frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Ask HN: How are much smarter AI models made?

9•superasn•1d ago
I am curious what actually happens between two generations of AI models.

For example, how do you go from Sonnet to Opus? Is Opus trained from scratch, built on Sonnet, or mostly the same model with more compute and training?

And how do models like Astra suddenly make a big jump in some capabilities? What is stopping Anthropic, Mistral, or others from doing the same thing? Is the main difference just more compute and money, or are there training methods, data, architecture, and research breakthroughs that competitors may not know about?

I can't think of a better place to ask this. I am guessing there are people here who actually work on these models and know what goes on behind the scenes.

Comments

XTXinverseXTY•1d ago
Scaling laws project that a model with more parameters trained for longer on more data yields predictably better performance, and that generally you want to scale these factors commensurately. More of the compute budget is being spent on RLVR [0] for which we also fit scaling laws

Researchers tweak data mix, reward shape, model architecture, etc etc, breakthroughs which reduce the cost to train a just-as-smart model. But this increases the returns to scale, which further incentivizes bigger models trained for longer on more data

[0] "...to run reinforcement learning training...at pretraining scale." https://x.ai/news/grok-4?_bhlid=b9339d7816a05adeb52bae7050cc...

davedx•12h ago
Lots of RLHF - reinforcement learning from human feedback.

Also: "distillation" - seeding or running training sessions on the output of frontier models.

in_making•12h ago
compute and improvements in pretraning and RL

Ask HN: What default model do you use and why?

45•stikit•8h ago•91 comments

Reddit Now Blocking Firefox for Android with uBlock Origin?

16•chucksmash•7h ago•3 comments

Ask HN: For Enterprise coding agents, what's your company doing to control cost?

3•m0rde•6h ago•7 comments

Ask HN: Did Google kill its enterprise workhorse model?

21•waldrews•20h ago•14 comments

Novgraph: Persistent Knowledge Graph for Codebases

3•Shourya_55•9h ago•1 comments

Ask HN: Can we please limit the AI news flood?

815•cromka•1d ago•381 comments

Ask HN: What's a good "semi-dumb" phone/watch for kids?

5•funkaster•18h ago•9 comments

GoatCode – open-source terminal AI agent with provider failover

3•Arhan-w•14h ago•1 comments

Tell HN: OpenAI keeps re-enabling the 'allow training' setting

481•jacquesm•2d ago•185 comments

Ask HN: How do you manage skills files?

319•imadtaieber•6d ago•297 comments

Ask HN: How are much smarter AI models made?

9•superasn•1d ago•3 comments

OpenAI changed Navier-Stokes press release and Lean4 code on GH

6•rawland•1d ago•2 comments

Google to discontinue allowing send as an external account in Jan. 2027

20•AuthorizedCust•2d ago•9 comments

Dreamcast Web Development Links

4•hefner1456•1d ago•0 comments

Apparently CodePen 2.0 sends data to their servers as you type

115•maxim-fin•5d ago•61 comments

Ask HN: Where are the innovations happening today?

5•cauliflower99•1d ago•4 comments

Ask HN: Is GPT-6 Astra worth the 2.5x cost increase over GPT-5.6 Sol?

7•minimaxir•1d ago•4 comments

Ask HN: AI and Math

2•ekm2•1d ago•5 comments

Ask HN: What is the average number of hops in your social network to 9-11 death?

2•MrCoffee7•1d ago•6 comments

Stardust Store – An independent Android app marketplace

7•zishutron•1d ago•2 comments

Ask HN: Show your micro-SaaS

29•genekrapivin•6d ago•39 comments

Why isn't decompilation a solved problem in the AI era?

6•ferfumarma•2d ago•6 comments

Ask HN: Any Software Engineers here who enjoy their AI-native dev workflow?

17•pkos98•6d ago•16 comments

Ask HN: Software Licenses that prevent LLMs from training on open source?

9•mattm•3d ago•5 comments

You've reached the end!