The Silver Fern

Wait a week (or so) and evaluate Grok 4.

They are all very similar at this point (the top models). Claude is the best for coding, some are better for visual stuff, but for pure text stuff you won't see a huge difference so you can be driven by price.

The Grok team iterate faster. OpenAI are the 'leaders' right now, but with the exodus of talent and these huge data centers their competitors are setting up I'm not sure how long that will last.

I'd recommend picking a monthly plan over a yearly one so you can swap as you need to.

Nice hands for some of those trys. Didn't really have to get out of second gear to put 50 on them.

@Old-Samurai-Jack said in 2025 All Blacks v France series:

Geez, it is just a circle-jerk of butt-hurt provincial bias against Blackadder, isn't it?

It's the opposite really. If he played for the Highlanders he would be nowhere near the All Backs. Not the first or the last to be selected purely on club mana opposed to individual ability.

@reprobate said in 2025 All Blacks v France series:

@Kirwan said in 2025 All Blacks v France series:

Well that was a methodic dismantling of the Blackadder myth. Confirmed beyond all doubt the "busy but ineffective" label.

As a ball carrier, sure. He's almost McCaw-like in his lack of metres and evasiveness.
Is a ball carrier what we need to accompany and balance Sititi and Savea?

I don't care if he's a ball carrier, I want his ruck clear outs and tackles to be effective.

Who cares if he makes 15 tackles if he's escorting guys over the try line? Espcially at six, you need stoppers there. Obviously Kaino and Collins used to cut people in half, but I could mention another six that tackled harder than Ethan.

Reminds me when I started work on a helpdesk, I had awesome call stats because I took all the easy items off the queue. Not actually that useful but looks great on paper.

Well that was a methodic dismantling of the Blackadder myth. Confirmed beyond all doubt the "busy but ineffective" label.

@Bones said in Artificial Intelligence (Previously "Chat GPT"):

@Kirwan yeah we've got an internal gpt that's typically better, but can't produce docx - I'm sure I can probably get a bot sorted if it's gonna be a regular thing, was just annoying when I'd prompt it to add something and it would change two other things and drop a third.

Yeah ok, I've been playing with Task MCP servers for that. There's one called TaskMaster and another called Shrimp-Tasks. These help it plan, save the steps into a file and keep it's head straight about what it's doing. Funny to watch it argue with it self with thinking enabled.

I have a more optimistic view than that. The amount of productivity about to be unleashed is hard to comprehend.

Add in a limitless workforce on the horizon with robotics shits about to get weird.

Truely think we about to enter an age of abundance. But the people aren’t ready for the disruption we are about to experience as we transition.

@Bones look at MCP for Context7, might help with syntax with the right docs in context.

And for that sort of work, you should have custom instructions to guide it so when you do ask it you don’t have to argue with it.

I mean the 1950 style teacher classroom model should be completely reformed.

1:1 AI Tutors shepherded by far less teachers.

This model is being tried in Texas and they already are having great results.

And, no, AI is not easy to spot. Lazily used AI is. I could set up a model for a kid that would be indistinguishable from any of their other work in a few weeks.

But it would take them minutes to generate the work.

The way people are assessed needs to change.

The top models are ticking off exams pretty easily now.

Take a look at SuperWhisper, I’ve started playing with the free model. Pretty accurate.

Some people are battling the new CLIs, raised a smile

Isn’t school supposed to be preparing you for a career? These tools are going to be used everywhere, education needs to reform completely.

I had three AI bots working as a team this afternoon, one searching for a root cause of an issue so I could resolve it.

Another reviewing pull requests from the team, and another speeding me along writing code.

They are increasingly becoming autonomous and if used properly, undetectable. The kids will work out how to bypass any checks, and like with most cheating you’ll just catch the dummies.

@gt12 said in Artificial Intelligence (Previously "Chat GPT"):

@antipodean said in Artificial Intelligence (Previously "Chat GPT"):

And in another criticism - I got chatgpt to review a report I uploaded. I asked it questions and had to continually prompt it and fact check. When told to restrict its answers to the report, it made shit up and offered the excuse it must have got the answers it provided from other sources.

I wonder how many people are taking what these models are saying at face value.

Huge, huge, issue for us in education.
AI slop is everywhere.

You guys need to transition to AI Tutors ASAP. I won’t hold my breath

@antipodean you’d still get hallucinations with a local model. It’s baked in.

You need the massive data centres and the top models to reduce it. I’ve noticed that MCP task trackers help enormously, helps it keep track of its chain of thought.

Cullen Grace? Hell no.

Was bad enough when people were talking about the woman beater coming to the Blues.

I’ll take the up and coming talent in the region over never was/has beens.

Is that the free chatGPT? Wife was just saying the same thing about it, I don’t use it anymore.

Claude-4, Grok and Gemini-Pro 2.5 are the best models.

And yes people are definitely just taking it at face value. It makes stuff up confidently.

So one more injury and we get Sotutu? Where did I leave my voodoo dolls?

Jagged intelligence is a very apt description. Incredible results mingled with utter retardation.

The reasoning models that cross check are better, but the trade off is speed.

New voice is coming soon too.

Have you tried the vision? I got a bunch of stuff out of the fridge and chucked on the camera for and we discussed what we could make with it.

When this shit is in AR glasses and off a phone it’s going to be crazy what we can do.

Only one specialist 8, two specialist wingers. Not a fan of utility players all over the place.

Same as 10s playing 15, they fixed fullback now that maliase is spreading around the team. Locks that aren't that great at 6 covering six.

A giant cup of meh.

Need to a CEO that can mend some fences that Robinson kicked over.

@No-Quarter said in Artificial Intelligence (Previously "Chat GPT"):

I am finding Grok extremely useful for sifting through the bullshit on X at the moment. You see a political commentator (from either side) make some claim, then in the replies someone simply says "@grok is this true"? and it gives a big rundown of all the nuances and why said commentator is either exaggerating their claims, or straight up lying (e.g. using a video from years ago claiming it is current etc). I know it is not perfect but it's a damn sight more accurate than what a lot of the grifters are putting out there.

It's very good at showing both sides, even if it doesn't agree.

The next update is focused on improving the quality of sources for that, as it still has a bias based on the media sources it's using.

It's great at threads where you want to dive in a bit, or don't know a term. A very useful feature.