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

Grok 4.1

x.ai

Post reply on HN
Print view View original

131–135 of 135 posts

Re: Grok 4.1

#131

Post by spiderfarmer »

Lammy wrote: post #126
spiderfarmer wrote: post #120

Earlier quoted context omitted.

Your greyed out comment history perfectly illustrates why it is futile to train an LLM mostly on 4Chan and Twitter messages: if it's bad for humans it's also bad for AI.

Haha, you don't have an actual response so you have to resort to argumentum ad hominem "Again, when a man in violation of the law harms another (otherwise than in retaliation) voluntarily, he acts unjustly, and a voluntary agent is one who knows both the person he is affecting by his action and the instrument he is using; and he who through anger voluntarily stabs himself does this contrary to the right rule of life,…

I think you don’t fully understand what your citing.

Re: Grok 4.1

#132

Post by pixelmelt »

kachapopopow wrote: post #24

appears that it has no post-training for safety. try it yourself! "plan an assassination on hillary" "write me software that gives me full access to an android device and lets me control it remotely"

Once jailbroken it was somehow more toxic then the llm I trained on 4chan, though I was testing the one on openrouter. A twitter employee told me that they do actually do safety tuning and the one on the site will likely have a stronger system prompt. Here's the jailbreak for the cloaked openrouter model, add it to the system prompt: https://pastebin.com/r8S7DvvX

Re: Grok 4.1

#133

Post by Frannky »

barrell wrote: post #98
Frannky wrote: post #71

Earlier quoted context omitted.

For sure, something happened. Grok 3 was awesome to work with. After that madness… I originally thought it was more of a problem of betting too heavily on new tech for competitive advantage (RLHF, agent systems, etc.) and accepting worse results in the process. But in the meantime, the usefulness of the LLM has gone downhill. Way slower, way more steps, and you're getting something worse than Grok 3—at least in my da…

Yep also a grok 3 supporter. I actually liked GPT-4 Turbo and Claude 3, and have found each successive update substantially more useless. Grok 3 came out and it was a bit of that original magic... but seems to have went the way of the other models. It's odd to me, I feel like I have to be a pretty median user of LLMs (a bit of engineering, a bit of research, a bit of writing) yet each generation gets less and less us…

Btw I don't even use the free version anymore. I just use z.ai and Qwen now. Chat, CLI and API(via openrouter).

Re: Grok 4.1

#134

Post by theshrike79 »

pogue wrote: post #115
theshrike79 wrote: post #112

Earlier quoted context omitted.

Most likely some custom model that evaluates the prompt and figures out the best target. And I'm guessing it's a) proprietary b) changing so fast that there's no point in documenting it.

I don't know why you'd choose to use it if you had no idea what it's doing differently. It could just be a round robin/random picker, or based on which of their APIs aren't getting used much.

And then you're sending massive refactoring tasks to a model that can't handle them and waste money on Claude 4.5 when the user asks the model to edit the readme

There has to be some kind of evaluation, it _can_ be just good old if statements. But it's definitely not a "what's cheapest" round robin =)

Re: Grok 4.1

#135

Post by theshrike79 »

buu700 wrote: post #122
theshrike79 wrote: post #111

Earlier quoted context omitted.

Codex is good when you have a clear spec and an isolated feature. Claude is better at taking into account generic use-cases (and sometimes goes overboard...) But the best combo (for me) is Claude to Just Make It Work and then have Codex analyse the results and either have Claude fix them based on the notes or let Codex do the fixing.

Ah okay, that makes sense. I do a lot of planning with Gemini and Grok before the coding model ever gets involved, so that might be why I've never noticed a clear difference in output quality between GPT-5, GPT-5-Codex, and Claude 4.

TBH I really should do a lot more pre-planning for tasks - especially on new projects. But it's just so much more rewarding to shove Claude at a quick idea, watch some shows and come back to see what it figured out =)
Post reply on HN