Microsoft Decision-1: 9B Qwen Model Launch Explained

Posted by gmays 3 days ago

Counter72Comment31OpenOriginal

Comments

Comment by mstkllah 3 days ago

Reads like an ad for Microsoft and its model because it's an article about a decision model that doesn't mention Jev or other decision models at all.

Comment by girvo 3 days ago

And reads like AI generated text, to boot. But then that might just be because it’s marketing, and LLMs write in non stop marketing speak

Comment by yas_hmaheshwari 3 days ago

Exactly my thought. I read the whole article, did Cmd +F for Jev - and felt uneasy on did the author did any research at all

Comment by martimarkov 3 days ago

This is really AI slop - so I don’t really know if it makes any sense for comments given that there is no real way to validate the crap that was put out.

It seems somebody bought the domain due to its rating. All the authors share the same type of image for their profile pics…

It’s sad that this content now come to the top of HN so often

Comment by yakbarber 3 days ago

Unlike the Internet, Microsoft did recognise the pivotal moment. But they made the wrong play… that and they’ve annoyed pretty much everybody with their products, I don’t think I’ve ever seen such a successful company with such little good will from their community.

Comment by Neikius 3 days ago

They are a de facto monopoly for decades. They can afford annoying everyone.

Comment by Forgeties79 3 days ago

EA

Comment by Den_VR 3 days ago

No pelicans riding bicycles on this one, eh?

Comment by kennywinker 3 days ago

Could one reverse the test? Feed ai generated svgs into the model and ask it to decide which of 10 animals (dog, cat, pelican, crocodile, etc) doing which of 10 actions (moonwalking, paddleboarding, sitting, riding a bicycle, etc.).

Comment by dannyw 3 days ago

AlexNet (2012) can do the first part :)

Comment by kennywinker 2 days ago

From an svg not rendered to pixels?

Comment by bushbaba 3 days ago

Instead the right app is hotdog / not hotdog

Comment by abhis3798 3 days ago

Pelicans are deciding.

Comment by jumperabg 3 days ago

Pelicans deciding if they are going to ride a bicycle.

Comment by egeozcan 3 days ago

Bikes with integrated decision models deciding which pelican needs a ride the most.

This made me realize it's been a long time since I did some brainstorming and how I missed it.

Comment by cronin101 3 days ago

Put a decision model in my bike that decides when it is time to automatically shift up/down based on signals (FTP, Watt avg, HR, cadence, upcoming grade) and I will buy a subscription to your SaaS.

Comment by v3ss0n 3 days ago

[flagged]

Comment by eblume 3 days ago

It's a meme. Simon Willison started doing it because, I'm paraphrasing, it's something that (at the time) was just sort of at the edge of what you could expect an LLM to be able to do, and also it's somewhat of an impossible task since, of course, Pellicans can't ride bicycles. From there, he started writing about it along with every major LLM release he covered. His coverage is generally well regarded (hi, Simon!) and so it sticks around. It's never been intended as a serious benchmark.

Comment by adamsmark 3 days ago

The 3D visualizations have always made it starkly apparent the difference in quality between quantizations and thinking.

The pelican meme is so much better than benchmaxxed stuff. Unfortunately we are likely now in pelicanmaxxing territory.

Comment by v3ss0n 3 days ago

already , i self hosted Deekseek 4.1 Flash , Qwen 3.8 Flash Next , Glm 5.3 flash , tested it , all benchmaxxed to the point that all of them can draw that well , much better than other logos .

Comment by v3ss0n 3 days ago

Sorry I didn't knew it was started here and , lol I didn't know it was viralized by Simon of django! Now these days , many LLMs are benchmaxxed to it.

Comment by locknitpicker 3 days ago

> Yeah the most usesless and stupidest bench ..

I recall that when this benchmark came out there were concerns that AI companies were gaming benchmarks to fake their progress. The purpose was to find something silly and outlandish that AI companies would definitely not have trained their models on, and thus have a way to check a model's capabilities.

A pelican riding a bicycle is pretty random. That was the whole point.

Comment by v3ss0n 3 days ago

https://news.ycombinator.com/threads?id=v3ss0n#:~:text=alrea...

Sorry i reply to wrong comment.

Comment by liuliu 3 days ago

I haven't evaluated this one, but for what I have, it seem the pplx-decider-v1.2 (and its -flash variant) are pretty good. Anyone evaluated all these open-source decision models on their own tasks? (I evaluated on MonitoringBench).

Comment by GavinEke 3 days ago

I didn't look specifically at open-source but I evaluated as many decision models as I could on openrouter against how good are they at detecting phishing, which isn't the best use of these models but I was interested to see how they compared. https://emdashcyber.com/phishnchips-decision-models/

Comment by theanonymousone 3 days ago

I had my first use case recently, and did some very non-systematic benchmarking of models on OpenRouter.

The only one that came close to Jev itself was the one from respan (respan/span-02), which is also open source and cheaper too.

Comment by mgrandl 3 days ago

Have you checked mammoth-1/elephant-1 models? They are benchmarking quite well compared and are currently free on openrouter. I currently use mammoth-1 for a small internal app and it works flawlessly for what I am doing.

https://huggingface.co/baselabs/mammoth-1

Comment by theanonymousone 2 days ago

No. I will, and I will report here.

Comment by theanonymousone 2 days ago

Did some ad hoc/small test. Unfortunately not very impressive. At least no on German short text, which is my use case.

Comment by mgrandl 2 days ago

Ah damn and also interesting. I am using it for classifying ads in podcasts. I am listening to a lot of English podcasts, but I am in Germany so often German segments are cut in. Mammoth-1 has had zero issues classifying those segments, but I guess my usecase might just be a lot simpler than yours.

You might also be interested in this index for decision models: https://huggingface.co/spaces/multimodalart/jev-decision-ind...

Comment by theanonymousone 2 days ago

Mine deals with sentiments of news headlines.

Thanks for the index. I will try to navigate it. For now I cannot even find this respan/span-02 in it :/

Comment by ricardobeat 3 days ago

> a model that refuses to write you a paragraph

AI slop reporting. Actual news: https://commandline.microsoft.com/microsoft-decision-1-model...

Comment by selcuka 3 days ago

[dead]