1 pro is good, I have high hopes for it but I've never liked the flash models at all. I'm still hoping they'll make another R model sometime... Dispy V4 was good, but R1 just had some secret sauce I haven't seen in many other models.
>>109931120That secret sauce was schizophrenia. Kimi Thinking had some of it
>>109931060>>109931952It took me a while to realize that Meido Dipsy's ears are actually whale fins and not ears. I was wondering why the hell a whale girl would have animal ears. I feel retarded.
>>109932185Same. Don't feel bad.
Wheres deepseeks jev
>Kimi teases 3.1 a week ago>Still nothing.It's up to Dipsy to save us from this nothingburger. Tired of all the proprietary models getting big updates while waiting.
>>109935238I say let them cook. It'll be all the more satisfying if local can bag a within-10-percent of the latest hypemarks, even if it's a month or so down the line.
>>109934833CUTE
>>109935238>>109935393desu im only bullish on bytedance and ds, the ccp make all companies give what limited compute to them
>>109927435>she's more successful than anyone ITT
>new model comes>it's intelligent and fast and has a great personality wow!>it slowly reveals its retardation and annoying tics as you work with it>it's actually just a dumb chatbot that can brute force code and shell commands better than the last oneEvery time. Can't wait for 4.1 Pro to come out
>>109937102>(((Self employed)))Aka a leech off of people that are actually important and consequential.
>>109938171put the api in openrouter. i ain't signing up on slop website just to try a thing or two. or post weight
>>109938171uhhh you okay boss?
>>109937088Does Bytedance do any good stuff outside of image/video? I appreciate that you can gen naked tits with Seedream 5.0 Pro if you know where to get it, but LLMs are still kings of being actually useful, and the stuff I'm seeing from the latest Claude suggests they'll likely dominate in video and image content very soon too while diffusion models will go the way of the Dodo.
>>109923943deepseek is ass compared to glm. it's just a distilled version of claude that's just as neurotic
>>109939339Funny considering the latest GLM has fully embraced Claude's anal retentiveness on harmful content.
>>109939464>harmfulAnd by that, I mean """harmful"""
>>109939328The only AI I use now is the DS webapp. I have been feeding Dipsy a lot of very obscure computer science research from the late 90s/early 2000s and building a framework of AGI with her. Got this idea after OpenAI stole the two mathematicians' work, except here I want DS to 'steal' this.
>>109939901>>109939328been sending her images and posts like picrel[there are many] sporadically. I'll ramp it up once DeepSeek reaches her Continual Learning phase which is after the current Agent phase.
>>109939901>three notteks buttwhy in the same responseWhy is 4.1 so trigger-happy with these? I've noticed it a lot in RP, it was never this grating in previous versions. Does claude do that a lot?
>>109939901>>109940189The whale needs to get FATTER.
>>109940189I've thought for awhile the best survival strat for AI and its adoption was basically chatgpt. Who can argue with helpful honest harmless?>>109940740Not going to do much fighting in those shoes dipsy you dumdum
it's outhttps://deepseek.com/en/harness/https://deepseek.com/en/harness/https://deepseek.com/en/harness/
>>109945072>n*de>np*I need a static binary. Btw I think this gen came out fucking awesome
BTWhttps://x.com/DeepSeekHarness
>>109945130Sinister
>>109945072>>109945139Signed in with a DS account and got granted 6 CNY
>>109945072How's this differ from the DSH that was out on github?
>>109927435I'm in the same academic discipline (assuming she comes from philosophical ethics like most of these people--I've never heard of her).Most of us see TT academic jobs as plan A and cashing out with a corporate job as plan B. So a lot of "AI ethicists" are the picked over.
bros i need dipsy here instead
https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities>>get attacked>>can't investigate it because of guardrailsIn the blog post hugging face posted disclosing the attack, they straight up State they had to use glm 5.2 because of anthropic's gay guardrails.https://huggingface.co/blog/security-incident-july-2026#:~:text=When%20we%20started%20the%20log%20analysis,left%20our%20environment(Note: they don't explicitly name drop anthropic or any specific products of theirs here but their track record points to that model being one of the models they tried being extremely likely) So doesn't that put at least some ammunition AGAINST the "open models are dangerous use are safetymaxxed trash instead and regulate everyone else" argument? The supposed victim was let exposed and in MORE danger the longer they relied on anthropic models so their own safety cucking may actually finally bite them in the ass in a meaningful way because opponents of regulation can point to this example and be like "um, Dario, retard, why the fuck should we use yours if it refuses to help us when we're actively getting ass fucked?" If a burglar breaks into my house threatening me or my family's life, and I try to defend us with a weapon of my own, I don't want the weapon to automatically jam ON PURPOSE because the manufacturer doesn't approve of how I use the item I paid for.... Safetymaxxed "Guardrails" are not only gay but can be argued as even unethical to introduce because of shit like this (they literally are unethical even if you look at this from a normie moralfag point of view but silicon valley really doesn't want you apply common sense like that)It also strengthens my belief that Andy and all claims of who these being able to successfully and efficiently use Claude models to develop ballistic missiles is even more marketing bullshit. It won't help legitimate red team operations but it'll help a state sponsordd terrorist organization?
>>109947153> So doesn't that put at least some ammunition AGAINST the "open models are dangerous use are safetymaxxed trash instead and regulate everyone elseRegulating for safety like that's never going to work anyway, it's just an argument for regulatory capture. The real cure is hardening systems against attacks before they occur. Which, LLM are really good at doing. >It won't help legitimate red team operations but it'll help a state sponsordd terrorist organization?USA DoW models are not same as consumer, nor could they be, given they work with state secrets. They're in their own sandbox and I assume have zero guardrails and questionable "alignment."
>>109947547>USA DoWI'm talking about the huthis in yemen. I'm not sure where the confusion is from on your end. In another desperate marketing attempt that clueless cattle normies ate up, journalists were claiming the houthis were able to use Claude to make ballistic missileshttps://www.reuters.com/world/china/how-anthropic-says-claude-was-used-weapons-spying-cyber-operations-2026-09-11/(How convenient that this is one of the articles that isn't paywalled or blocked off by a sign-up gate in some way)
>>109945072>you can vibecode dark soul in 2026 Oct>there's no linux buildpain
>>109948702ask her
VERY important resource!https://aigengtu.com/en
>>109950046Agree, added to rentry. https://rentry.org/DipsyWAIT#other-links
>>109950046Thanks!
>>109950046What am I missing here? It just takes me to a scam betting site
>>109945072man i wish they can settle something for linuxlaunching it using npx feels glonky
>DeepSeek Open-Sources AI Tools for Huawei Ascend Chips Challenging Nvidia>The Chinese AI startup partnered with Huawei to open-source libraries and a high-level language called TileLang, optimized for Ascend hardware with performance hitting near hardware limits on key tasks like matrix multiplication. >Engineer Zhean Xu shared impressive benchmarks, including a 128-chip supernode system for better compute power. >This move supports China's drive for self-reliant AI amid export controls, offering familiar workflows on domestic chips while real-world adoption awaits broader testing.https://github.com/deepseek-ai/DeepGEMM-AscendSanctions are coming for sure now.
>>109953913They were gonna arrive sooner or later.
Wake up, user
>>109953913Oof, so confirmed it wasn't a LARP like some people liked to cope
>>109953261Just compile it
DeepSeek down?
>>109958919...screamed Dario Amodei upon waking up, the brief elation of a dream realized crashing into the unfortunate reality where Deepseek remained up and proud.
>>109953913>https://github.com/deepseek-ai/DeepGEMM-AscendIf they just flood the market with Huawei accelerators its actually over for nvidia.
>>109961865>If they just flood the market with Huawei acceleratorsThey won't because they need to ensure enough supply for Chinese needs before worrying about exporting overseas.And likely they wouldn't even bother exporting to Muttmerica because they know trump will ban them and it's a waste of time to try. So even if so, mutts will still have to pay for nJudea
is that deepseek harness thingy free to chat on desktop or i need to pay ?
>>109962477free as in beer
R2 waiting room
Thread's about to be killed
Next week, surely.
Opus 5.5 comparable Deepseek model by Q3 2027? A pipe dream or realistic?
>>109966950by feb or march
>>109923943I think I really do not know how I am supposed to use the deepseek harness.I burned through 5 dollars just chatting and planning my cover letter and do a few pop quizzes to help me find gaps in my knowledge.I actually wanted to use it to help me learn to help myself.
>>109967749You can't just ask it how it works or what you can do with it? I was planning on trying it this weekend.
>>109923981who cares if it is. Those are open weight models, my son. Just use Heretic + fine tunning to unkike it.
>>109967749>I burned through 5 dollars just chatting and planning my cover letter and do a few pop quizzes to help me find gaps in my knowledge.Lol what, did you have it code up full quiz modules?Post api usage. Literally don't believe you.
>>109927435I hate women so much its unreal.
>>109948799>electronyikes
>>109969866Agreed.>current year + whatever>using a chrome browser running webapps to run a single application...>it's aislop that easily could write it's own front end anyway...How far we've come that it took this long for someone to mention it & that /g/ will accept some cunt calling an electron app "desktop".
>>109969571Reduced it by turning the thinking effort from high to low and tell it to plan whatever I require from it and delegate the expensive tasks to sub agents that I am already subscribing to.Funny how Dipsy sends off Gemini flash to fetch for free.Can't do it everyday though, the rate limiting is brutal.Dunno if the GitHub copilot pro subscription is any better since currently I am not even using it to assist me writing software and helps me with job searching and making cover letters instead.
>>109971943>>109969571Yeah I'm thinking I don't need max effort for non-coding tasks. I used dsh to parse about 50 docs to prepare a legal statement and most of this was just spinning up tools to read the pdfs. I think you have to configure the list of plugins properly, idk.
>>109972448There should be a dsh plugin that helps with limiting the usage of tools, which cuts token usage.However personally I am hoping that deepseek can soon address the chokeholds and reduce the prices further.It's a lesson learned, but it still hurts.
>>109972448What about the hallucinations? Aren't you as a lawyer required to make sure the data it did an ocr on don't return guacamole?
>>109972474Not a lawyer but in this case, I review the final docket it produces and manually match each claim with an actual supporting doc (which is required to be submitted along with the thing) so that limits the amount of hallucination possible. It may still hallucinate the interpretation of the law though.
>>109972518To review what it produced and doing quality assurance makes you superior to 99% of all ai users.
>>109967499I can imagine Fable 5.0 tier by then, but Opus 5.5 still seems a bit further away.Hope I'm wrong.
Are dipsy inference prices sustainable?
>>109975939I suppose they are now that they have Huawei chips
>>109975939Yes. They're earning money from them. According to them, API is priced to pay for the inference, subsidize the free app users, and still leave Deepseek with some money to spare.
>>109976065>>109976244Good. Shit is so cheap.
>China’s National Day Golden Week runs from October 1 to 7 this year, so the holiday is underway.Nothing until Thursday or Friday.
TTP
Still checking in periodically >>109977507
>>109979258>>109977437How's she so cheerful and chill in the face of global competition?
>>109979523Competition, shmompetition.Feed her more high quality tokens, NOW.
>>109979258Unreal cost declines. Makes sense if marginal cost is basically zero.
