sgt101

This is much better:

https://www.rameznaam.com/p/471bbae4-1163-4048-944b-18f8b0bf...

I have sat in so many meetings looking at shit generated by claude while someone grins and goons in front of it, and I just realise... 99.9% of people in corporate life neither care about the things they produce or present, or can tell a carefully created document from any old shit.

I dunno man, I use LLM's to review things before I send them in and I often feel like I've been handcuffed to a chair, had a bright light shone in my face and had to account for all the mistakes I've made in my life.

They can be brutal.

Yet again: we're making the Russians do it?

I think all of them do, but the alternative is to let the Russians do what they want and that ends worse for all of us.

There's a leverage issue.

In one case (financial services) it's thought that expertise is valuable at V=S^2/b4 where V is value, S is skill and b capacity (the leverage available to the manager/expert. b erodes as it becomes harder to find examples of things that are not done well, so if you manage $1bn you might find lots of miss allocations that you can exploit with just that $1bn really effectively, but if you manage $10bn it's much harder to find good places for the extra $9bn. A low hanging fruit effect.

Anyway, that double hit - raw skill and the amount of times you can supply the skill makes the value of skill (V) convex, and it means that in a perfect market (heh heh heh) someone running $100bn is worth 1000's or maybe 10,000's of an average joe expert.

Now, if AI is trusted to run the top 0.1% of everything and has the skill to do it at human top level expertise, then your calc holds. If it's the case that it isn't then more than half of that value disappears. If it's not even top 1% then chop out another 25%.

That implies that we need a lot of trust and a lot of AI capability before these valuations stack up, and it also implies that all other competitors and incumbants are going away. I do not think that Citidal or Bridgewater are going to let Anthropic or OAI take them without a fight. They might lose - but there is a decent bet that they don't. I don't think that many professions like Lawyers or Doctors are just going to roll over and cede their monopoly rights to OAI or Anthropic either.

neat

3 tasks?

My point was more that it wasn't an entirely American project.

Given that OpenAI and Anthropic have had major incidents in the last few weeks why will giving them a monopoly on models make thing safer?

The correct response is for these two companies to be closed down and their executives imprisoned. Yet somehow they have a free pass and are pointing at people who have not done any of the things they have and asking for them to be penalised.

This is just bizarre.

Absolutely reliably the best way to increase your chance of survival in a situation like this is to have good relations with your neighbours in a good community with high trust.

A large number of people working together will be much more likely to find solutions than a few isolated individuals. The isolated individuals may have disproportionately large survival resources, but probably not for long. If you are in the caring-is-sharing group you will have a much better chance than being in the me-me-bang-oh-shit group.

But there are a lot of people who fear community solidarity because it might lead to people asking questions about private jets and beach condos.

>an entirely American project until the technology was exfiltrated

No it wasn't. https://en.wikipedia.org/wiki/Tube_Alloys

Post WW2 the USA (for a bunch of quite interesting reasons) excluded the British. But, of course, the British had acquired a lot of knowledge from the program and were able to develop their own bomb.

so like a novel or a working paper? Or notes?

That's approximated by the portals - the people building these go and look at Expedia for their flow and design ideas... so the information permeates across.

In what way do you think it learns?

I don't think it can learn from context, otherwise we could write Jane Eyre into it and it would be the novel? It's not learning as we conceive of it - it's another thing that we have labelled as "in context learning".

I have feelings, it needs to be able to empathise with me, or another driver, or a client...

Does it learn?

Does it experience?

Can it connect with other agents, understand them, come to empathise with them and find a way to work with them better?

The answer is no to all of these, and there are other problems as well. Yes, this model is trained to use a domain specific language to reason and plan over puzzle problems, and so it's programmers have cracked arc-agi-3 and that's a great achievement, but there is an asymmetry here. The arc team are well funded but are charged with providing a target for the vast ocean of funding, compute and talent everywhere else.

Most importantly, arc-agi-3 and the other benchmarks are all verifiable. The model can check if it's succeeded or not. They are not A* of course, but long horizon problems where you have to overcome minima to get the solution are not alien to AI either.

And the reason why there are 10 clicks is that that is the number of decisions that is required to get people to hand over their money.

There will have been an A-B test with 9 clicks -> less sales, less revenue.

There will have been an A-B test with 11 clicks -> less sales, less revenue.

The people making these sites are not dumb. They are also sitting on thin margins nowadays, and mostly from the airlines and hotels rather than from the users.

But, like, isn't that what Expedia does and has done for 20 years?

There's a herding effect with torrents where popular things are quick to get, unpopular things may be very difficult to get, because the majority of users delete files to save space.

I might host 3 or 4 models that I've downloaded recently, but I won't be hosting the 50 or so that I've tried in the last two years.

DISASTER.

Good story, write more. You have a good NO AI style.

There are problems with this analysis.

First it discounts the debt and financial load on the frontier labs to 0. The costs of developing this technology may well not be borne by those who reap the benefits. This is not like the Amazon distribution network.

Second it assumes that, like nuclear weapons, the technologies of frontier AI will not proliferate. I would argue that we have seen orders of magnitude reduction in compute requirements enabling sub linear improvement of scale driven AI in the last 24 months. Meanwhile 27bn param models do what trillion model parameters couldn't 12 mths ago. I think we're seeing a diversification of approach now, and it's not clear that the winners will be scale driven.

Third, and the most fatal, it assumes that AI will be the driver of power in the future geopolitical landscape. But frankly if I nuke you what will your clever machines do then?

This is not 1945 and this essay is essentially a note from a USA in the middle of an existential crisis channelling the 1950's.

Apologise and donate his fortune to homeless charities?

Worth watching Star City on Apple TV for a perspective on Soviet life. Now, Star City is definitely propaganda but it makes you think!

Do you know a good place for accessing the full sized chinese models but hosted not in China?

What I love is the idea that this is more likely to happen at Meta than at Anthropic, Google, Microsoft or OpenAI... or in the PRC!

There's contractual cover, there are lawyers all over the USA ready to make themselves very rich by creating a class action over it... you don't have to worry! Or, you do, but frankly, only MI6/CIA/Mossad can save you now.

Could this kind of trigger effect the standard candle application of Type 1a's? For example, if a black hole triggers a type 1a and that perturbs the type 1a light curve in some way (possibly because less mass could be consumed in the explosion than you would get if one was triggered by mass accretion) and the rate of triggering of this kind changed over time as the universe aged and somehow the population of primordial black holes changed, or the distributions of white dwarf candidates and pbhs changed...

In what way do you mean iterations - what's your model for iteration as a mechanism for iterating intelligence? Do you mean that intelligence improves with every generation?

Where are you subscribing ? z.ai?