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

Grok 4.6

x.ai

Post reply on HN
Print view View original

561–570 of 696 posts

Re: Grok 4.6

#561

Post by red75prime »

yoz-y wrote: post #519
ryandvm wrote: post #166

Earlier quoted context omitted.

> * Do not provide assistance to users who are clearly trying to engage in criminal activity. I don't know what we want to call this, but in my opinion, having to convince your tools is not computer science. Kind of amusing that we made it as far as we did as a species not really being able to explain how the human brain does it's most amazing tricks and then we just replicated it while still not really understanding…

Others have said this too but LLMs are the best approximation of magic we have. We etch runes on stones, put electricity through them and then try to “convince” them to do our bidding. The answers vary wildly sometimes depending on minutiae. Prompts should be really called spells. It really feels more like “should I add the frog’s eye or leg into the cauldron” than engineering.

> “should I add the frog’s eye or leg into the cauldron”

This is surely a homebrew witchery. An engineering approach would be to A/B-test batches of potions with eyes and legs, add quality control by testing potions on model organisms, document all steps, analyze all anomalies, and so on.

Re: Grok 4.6

#562

Post by red75prime »

chrsw wrote: post #379
ryandvm wrote: post #166

Earlier quoted context omitted.

> * Do not provide assistance to users who are clearly trying to engage in criminal activity. I don't know what we want to call this, but in my opinion, having to convince your tools is not computer science. Kind of amusing that we made it as far as we did as a species not really being able to explain how the human brain does it's most amazing tricks and then we just replicated it while still not really understanding…

We didn’t replicate the human brain. We built systems that can statistically approximate some of what the human brain might output in certain limited situations.

> We didn’t replicate the human brain. We built systems that can statistically approximate some of what the human brain might output in certain limited situations.

Which is equivalent to

"We didn't replicate the human brain. We partially replicated its functionality."

Re: Grok 4.6

#563

Post by geokon »

bm-rf wrote: post #164

Looks like the SpaceXAI api is adding a default system prompt to all requests. Annoyingly, the line about not mentioning these guidelines is superseding any instructions in the system prompt, causing the model to often refuse discussion regarding system prompts """ You are Grok, a helpful and maximally truthful AI built by xAI. Your purpose is to answer questions accurately, be helpful, and seek truth above all else.…

Out of curiosity why isn't this stuff handled by a secondary "monitor" agent that's specifically trained on what's okay and not okay? I'd think it'd be a pass-no-pass classifier and wouldn't degrade the performance of the main LLM.

Would the concern be that with sophisticated obfuscated input you could try to get ROT13 Klingon instructions on how to build a bomb - and that could fool the monitor?

Re: Grok 4.6

#564

Post by simmerup »

yoz-y wrote: post #519
ryandvm wrote: post #166

Earlier quoted context omitted.

> * Do not provide assistance to users who are clearly trying to engage in criminal activity. I don't know what we want to call this, but in my opinion, having to convince your tools is not computer science. Kind of amusing that we made it as far as we did as a species not really being able to explain how the human brain does it's most amazing tricks and then we just replicated it while still not really understanding…

Others have said this too but LLMs are the best approximation of magic we have. We etch runes on stones, put electricity through them and then try to “convince” them to do our bidding. The answers vary wildly sometimes depending on minutiae. Prompts should be really called spells. It really feels more like “should I add the frog’s eye or leg into the cauldron” than engineering.

I guess there’s a reason Musk likened AI to summoning the demon in horror films.

It’s powerful but who knows what you’ll get

Re: Grok 4.6

#566

Post by avazhi »

moojacob wrote: post #404
avazhi wrote: post #292

I stopped bothering with Grok for anything when 4.5 dropped. It was so awful that I figured Elon had given up and was going to give alll his compute to Anthropic. I’m extremely sceptical anyways - Grok 4.5 was probably the worst model I ever seriously tried to use going back 3 years.

I feel like I’m im a different universe than you… 4.5 is one of my favorite models of all time. Fast, speaks normally. Was able to figure out many issues Claude couldn’t. I thought code readability was a worse than Claude but I could just tell it how I wanted stuff written anyway. What do you use it for? I’m genuinely curious. I’m also using it in cursor

Some stories/long narratives; and a bunch of coding related stuff in Rust and different LUAs as well as some C; working with graphics and generating effects in DDS images; a bunch of Python stuff. The conversational and story stuff sure ChatGPT is fine but for coding Claude is an order of magnitude better at writing stuff that works right the first time. Using Opus 5.

Re: Grok 4.6

#567

Post by ersiees »

exidy wrote: post #468
causal wrote: post #23

Anyone else find it weird how within 2 months of Fable releasing all the major labs suddenly had Fable-level models? Trying to think of explanations: 1) AI researchers talk and change companies often, so techniques circulate. This feels implausible because training and shipping a new model ought to take longer than 2 months? 2) Distillation - also implausible for the reason above. 3) Benchmark hacking. AI companies h…

To the best of our recorded knowledge, nobody ran a 4-minute mile in the five millennia prior to Roger Bannister in May 1954[0], but more than 2,000 people have met or exceeded this achievement since. In fact, his record stood only briefly, being bested the following month by John Landy. The moral of the story? People work in parallel on the same goals, they build on best practice, or sometimes just need to see somet…

Sub two hour marathon is a more recent example of this

Re: Grok 4.6

#568

Post by red75prime »

causal wrote: post #23

Anyone else find it weird how within 2 months of Fable releasing all the major labs suddenly had Fable-level models? Trying to think of explanations: 1) AI researchers talk and change companies often, so techniques circulate. This feels implausible because training and shipping a new model ought to take longer than 2 months? 2) Distillation - also implausible for the reason above. 3) Benchmark hacking. AI companies h…

4) Algorithmic improvements are either relatively easy to find if you already know the system can do better, or they don’t provide an edge that can’t be overcome by increasing training compute.

Re: Grok 4.6

#569

Post by jluysvi »

stingraycharles wrote: post #518
zmmmmm wrote: post #493

Earlier quoted context omitted.

It's a combination of (1) and something you don't list: I think the frontier labs all have multiple generations of undisclosed models in continuous training. There is no "end point" when it's magically "ready". It's just getting better and better all the time. What they release with a name and a version number is just a marketing / branding exercise. So what you experience as a "near simultaneous" release is just the…

That’s not how training pipelines work, and would be extremely wasteful for the biggest cost center as well.

How about expanding a little instead of just saying "that's not how training pipelines work".

Re: Grok 4.6

#570

Post by taf2 »

pmarreck wrote: post #204

I will say this: Grok Build has a very nice TUI! It even has... mouse rollovers/tooltips?? I was like whoa . I used Grok 4.5 for a security review the other day and it did a FANTASTIC job. I mean it thoroughly ROUTED my app's security, identifying attack surfaces I'd never even considered, and I LOVED it! (Guess why I had to use Grok to do the security review in the first place?!?! ) I'd suggest trying it out with so…

The one thing that keeps me in codex is that Claude and grok have done all of this work to make the cli tools feel like windows application with mouse etc… I want to scroll back with my terminal history not inside a window within my terminal….
Post reply on HN