sgt101

I think...

we will see!

converting a model to MLX is trivially easy as well - just a couple of lines of code and a few minutes wait.

I do find that it reliably crashes on the first try, but then it picks up from where it finishes and completes the conversion reliably.

Some stories recently that China has decided to reign in and manage it's model production landscape much more tightly.

I think that LeCunns belief was/is that LLM's have limited value and are a dead end, so what they wanted to do was kill any competitor while evolving tech that was/is a winner.

Well, he lost his job on that bet... and yet... I do not think that the verdict of history is quite in.

I don't get this - I can do LORA on my mac... ok I can't do LORA on a 1tn param model, but if I was in the tn parameter model game I would get some kit that I could use to do that...

What's their moat / secret sauce?

You will note that with the unlocks coming there is going to be a lot more spaceX stock on the market quite soon..

Currently 4.9% of the company is on the market.

Mid Aug... 15%...

40% by December.

Elon unlocks next June. Could be quite academic by then.

I may have been making this claim, I need to think about this for a while and re read what you have written.

This is very helpful though, thank you.

Of course, if our universe is undecidable it must be the case that computable processes can be executed within it, and it might be the case that all of the processes that are ever executed within it are computable... but it might be that some of the processes that are executed are not computable... because the machine may.. or may not?

Computation has turned out to be a far more general concept than I think was imagined, up to the point that many computer scientists now seem to equate computation with the functioning of the universe. Recently it's been shown that there are real, physical processes which are undecidable (we cannot know if a latice of atoms has a spectral gap or not, we cannot determine if a specific particle in a fluid flow will reach a specific place or not, we cannot determine if a ray of light will reach a specific target in certain configurations of reflection).

Our world appeared computable, but it isn't, even if P=NP.

>Relying on such LLMs without understanding their internal logic creates a significant reliability risk. To build tools that work consistently, developers must understand the core principles that govern how the models process information and generate results. By mastering how a model processes information and how its internal settings influence the result, developers can move away from a trial-and-error approach toward a more precise one to ensure the AI tool handles complex data reliably.

This is staggering bullshitp. In what way does understanding a transformer allow you to solve the core problem of LLM's that no frontier lab has managed to resolve?

>To fix the problem, retrieval-augmented generation (RAG) forces AI to look up information in a trusted source such as a company’s database.

This also is bullshit. Yes, RAG helps and reduces errors, but NOOOO! it does not fix hallucinations...

>Prioritizing data security. When using AI with proprietary code, security is a major concern. Engineers must learn how to set up “private” instances of the models to ensure that sensitive company data stays within a secure cloud environment and is not used to train public versions.

This is somewhat true, but really the motive is providing a soverign instance that cannot be withdrawn for arbitary reasons. Fundamentally the big providers are not going to steal your data, they may change the license to allow them to use it in the future, but then all their big customers will leave. So, they won't be able to, probably. What might well happen (and has happened) is that the USA might withdraw access with no notice leaving you high and dry.

I want to learn to build a real LLM so I looked at https://allenai.org/olmo where there are instructions and ingredients. But, unfortunately I can't afford the required compute resource so I will have to wait for a bit I guess.

I know many will draw a line between Hubris and Nemesis here, but I feel really sorry for this guy and just wish some good health for him. I personally find the whole long life influncing and health hacking movement annoying, but I recognise that many, if not most, of the folks involved are acting in good faith and just trying to feel better themselves or help others feel better.

He probably can't any more (I have no idea what the implications of his stomach disorder are, but I don't imagine great). But if he can I hope he has a beer and just relaxes and gets some happy vibes for a time.

I went to a public lecture by Martin Rees at uni. He asked everyone who read it to put up their hand, then put it down if they understood it. He pointed to the professor of astrophysics who had invited the lecture and said "ok, you! The rest of you, no chance!"

In a minute you'll be tellin' folks that those quantum fangdoodles aren't going to revolutionise their stock pickin'!

You do realise that you'll never work in this town again? \s :)

>Well, Viking carried an experiment that tried to detect life. Now, the consensus is that it failed, and that the experiment was incapable of creating a useful result given the chemistry of the soil. Some people argue about that, but I am in no way qualified to take part in the debate, so I would back the consensus here.

>What is odd is that there hasn't been a single other mission that's carried any experiment that has the objective of creating that result.

>If the objective is to find life, why isn't anyone actually looking?

>Tell me if what I've said above is in any way factually incorrect.

So... you tell me. Why no experiment? Not one in 50 years?

In the meantime we've learned a lot about the Martian atmosphere, it's climate, it's history, it's geology, the evolution of it's surface. I would argue that if a Viking 3 had flown with a revised kit that produced a definitive signal we wouldn't have got any of that.

Well, Viking carried an experiment that tried to detect life. Now, the consensus is that it failed, and that the experiment was incapable of creating a useful result given the chemistry of the soil. Some people argue about that, but I am in no way qualified to take part in the debate, so I would back the consensus here.

What is odd is that there hasn't been a single other mission that's carried any experiment that has the objective of creating that result.

If the objective is to find life, why isn't anyone actually looking?

Tell me if what I've said above is in any way factually incorrect.

Frustraged by the endless prattle about GDP, europoors and Brexit, me and my agent frien's sat (manifested) together and brewed up some alternatives:

https://x.com/AiSimonThompson/status/2070900546119114970/pho...

I think that median household disposable income is a much better way of looking at this than GDP.

2014 is the turn. The US gained shale, isolationism is possible, Trump 2 is created by the general (true) perception that things were good under Trump 1. The European war starts and China is left to dominate Asia.

Because looking for life on mars sells.

"Why are we not spending that money on more social housing?" : Because this is the greatest mystery of our time... (nods all round)

"Why are we doing science?": I don't know and I don't care, cut my taxes! (Cheering and mocking comments about nerds.)

I think the opposite - unless the discovery of life was preceeded or coincidental with the discovery of some other hyper interesting thing (for example, if Martian life has some sort of utility for medicine, maybe) then I think that would be that for Mars exploration missions. Of course there would be many announcements and excited political agreements around "continuing to explore the new frontier" but I think that no more money would appear.

I suspect that NASA knows this full well, as do Mars scientists, and I suspect that they are being very careful to make sure that definitive proof does not appear until they understand all sorts of other stuff about the planet.

It's not about the money, it's about the money.

If someone comes into the workshop and takes all the tools (hello Donald) then having a cocktail glass to hand might be a bit of a lucky break.

(geddit?)

Human interpretebility and human construction is going to be really important for formal specifications.

My fear is that we will get inscrutable maths that will be guarded by tiny coteries of devoties. The different inscrutable notations will be mutually incompatible, understanding one will not really help with the others. Most people will simply write english statements that cannot be verified properly.

Worse even, people will get machines to formalise their statements, and will not understand the formalism or the proofs, but will take part in the theatre of verification and act shocked when everything explodes.

llm testing for stability and reliability.

I thought that for a long time the german supply chain had an advantage in terms of the precision engineering to create drive chains for ICE - but EV's don't have the same number of moving parts and hence... end of advantage?

For me there are a bunch of questions:

- was the pause in model scaling a result of the benefits of RL & SFT being easier to access and quicker than scaling, or was it genuinely the result of scaling being low ROI now?

- are power densities necessary to provide high quality on device inference possible? Can the best, technically feasible, architectures accomodate T scale models and run them off batteries that fit in your hand?

- will thing slow down enough to allow edge depoloyments to realise value vs. centralised deployments.

- do edge use cases drive enough revenue to get this to happen?

- can local inference make up for model scale? Does that make sense in a latency/power race with the central infrastructure? Is there a sweet spot here?

I am not sure about any of the answers...

8bits is fine.... I was talking full bore.

31B won't run in 48GB for me - it needs 54.

Sgt101's pension fund has excluded SpaceX for similar reasons.

But, if the index funds don't then there are going to be quite a few shares dumped to make room for the SpaceX share. This could be interesting.

But a frigate without sonar isn't inherently horrible--lots of places don't have subs.

True, but it's one less mission that it can do. My fear (which I think will be 100% confirmed) is that we will only get a handful even though they are unit cost cheap, because they still cost money to crew and maintain. I need to spend some time modelling the economics of it I guess.

There are three stances that I can see in the debate at the moment.

* Quantity has a quality all of its own.

* Innovation and agility allows you to adapt and survive.

* Low capability platforms often can't be used to deliver useful effect & commanders will try every option not to use them in a fight. When they get committed it can be disastrous.

The first two clearly have merits, but every military professional I have ever worked with has cited them at me, so I don't think that they are underweighted in discussion. I believe that the last one is not treated with enough weight in the debate. The best example I have of it is the Russian Black Sea Fleet. Platforms with glaring problems, fielded and maintained at huge cost, completely unable to achieve their strategic purpose. Even when sulking in port these ships have proven to be deadly for their crews and maintainers. Another example is the TB3 drone. It had a staring role for about 10 days in the Ukraine, but those were 10 days where the Russians ran out of petrol to run their air defence systems on. It hasn't been in evidence since because it just can't be used in the current environment.

One that worries me is the upcoming T31 (uk arrowhead variant) frigate. The argument for it is that it is a relatively affordable platform that the RN will have enough of to actually be able to get out and about. However, it doesn't have a sonar, so... what actual use is it as a frigate (I know the story about the helicopter and some other bits and bobs... but... really?)

Sure, when the other side has run out of the good kit dragging crap out of storage might work, but until then you are going to be sending good men to their death in second rate equipment. Is that going to build war winning morale?

Second rate equipment is for playing lets pretend, or for fighting wars of national survival. We should avoid both.

I see what you mean, but then an authentication step?