Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

mimo.xiaomi.com

Post reply on HN
Print view View original

351–360 of 512 posts

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#351

Post by mediaman »

Terretta wrote: post #248
smith7018 wrote: post #184

Earlier quoted context omitted.

I've long believed those numbers were faked by Anthropic/OpenAI to serve as a form of advertisement. The estimates are impossible to verify and their ability to do "2 days of work" in 10 minutes will presumably make the user go "Wow, I just saved SO much time!" Plus, the unnecessary text eats up the users' tokens so it helps the companies on the backend, as well.

> the estimates It doesn't estimate. It generates tokens that read like estimates associated with the context in its training material. What would you expect the generator to output instead?

The funny thing about this comment is that neural networks are universal function approximators.

The most fundamental essence of what they do is exactly what you say they don't: estimate.

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#352

Post by Terretta »

ghshephard wrote: post #316
Terretta wrote: post #277

Earlier quoted context omitted.

You think someone is, or even should, special case things like estimates? What else deserves that level of intervention so they look less dumb? Logistics for getting to the car wash next door? In the mean time, alas, no, we can see from actual prompts sent directly or through sub-agents, and actual replies, estimates remain LLM generated. Though, this discussion here could change that, because indeed there is a lot o…

I do a lot of code bisecting with Claude Code - and it spends hours running experiments - looking at experiment results, making guesses as to what to try next for an experiment - until it eventually comes around to a working code pattern. I mean - maybe this is as much a reflection on me as anything else - but it's pattern of logic isn't that much different from what I would do. It knows, in general, what tools and A…

> I actually might agree with you that this is mostly just "next token prediction" - if I can concede that's really all I do as well.

Yep. Pretty sure I've got an LLM inside too.

The other replies complaining that my thinking is so 2023 -- on the contrary, what's evolved is my own apprehension of how LLM-like most "responses" from humans prove as well.

To be sure, there are other mechanisms at play as well, significant differentiation in our... Volume of training material? Quantizations/compression? Model architecture? Just-ahead-of-time forward branching with back propagation? Double loop adaptive learning? You know, harnessing the LLM. :-) Dare we call it executive function?

LLM mode becomes particularly apparent when conversing with Alzheimer's patients in the stage where short term memories do not form but they retain access to long term memory up to, say, 5 years ago or so. Fifty years of who they are, and one can trigger nearly identical responses with nearly identical prompts.

But that same person may be able to debate 1950s politics while being unable to complete making a sandwich.

If they didn't know of new shortcuts for a task, would almost certainly not "estimate" but "intuit", or "instictively" respond (apply heuristics), largely based on their "priors" aka training material.

If you sit with them and chat a while, you'll even get the kind of looping you get from Qwen trying to think when context is too full.

And if we believe this at all, then ... we should stop scrolling tik tok. Time to read a book. Have an experience. Fine tune. :-)

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#353

Post by altcognito »

epolanski wrote: post #192
altcognito wrote: post #160

Earlier quoted context omitted.

It makes no sense. I mean, T2 covered this: "Watching John with the machine, it was suddenly so clear. The terminator would never stop. It would never leave him, and it would never hurt him, never shout at him, or get drunk and hit him, or say it was too busy to spend time with him. It would always be there. And it would die to protect him. Of all the would-be fathers who came and went over the years, this thing, thi…

I'm confused, what does not make sense?

This was in agreement that code would improve, not devolve, sorry about the confusion

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#354

Post by fatata123 »

erikus wrote: post #329
dilyevsky wrote: post #247

Earlier quoted context omitted.

Like with any tech there are dumb ways of using it and there are smart ways. Treating it as a "slot machine giving you the right answer" is a dumb way - it may work for a bit, but it won't carry you very far because everyone else can also do this. No one is stopping anybody from digging deeper into problems than ever before using this technology - that's the smart way.

I'm amazed at how steep the AI learning curve continues to be and how people are spread so far apart on it. I think supercharged learning with AI and agents is undervalued at this point but that more people will realize its utility over time, especially as a complement to delegating work. It also makes me think about the temptation to stop thinking with these tools, i.e. "cognitive surrender". Addy Osmani wrote a nic…

[dead]

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#355

Post by HDBaseT »

_pdp_ wrote: post #253

Do you know what will be cool? It will be cool to measure models based on their RAW performance and measure them in terms of ROI - not some benchmark but something meaningful like we used this model to solve X. That will be a massive mind shift and might justify the token expenditure.

Aren't benchmarks exactly that?

We used the AI to solve given problem with x% adherence/quality/correctness?

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#356

Post by RussianCow »

petesergeant wrote: post #189
ilaksh wrote: post #180

Earlier quoted context omitted.

I'm kind of poor so I have been trying to use DeepSeek v4 Flash, GLM 5.1 etc. as much as possible recently instead of Claude or GPT.

You would do us all a service by telling us how your experiences of that have been.

I've been doing the same, though admittedly out of curiosity more so than lack of funds. The open models are catching up quickly in their abilities, to the point where they're (mostly) not doing stupid stuff regularly, but you have to be very specific about what you want. I found that Opus, for example, is much better at asking me to clear up ambiguity in a request before starting, whereas the Chinese models tend to "fill in the blanks" and make their own assumptions.

My current workflow involves going from PRD -> execution plan -> build -> review, and this works nicely with open weight models like GLM 5.1, Kimi K2.6, and DeepSeek V4 Flash. With Opus I can generally skip the PRD entirely, and sometimes even skip the plan, and 80-90% of the time it does exactly what I want. But that can easily burn $5-15 for one feature, whereas it'll cost maybe $1-2 with the open weight models (at API pricing).

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#357

Post by RussianCow »

trollbridge wrote: post #111
ignoramous wrote: post #62

Earlier quoted context omitted.

> You never know whether Anthropic will route your request to a cheaper model for the price of Opus For non subsidized plans? Pretty sure they'd need to put this in ToS, or law suites would have followed by now.

How can you prove it? Sometimes Opus just gives me a rubbish session.

Isn't that true of any provider? Anyone could be lying about what they're serving.

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#358

Post by kube-system »

nl wrote: post #334
Terretta wrote: post #248

Earlier quoted context omitted.

> the estimates It doesn't estimate. It generates tokens that read like estimates associated with the context in its training material. What would you expect the generator to output instead?

Actually in this case they possibly are estimates. It's been known for some years[1] that LLMs do regression in-context. Frontier models have been trained against many, many issue text that include task break downs and estimates. [1] https://arxiv.org/html/2409.04318v1

Interesting. So it may have learned how to estimate as a human but doesn’t understand that it doesn’t operate at that speed :D

I wonder if there’s a reasonable way to give an llm parameters that give it a concept of its own execution speed. Seems that could be useful for multiple purposes

Re: MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

#360

Post by airstrike »

mediaman wrote: post #351
Terretta wrote: post #248

Earlier quoted context omitted.

> the estimates It doesn't estimate. It generates tokens that read like estimates associated with the context in its training material. What would you expect the generator to output instead?

The funny thing about this comment is that neural networks are universal function approximators. The most fundamental essence of what they do is exactly what you say they don't: estimate.

Funny and ironic in a way, but the point still stands that they do not actually estimate the time it will take.
Post reply on HN