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

OpenAI releasing new open model in coming months, seeks community feedback

openai.com

Post reply on HN
Print view View original

61–70 of 82 posts

Re: OpenAI releasing new open model in coming months, seeks community feedback

#61

Post by refulgentis »

bigyabai wrote: post #3
CamperBob2 wrote: post #2

What would you like to see in an open-weight model from OpenAI? Explain what you would use it for See... that's kinda the idea behind an open model. I don't have to explain what I would use it for.

The idea that they want to train a new custom model for open release instead of just... giving us GPT-3 already suggests a terrible start. I'm calling it now, this is a strategic counterplay against Google's Gemma models so @sama can sell Tim Cook a "frontier" local model that doesn't compete with anything coherent. A fig leaf for their "Open" identity and a paper tiger for the Apple Intelligence panoply. OpenAI does…

Is there anything about training a new model? I just assumed they were asking for all the little bits that companies forget to do at release with an open model.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#62

Post by turnsout »

minimaxir wrote: post #15
turnsout wrote: post #9

This definitely feels like a move made out of desperation, but I admit I'm curious to see what they release. It's fascinating that some of the best-funded startups of the 2020s are all rushing to commoditize their core technology as quickly as possible. They all seem to have the same business model as the Change Bank from SNL

Desperation? How so? The cynical late-stage-capitalism argument for releasing an open source model is for PR/goodwill and hoping for the chance the model becomes a foundation model in the OSS community. But OpenAI definitely isn't operating around good PR and the chance of releasing a foundation model is unlikely given the competition.

Desperation because they face an existential risk from trillion dollar companies (and others) creating open models that are starting to catch up to their flagship.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#63

Post by 9rx »

wewtyflakes wrote: post #57
9rx wrote: post #47

Earlier quoted context omitted.

Given that they come in good faith (trolling is already excepted), I would ask. Burger toppings are predominately plant-based, so perfectly in their wheelhouse. In fact, I'd expect better suggestions from them than the average burger eater.

This seems like a stretch. Another example may be, "If I am throwing a party, should I ask people that I do not invite, what they want?". Sure you could say that they could approximate the wants of the people who _are_ going, but why not just ask the people who are going, directly? It's just noise, otherwise.

> why not just ask the people who are going, directly?

"If I had asked people what they wanted, they would have said faster horses"

You don't poll people to find out what they want. You poll people to gather their ideas. Ideas that you can then leverage to deliver what your intended audience wants, even when they didn't know that they wanted it!

If we assume this party you are throwing has 10 guests, you think you're going to get all the best ideas from those 10 specific people and nothing from the hundreds of people you could have asked? Maybe if you're throwing a party for professional party planners, but otherwise...

Re: OpenAI releasing new open model in coming months, seeks community feedback

#64

Post by mackenly »

minimaxir wrote: post #5
bigyabai wrote: post #3

Earlier quoted context omitted.

The idea that they want to train a new custom model for open release instead of just... giving us GPT-3 already suggests a terrible start. I'm calling it now, this is a strategic counterplay against Google's Gemma models so @sama can sell Tim Cook a "frontier" local model that doesn't compete with anything coherent. A fig leaf for their "Open" identity and a paper tiger for the Apple Intelligence panoply. OpenAI does…

> instead of just... giving us GPT-3 If you're referring to the GPT-3 from 2020, modern open source models five years later are a) better at benchmarks b) much smaller yet still better at said benchmarks c) much, much cheaper/faster due to architectural improvements. The real hard thing for OpenAI to do is to release an open-weights model that's better/more differentiated than Gemma 3 (at the small scale) or DeepSeek…

But if OpenAI was Open, they'd open-source those old obsolete models. You're right that no one really wants it, and it has little to no commercial value at this point, but that's all the more reason they should just put it out there and actually live up to their name.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#65

Post by xiphias2 »

syntaxing wrote: post #18

Easy. A reasoning model with better performance than QWQ but at 21B (like Reka Flash 3) and good tooling call support. A model as “intelligent “ as Qwen2.5 but personality and creativity of Gemini (or Gemma at a minimum)

Something even cooler would be a model trained for 4 or less (1.33) bit weights instead of quantized after pretraining.

Math units are completely underutilized when I'm inferencing with batch size of 1, and post-training quantization under 8 bits loses too much of the precision to make a real difference compared to smaller models with higher precision.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#66

Post by BoredPositron »

Spivak wrote: post #43
BoredPositron wrote: post #36

Don’t kid yourself about the intentions. It’s only so enterprise customers can deploy a shitty chatbot on premise for their "secret data". The models will be free for commercial use until 1 mil turnover or something like that.

I mean for large enough buyers that was already an option, they wouldn't need to release a new model just for that.

Citation needed.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#67

Post by nyrikki »

dilippkumar wrote: post #26
mirzap wrote: post #19

OpenAI should release its frontier model as an open-weight model. There are already open-weight models that match OpenAI's best models (at the time of their release), so the idea that OpenAI would lose something by making its frontier models open-weight doesn't hold up. With an open-weight model, they would instantly kill any proprietary competitor, similar to what Google did to all its competitors with Android. IMO,…

> There are already open-weight models that match OpenAI's best models (at the time of their release) Did I miss something? Do we now have o1-pro level performance in an open source model?

Depends on your use case.

DeepSeek R1 is far better at producing maintainable, modularized code as a coding assistant as an example.

The big deal is that the distilled versions like DeepSeek-R1-Distill-Qwen-32B are good enough that anyone with a few old 1080 Ti's sitting around can run them and get most of the performance.

When you can run gemma3/qwq/DeepSeek-R1-Distill-Qwen-.../etc... you can easily switch models when one fails too.

And you have consistent performance that doesn't degrade over time, have the ability to avoid leaking prompt data between client etc...

It is all horses for courses though. For me o1-pro is roughly the same as o1 with just higher limits etc... but is still worse than o1-preview IMHO.

In my experience the few percentage points on synthetic benchmarks that o1-pro was claimed to have doesn't matter much in real world problems.

R1 pretty much matched o1-1217 on every benchmark and the distilled models like DeepSeek-R1-Distill-Qwen-32B only lost a tiny fraction.

A few months of o1-pro costs will get you a local usable model of GPUs if you are fine with ~20 eval tokens/sec.

But if o1-preview wasn't better for your use case than o1-proe...the calculus can change.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#68

Post by EGreg »

vessenes wrote: post #34

For those wondering how to answer "what do you want to see from an open model" I put this in: an open weights end to end multimodal model, a model large enough to act as a functional teacher along with a range of nicely distilled smaller sizes, code repo to make training / finetuning easy. As I write this, I'd also like to request a set of tool calling LLMs in various sizes. Feels to me like a small fast local tool c…

and guess what

they did announce they'll release open weights :)

https://x.com/sama/status/1906793591944646898?s=46&t=6NqVriD...

to be honest many of us never saw that coming... LOL

Re: OpenAI releasing new open model in coming months, seeks community feedback

#69

Post by wewtyflakes »

9rx wrote: post #63
wewtyflakes wrote: post #57

Earlier quoted context omitted.

This seems like a stretch. Another example may be, "If I am throwing a party, should I ask people that I do not invite, what they want?". Sure you could say that they could approximate the wants of the people who _are_ going, but why not just ask the people who are going, directly? It's just noise, otherwise.

> why not just ask the people who are going, directly? "If I had asked people what they wanted, they would have said faster horses" You don't poll people to find out what they want. You poll people to gather their ideas. Ideas that you can then leverage to deliver what your intended audience wants, even when they didn't know that they wanted it! If we assume this party you are throwing has 10 guests, you think you're…

I think asking 10 people instead of 8 billion people is a more reasonable way to discover the preferences of those 10 people. That is not to say there is not interesting information at the margin, but it is the margin.

Re: OpenAI releasing new open model in coming months, seeks community feedback

#70

Post by paxys »

mbrezu wrote: post #11

> Let's start with your details

In this case I don't see a problem. If you want your opinion to hold weight then put your name/background/credentials/reputation behind it. Not every discussion needs to be fully anonymous and upvote-based.
Post reply on HN