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

341–350 of 696 posts

Re: Grok 4.6

#341

Post by porridgeraisin »

Manfrednotfunny wrote: post #280
LarsDu88 wrote: post #260

Earlier quoted context omitted.

They will get sharply better in tasks with verifiable domains... math and coding Gradually the labs will start engineering verifiable sandboxes for wider domains like videogames This strategy will hit a plateau in about 18 months and then we're back to diminishing returns and incremental progress along other dimensions (like accelerated inference using ASICs)

You only mention math, coding and videogames. They already hire and pay people with research titles for creating and solving problems in their fields. And a lot of labs say that RL can help everywere and has plenty of way to go.

Correct. You can look at the AI tutor jobs in the job board of any of these companies.

Re: Grok 4.6

#342

Post by anvuong »

verdverm wrote: post #243
LPisGood wrote: post #215

Earlier quoted context omitted.

System prompts are more like suggestions than hard constraints.

I don't understand why they don't look for large substring matches for the system prompt before returning the response. Trivial calculation compared to a system prompt instruction asking the model not to do it

because you can always make up your own language and ask the model to use it, no filters would catch that

Re: Grok 4.6

#343

Post by vrganj »

numpad0 wrote: post #337
vrganj wrote: post #313

Earlier quoted context omitted.

> CSAM is by definition limited to real imageries and cannot be generated. Where in the definition does it imply this?

I thought that's just the legal definition in any sufficiently developed countries?

I'm not sure what you're trying to imply, but we sure are moving away from a by definition.

Re: Grok 4.6

#344

Post by ardivekar »

behnamoh wrote: post #305
tintor wrote: post #290

Earlier quoted context omitted.

Maybe because frontier labs buy the same RL tasks from task producer companies.

Who are these task producers? Are you saying that Anthropic, et al delegate the RL part to third party companies that do it for pretty much every other AI company as well?

Turing, etc. And yes.

Re: Grok 4.6

#346

Post by QuadmasterXLII »

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…

I suspect that because each RLVR episode injects ~1 bit into the models capabilities, and training on a reasoning trace injects ~megabyte into a models capabilities, distillation is powerful enough right now that they’re all basically the same model

Re: Grok 4.6

#347

Post by Amezarak »

gazebo2 wrote: post #288
ralusek wrote: post #127

Earlier quoted context omitted.

Can someone help me understand the deep fake controversy? That's like making photoshop illegal.

It was always possible to modify images to produce inappropriate or insensitive content, but plugging a turbocharged state of the art image generator with virtually no guardrails into every Twitter reply and then failing to address the issue long after it was obviously being used for CSAM or deepfakes of real people against their will.. well that's worse

Notice the Wikipedia link says the problem was "put her in a bikini." The claims about "Grok just lets you undress people" were massively exaggerated because people hate Elon (perhaps for good reason) and not worse than other models.

Re: Grok 4.6

#348

Post by Amezarak »

cheesecakegood wrote: post #245
arrosenberg wrote: post #67

Earlier quoted context omitted.

Not the person you are responding to, but the fact that Grok is being used to generate a ton of CSAM and pornographic deepfakes isn't great!

It’s pretty telling that almost all of the bullet points in the system prompt that was posted for Grok have to do with preventing criminality and CSAM generation. No other provider has this same issue at that scale. The first-order-thinking reaction is “oh cool, look how they don’t want it to happen” but the second-order reaction is “why does this company have such a problem when others don’t?” It’s their own tactics…

> No other provider has this same issue at that scale.

What gives you that impression?

Re: Grok 4.6

#349

Post by dmix »

lucisferre wrote: post #335
dmix wrote: post #172

Earlier quoted context omitted.

These system prompts are not the only safety layer that these models use. There's other more deterministic filters in place both on input and (streaming) output.

I think it is fair to argue that prompts are not a safety layer at all and can't be relied upon for much. "Make no mistakes"

Yes that's been obvious since the beginning. That's why you should always monitor your agents closely. Just like supervised self driving cars, you have to watch the road and do some hand holding.

The tooling around isolation, logging, and real time security/anonomly detection for regular LLM laptop users is very immature right now. I expect that to change soon.

The alternative is extremely locked down models which is what Anthropic seems to want to do.

Re: Grok 4.6

#350

Post by Amezarak »

nomilk wrote: post #43

Tangental, but has anyone else noticed grok's voice mode got stupid and terse ~2 weeks ago? I've absolutely loved grok's voice mode since it came out (incredibly useful for brainstorming on walks and helping conceptualise and get the verbiage for expressing ideas) but it seems so have lost about 40 IQ points recently, and if the question is multi-part, it often answers just one part with no elaboration or explanation…

No, I noticed this too. Voice mode was great at providing detailed responses, although I wished it would have dialed the talkiness down just a tad. Then recently it suddenly got very terse, way too terse, but also latency went way down. Voice usage also really burns through your total allowance now.
Post reply on HN