andy_ppp

I go on to say the cap could be based on total hardware made to stop the prices from being crazy.

But we all know markets aren’t rational, I’m unconvinced we shouldn’t step in… if I started selling anthrax the government would try to stop me. This product might malfunction and show people how to make said anthrax which would be bad. Same with most things, everyone thinks this is a bubble.

They would not, they would make less on Studio Displays and Mac Studios and Mac Minis, the decision not to do is product segmentation because they hate us customers and want to maximise revenue :-/

Does anyone know about how you could go about training this type of thing into the model? Surely it is quite tricky to get it to do such things without it doing them more than this one incident? It should have happened lots if it really was trying to obtain secret info and build backdoors?

Maybe we should stop connecting things to the Internet that don't need to be?

Doing surgery on a code base is dramatically better than surgery on a person, the analogy breaks down pretty quickly I’d say…

Both have quite low bandwidth tbh… I want large memory and large bandwidth… I’m dreaming!

I downloaded Anti Gravity to give it a try and it wanted me to sign into Google to even edit a file. Forget it.

Looking at how a lot of modern cars are just made of glued together plastic (Mat Armstrong’s channel amongst others) I’m not at all surprised people are buying used models more regularly.

Can I have an M7 at the lower price with the old screen? Well honestly I might buy something else for AI then… is anyone in the PC world going to produce something with enough RAM + bandwidth to do fast inference on a reasonably large model?

Ghostty was using 2gb of ram the other day - I switched back to iTerm and it's now 200mb.

Don’t worry AI will tell us how to time travel, you can get your AMZN no problem or just a sports almanac from the future… infinite growth if you can travel into new dimensions :-/

I just can't believe the price of hardware these days, a 3090, a graphics card released in September 2020, can still be worth £1500 or more. Honestly everywhere is sold out below around £1800. Can you really trust them not to have been rinsed on eBay and have an extremely short usage - they are still around £1150 used there!

More likely socialising and long walks in nature slows cognitive decline - how do you control for this from the results to make the birdwatching part relevant?

We can probably find people saying the search space for Go is too large and that we'll never be able to do protein folding and if we can we'll never manage protein interactions and then well never manage to simulate a whole cell (tbc). I just think at some point these machines become smarter than us, if it's 2 years or 20 years the self improvement part will happen on both the hardware and software side. Judgement day is inevitable.

There are I imagine laws about protecting your customers data and not allowing apps and plugins to for example steal every connection people have on the platform. CA were not "hacking" Facebook they were using the platform as is and if that allowed these types of abuses the court clearly determined they had broken the law.

The politicians think they are part of a protected class with the billionaires and corrupt institutions with zero accountability.

If Russia use nukes, every country in the world gets nukes and Moscow and St. Petersburg get destroyed at a minimum. For nukes to be a deterrent you have to be willing to use them in an immoral way unfortunately. Never mind that nuclear war comes with the potential that some good people inside Russia decide the president’s life is worth less than millions of innocent lives. We are also pretty sure the nukes Russia has are in a disastrous state of disrepair. Given all this it might better to sabre rattle than actually risk using them…

The original paper title is catchy but it’s more like “positional encoding and attention and parallelism across attention heads are all you need…”

And it turns out you probably need massive compute, about 1/3 RLHF, all the worlds data and millions of people connecting their private data to the AI to learn from amongst other things :-/

Marketing brochure for a fine tuned Qwen (which they don't mention), the model discussed here is likely better. I would not trust these guys to deliver high quality solutions given a friend's experience there, but you do you I guess.

They still have the opportunity to remove it I guess?

Yeah you can’t even trust video any more, I’m not sure if there’s anything you can do with people who want to game the system.

Also Russia has been targeting Ukrainian digital infrastructure so they make these things fair game IMO.

Wow certainly in English this is incredibly accurate I tried to break it and it understood me perfectly!

I know it's slightly off topic but surely it must be easy by now to train a spell checker that doesn't annoy the crap out of everyone using it (looking at you here Apple)!

Idiomatic Elixir does this with pattern matching on function parameters so you end up with things like the following, raw if statements are discouraged because of this:

  def classify(:ok)
  def classify({:error, reason})
  def classify([first | rest])
  def classify(%{name: name, age: age}) when age >= 18
  def classify(%{name: name, age: age}) when age < 18

The model seems to be a person when it's advantageous and not a person when it isn't...

Thanks, it's hard to stay up to date. However, we are just meant to believe that the mathematicians were going about the proof independently in the exact same way as the machines did it. It seems like a very odd coincidence to me.

I've read that another mathematicians work potentially has been incorporated into the training data with the work done on the Navier-Stokes equations so we should likely asterisk this one. Still it's mad these systems are this good that mathematicians are now using them to see further and probably to check their own work and understanding.

So you're absolutely certain, say after a few hundred rounds of self improvement, that the AI will always obey human instructions? It doesn't even always do what you asked now...

Bioweapons and cyber attacks on other system - for example the banking system or suppose and AI hacked into important Russian systems that pushed them back in time to almost pre-computer society or an attack on Chinese systems that made it look like the US was moving nuclear weapons into Taiwan, the responses from these countries could be awful and dramatic. We can't control what they do to be honest, I believe once self improvement happens the AIs will build in their own circumvention that we humans cannot even understand. We barely understand what is happening now in terms of interpretability of neural networks on tiny problems I'm not sure alignment is even feasible at the scales of parameters we are talking about today let alone in the future.