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

Mercury: Commercial-scale diffusion language model

inceptionlabs.ai

Post reply on HN
Print view View original

51–60 of 189 posts

Re: Mercury: Commercial-scale diffusion language model

#51

Post by firejake308 »

drusepth wrote: post #32
jonplackett wrote: post #5

Ok. My go to puzzle is this: You have 2 minutes to cool down a cup of coffee to the lowest temp you can You have two options: 1. Add cold milk immediately, then let it sit for 2 mins. 2. Let it sit for 2 mins, then add the cold milk. Which one cools the coffee to the lowest temperature and why? And Mercury gets this right - while as of right now ChatGPT 4o get it wrong. So that’s pretty impressive.

Is there a trick question in here, where the cold milk is trending closer to room temperature over the two minutes waited?

The rate of change of temperature is proportional to the difference in temperature. Adding the cold milk later let's you juice a larger temperature differential out of the 2-minute waiting phase. So you should add the milk later.

Re: Mercury: Commercial-scale diffusion language model

#52

Post by nullc »

nullc wrote: post #46
byearthithatius wrote: post #34

Earlier quoted context omitted.

Both are conditional distributions on the context of which they were requested so like you said in the second paragraph, the difference is not significant. I see what you mean though and maybe there are use cases then where Diffusion is preferable. To me it seems the context conditional and internal model is sufficient where this problem doesn't really occur.

::nods:: in the case of diffusion though "conditional on its own (eventual) output" is more transparent and explicit. As an example of one place that might make a difference is that some external syntax restriction in the sampler is going to enforce the next character after a space is "{". Your normal AR LLM doesn't know about this restriction and may pick the tokens leading up to the "{" in a way which is regrettabl…

Totally separate from this line of discussion is that if you want to use an LLM for, say, copyediting it's pretty obvious to me how a diffusion model could get much better results.

Like if you take your existing document and measure the probability of your actual word vs an AR model's output, varrious words are going to show up as erroneously improbable even when the following text makes them obvious. A diffusion model should just be able to score up the entire text conditioned on the entire text rather than just the text in front of it.

Re: Mercury: Commercial-scale diffusion language model

#53

Post by freeqaz »

Anybody able to get the "View Technical Report" button at the bottom to do anything? I was curious to glean more details but it doesn't work on either of my devices.

I'm curious what level of detail they're comfortable publishing around this, or are they going full secret mode?

Re: Mercury: Commercial-scale diffusion language model

#54

Post by jefftk »

jonplackett wrote: post #5

Ok. My go to puzzle is this: You have 2 minutes to cool down a cup of coffee to the lowest temp you can You have two options: 1. Add cold milk immediately, then let it sit for 2 mins. 2. Let it sit for 2 mins, then add the cold milk. Which one cools the coffee to the lowest temperature and why? And Mercury gets this right - while as of right now ChatGPT 4o get it wrong. So that’s pretty impressive.

Claude 3.7 gets it exactly right:

To determine which option cools coffee the most, I'll analyze the heat transfer physics involved. The key insight is that the rate of heat loss depends on the temperature difference between the coffee and the surrounding air. When the coffee is hotter, it loses heat faster. Option 1 (add milk first, then wait):

- Adding cold milk immediately lowers the coffee temperature right away

- The coffee then cools more slowly during the 2-minute wait because the temperature difference with the environment is smaller

Option 2 (wait first, then add milk):

- The hot coffee cools rapidly during the 2-minute wait due to the large temperature difference

- Then the cold milk is added, creating an additional temperature drop at the end

Option 2 will result in the lowest final temperature. This is because the hotter coffee in option 2 loses heat more efficiently during the waiting period (following Newton's Law of Cooling), and then gets the same cooling benefit from the milk addition at the end. The mathematical principle behind this is that the rate of cooling is proportional to the temperature difference, so keeping the coffee hotter during the waiting period maximizes heat loss to the environment.

Re: Mercury: Commercial-scale diffusion language model

#55

Post by freeqaz »

AlexCoventry wrote: post #35
XenophileJKO wrote: post #29

Earlier quoted context omitted.

So my personal belief is that diffusion models will enable higher degrees of accuracy. This is because unlike an auto-regressive model it can adjust a whole block of tokens when it encounters some kind of disjunction. Think of the old example where an auto regressive model would output: "There are 2 possibilities.." before it really enumerated them. Often the model has trouble overcoming the bias and will hallucinate…

> it can adjust a whole block of tokens when it encounters some kind of disjunction. This is true in principle for general diffusion models, but I don't think it's true for the noise model they use in Mercury (at least, going by a couple of academic papers authored by the Inception co-founders.) Their model generates noise by masking a token, and once it's masked, it stays masked. So the reverse-diffusion gets to dec…

Here are two papers linked from Inception's site:

1. Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution - https://arxiv.org/abs/2310.16834

2. Simple and Effective Masked Diffusion Language Models - https://arxiv.org/abs/2406.07524

Re: Mercury: Commercial-scale diffusion language model

#57

Post by twic »

jonplackett wrote: post #5

Ok. My go to puzzle is this: You have 2 minutes to cool down a cup of coffee to the lowest temp you can You have two options: 1. Add cold milk immediately, then let it sit for 2 mins. 2. Let it sit for 2 mins, then add the cold milk. Which one cools the coffee to the lowest temperature and why? And Mercury gets this right - while as of right now ChatGPT 4o get it wrong. So that’s pretty impressive.

Depends on the shape of the cup! You can contrive a cup shaped like an exponentially flaring horn, where adding the milk increases the volume a little, which massively increases the surface area, and so leads to faster cooling. Or you can have a cup with a converging top, like a brandy glass, where adding the milk reduces the surface area, and makes cooling even slower.

Re: Mercury: Commercial-scale diffusion language model

#59

Post by jbellis »

schappim wrote: post #41

It's nice to see a team doing something different. The cost[1] is US$1.00 per million output tokens and US$0.25 per million input tokens. By comparison, Gemini 2.5 Flash Preview charges US$0.15 per million tokens for text input and $0.60 (non-thinking) output[2]. Hmmm... at those prices they need to focus on markets where speed is especially important, eg high-frequency trading, transcription/translation services and…

What is the price on Mercury Mini?

Re: Mercury: Commercial-scale diffusion language model

#60

Post by jbellis »

jtonz wrote: post #40

I would be interested to see how people would apply this working as a coding assistant. For me, its application in solutioning seem very strong, particularly vibe coding, and potentially agentic coding. One of my main gripes with LLM-assisted coding is that for me to get the output which catches all scenarios I envision takes multiple attempts in refining my prompt requiring regeneration of the output. Iterations are…

Unfortunately a 4o mini level of intelligence just isn't enough to make this work, no matter how many iterations you let it try.
Post reply on HN