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

Grok 4.5

x.ai

Post reply on HN
Print view View original

411–420 of 1001 posts

Re: Grok 4.5

#411

Post by teravor »

NitpickLawyer wrote: post #88

(from Cursor's blog) > Training included trillions of tokens of Cursor data which capture a wide-range of user interactions with codebases and software tools. This dataset lets the model learn both from existing software as well as developer-agent interactions, capturing how developers work and how agents interact with their environments. This is what the big money was for. Cursor is the first big player that had rea…

    > You use the previous gen model to prepare datasets for the next model iteration.
you can also use a previous gen model to literally generate data for the next gen model. people used to believe that this is a bad idea but it turns out if you create a scaffold which sinks a lot of compute into generating and grading the data the quality turns out great.

Re: Grok 4.5

#412

Post by senraex »

michaelbrave wrote: post #335
FunHearing3443 wrote: post #316

Earlier quoted context omitted.

Studies on political bias in models consistently show that LLMs lean politically left. The only outlier is grok which leans right but by a smaller factor, according to this study for instance: https://arxiv.org/abs/2603.23841 Edit: adding some other studies that are easily retrievable with a quick search for those unsatisfied with the first one - https://arxiv.org/abs/2606.12922 https://arxiv.org/abs/2412.16746 Claim…

[flagged]

This is an incredibly dangerous and insane world view.

Are you really that ignorant to not understand that a form of this is at the heart of basically every atrocity in history?

Re: Grok 4.5

#413

Post by SmirkingRevenge »

anticorporate wrote: post #354
FunHearing3443 wrote: post #316

Earlier quoted context omitted.

Studies on political bias in models consistently show that LLMs lean politically left. The only outlier is grok which leans right but by a smaller factor, according to this study for instance: https://arxiv.org/abs/2603.23841 Edit: adding some other studies that are easily retrievable with a quick search for those unsatisfied with the first one - https://arxiv.org/abs/2606.12922 https://arxiv.org/abs/2412.16746 Claim…

You do realize that the scale on papers like this is the important part, and the creation of that scale is itself inherently biased? This particular study used a "conflict loyalties" approach - not necessarily a bad approach, but all it's really asking is when two values come into conflict, which one does the AI side with in its response? Conservative values tend to gravitate around perceived individual impacts, and…

I'm not sure asking the AI to support or oppose something is the kind of bias I would really worry about, unless those "opinions" degrade other kinds of queries.

I'd be more interested to see how well the AI's do when asked to assume a political view, and either steelman or debunk arguments

Re: Grok 4.5

#414

Post by alecco »

SirHackalot wrote: post #262
alecco wrote: post #159

Earlier quoted context omitted.

xAI > Gooogle & DeepMind I did not have this one on my 2026 bingo card.

That’s just not true. Google Brain/DeepMind came up with the attention mechanism to begin with… What a silly take.

Grok's latest model is objectively superior to any of the current Google models in the most relevant use cases. I don't like Elon musk but that does not change reality.

And Google came up with the Transformer architecture (2017 "Attention is all you need"). The Attention mechanism they based it on is from Bahdanau, Cho, and Bengio (2014, ICLR 2015). And there were many other self-attention variants by 2017. It was an amazing paper but let's not twist the story and give proper credit.

And not one of the people in that paper are still at Google, AFAIK.

Google has more compute, more data, and had the best 2 labs. And it seems they squandered it all. I'd blame their McKinsey CEO, the board, and management in general. It's a shadow of what it used to be. And it's a shame.

Re: Grok 4.5

#415

Post by hagbard_c »

carlob wrote: post #336
FunHearing3443 wrote: post #316

Earlier quoted context omitted.

Studies on political bias in models consistently show that LLMs lean politically left. The only outlier is grok which leans right but by a smaller factor, according to this study for instance: https://arxiv.org/abs/2603.23841 Edit: adding some other studies that are easily retrievable with a quick search for those unsatisfied with the first one - https://arxiv.org/abs/2606.12922 https://arxiv.org/abs/2412.16746 Claim…

a non-peer-reviewed preprint with an high schooler as the first author is the best citation you can come up with?

Appeal to authority is your best retort? It tends to not work out that well.

Re: Grok 4.5

#417

Post by gordian-mind »

otterley wrote: post #401
throwitaway222 wrote: post #393

Wait, do people now have X Derangement Syndrome. Like they can't even think or see the letter X?

Not wanting to be affiliated with Elon Musk or his business enterprises is not "derangement."

It's deranged to believe correctly naming something is "affiliation".

Re: Grok 4.5

#418

Post by davmar »

adamtaylor_13 wrote: post #407
davmar wrote: post #360

Earlier quoted context omitted.

[flagged]

A very quick search showed that USAID had extensive ethical compromise throughout the organization. Were we supposed to just subsidize that forever? It did some good, yes, but from what I can tell, it was better to shut it down. That's just one example, but frankly I get the feeling that if I dug into more examples they'd end up the same: easily explained and not entirely shocking.

[flagged]

Re: Grok 4.5

#419

Post by shangofox »

solarhoma wrote: post #317
jspaetzel wrote: post #309

Earlier quoted context omitted.

How's this going with the rest of the models?

My immediate thought as well. Every other AI platform has very left leaning guardrails installed. Grok is the only AI platform that has been shown to be center leaning.

There's been studies that actually show most AI platforms have right leaning bias in circumstances. It's definitely not "left". And Grok isn't center if Elon Musk's bias is involved

Re: Grok 4.5

#420

Post by SmirkingRevenge »

jjcm wrote: post #295
jesse_dot_id wrote: post #291

I just don't think that I can ever trust an xAI model knowing that they are actively trying to shape its replies to fit a political narrative. How can you trust their models to be reliable in a business setting with the foreknowledge that their models are being nudged around in the backend?

Depends on the domain imo. I work on a design tool - I don't think their political narrative will affect my work.

[flagged]
Post reply on HN