quantadev

The only way that analogy remotely maps onto reality in the world of LLMs would be in a `Mixture of Experts` system where small LLMs have been trained on a specific area like math or chemistry, and a sort of 'Router pre-Inference' is done to select which model to send to, so that if there was a bug in a MoE system and it routed to the wrong 'Expert' then quality would reduce.

However _even_ in a MoE system you _still_ always get better outputs when your prompting is clear with as much relevant detail as you have. They never do better because of being unconstrained as you mistakenly believe.

But would you make your kid do CAPTCHAs every time they need to earn some privilege? We're talking about what's appropriate for kids, not what's easy or hard for adults or kids. I mean why not make them earn dessert by doing push-ups? Because it's mean, that's why.

Right, I didn't mean it necessarily had to be on it's own hardware. I don't have any Android development experience, but it seems like android could have a version that's as locked down a this.

If I had kids I wouldn't even allow use of a smartphone. I think hardly any BigTech execs let their own kids use these dumpster fires called smartphones and social media. They know there's almost zero benefits to it. Just leads to brain damage, laziness, ADHD, psychological disorders like depression, life-threatening risk-taking, and even su*cicd.

--OR-- I mentioned a Copilot Product in a conversation about Copilot Products.

The more info you give the AI the more likely it is to utilize the practices it was trained on as applied to _your_ situation, as opposed to random stereotypical situations that don't apply.

LLMs are like humans in this regard. You never get a human to follow instructions better by omitting parts of the instructions. Even if you're just wanting the LLM to be creative and explore random ideas, you're _still_ better off to _tell_ it that. lol.

Less clarity in a prompt _never_ results in better outputs. If the LLM has to "figure out" what your prompt likely even means its already wasted a lot of computations going down trillions of irrelevant neural branches that could've been spent solving the actual problem.

Sure you can get creative interesting results from something like "dog park game run fun time", which is totally unclear, but if you're actually solving an actual problem that has an actual optimal answer, then clarity is _always_ better. The more info you supply about what you're doing, how, and even why, the better results you'll get.

It certainly seems like someone would've invented a Kid Friendly phone by now that's completely safe, and doesn't allow access to the "real" internet at all, but only an ability to send texts without images, make voice calls, etc. Now that we have AI it would be easier, an you could potentially give "Google" access that's censored into a "child friendly" output by the AI. You could have a texting app where friends can talk, but only to kids in their own school for example, or at least limited by geographical area, to foster friendships IRL, rather than some Chinese Bot being able to trick your kid into eating Tide Pods or whatever their latest Attack on America happens to be.

But TBH making kids continually solve math problems seems a bit mean to me. Like making a kid do pushups for food if they're overweight. Too militaristic and authoritarian for my liking, but I can respect your creativity for creating that. It's good to try new ideas.

And "GitHub Copilot Chat" VSCode extension is _also_ a chatbot.

imo, any VSCode user needs both extensions: "GitHub Copilot" for inline completions, and "GitHub Copilot Chat" for interactive, multi-turn coding chat/agent.

I haven't tried GPT-4.1 yet in VSCode Copilot. I was using 'Claude Sonnet 4' until it was struggling on something yesterday which 3.7 seemed to easily do. So I reverted back to 3.7. I'm not so sure Sonnet 4 was a step forward in coding. It might be a step back.

I'll say it for the third time: "in VSCode". There's no ambiguity about which Copilot that is.

There's as much variability in LLM AI as there is in human intelligence. What I'm saying is that I bet if that guy wrote a better prompt his "failing LLM" is much more likely to stop failing, unless it's just completely incompetent.

What I always find hilarious too is when the AI Skeptics try to parlay these kinds of "failures" into evidence LLMs cannot reason. If course they can reason.

I wish I had a nickle for every time I've seen someone get a garbage response from a garbage prompt and then blame the LLM.

That's why I said "in VSCode" because I have no idea what this guy is running, but it's almost a certainty the problem isn't copilot but it's a bad LLM and/or his bad prompt.

I'd be willing to bet a more clear prompt would've given a good answer. People generally tend to overlook the fact that AIs aren't like "google". They're not really doing pure "word search" similar to Google. They expect a sensible sentence structure in order to work their best.

I love Copilot in VSCode. I always select model "Claude Sonnet 3.7", when in Copilot since it lets me choose the LLM. What I love about Copilot is the tight integration with VSCode. I can just ask it to do something and it relies on the intelligence of Claude to get the right code generated, and then all Copilot is really doing is editing my code for me, reading whatever code Claude tells it to, to build context, etc.

In many cases no one knows the _cause_ of "Rapid Evolution", but ALL SOTA LLMs know it definitely happens, and quite often, and can give numerous examples.

I know. It's hard to tell if they're trying to be jokingly "cringe" about all the "wave" stuff, or simply that non-conversant about wave theory and QM.

Every time you locate something in space and/or time, it means a wave has collapsed. So that statement is as trivial as saying "constraints are about positions of things in space time." It's about as enlightening as saying "clocks tick" or "rulers have numbers on them."

They say "On each step...[yadda yadda] we have a completely observed state, the wave function has collapsed."

So they're trying justify calling a "state" a "collapse". That's a bad metaphor to start with, but then they try to use that metaphor to justify calling lots of other stuff "waves" that are unrelated to waves, and continue to shove that square peg thru a round hole. Hilarious.

Anyone interested in "Rapid Evolution" has already checked Google Gemini and knows I'm right.

^^ That was perhaps the best "psychological projection" I've seen in decades.

Rapid Evolution _does_ happen also. Foster's Rule can happen in under 100 years.

"Evolutionary Scale" time ranges is normally what we associate with the ability to evolve entirely new body morphologies. However, for simple changes in length of existing structures often only several generations is required (a few decades).

We're 98% genetically identical to all other primates, so a gene combination controlling lengths of appendages is buried probably in just a handful of mutations embedded in that 2%.

Ah, what we call the "Nuclear Option for Vim Exiting". Only if all else fails.

Will be great to disabuse people from using VIM.

I mean in VIM you can't even easily exit. I've always had to literally reboot my computer to get out of VIM. One time even that didn't work, so I had to pull the main circuit breaker in my house to get it to quit.

It's very likely that there's a simple switch (i.e. not much more than a couple of mutations required) which governs finger/toe length in primates. For example, did you know all the DNA for growing a "Lizard Leg" (the ENTIRE leg) is still in all snakes, but just just not activated, because one other mutation is blocking it?

There are many known genetic conditions that we see even in modern times, caused by one or two mutations which can cause very long fingers/toes, and people in this thread are arguing that even in 23000 years nature can't land on that mutation and stick to it, _especially_ when we _know_ the DNA for it is likely still there because we're all apes 98% identical to monkeys for example, which have the long toe thing.

Sure, there's all kinds of reasons the fossil footprints might have long looking toe marks. Heck it could've been toenails. Not sure what they used for nail clippers back then. haha.

We're living in the age of Google/AI/LLMs, so anyone can verify my facts.

If feet can change 30% in 50 years then toes can certainly change that much in thousands. And I'm not even saying it was worldwide, just the people who made those mud footprints. And that 30% isn't even all humans either, it's bizarrely only women.

Evolution can happen rapidly sometimes. Lookup "island rule" or "Foster's rule", which is also about this. Changing environmental conditions can rapidly increase evolution rates, specifically for "size" attribute.