Yes. Their architecture recomputes every time so at 150k context every request will have to spend 1.5 min waiting for the model to reread the context.
Say avg model response length is 1024 tok. At 50 tok/s normal providers do your turn will only take 20s (vs Cerebras 101s) and will cost 20x less. That time and cost is per single tool call.
They are increasingly being trained on generated tasks and even (parts) of the pre-training data is 'distilled' (e.g. Clibmix as an open-source example), so there are many ways in which the vocabulary can seep into the model.
SAP isn’t a company to be crazy proud of, but it doesn’t make it not a success nor does it mean “EU is fucked”.
May I remind you that in the US, Facebook is in that same list of top successes? Have you ever heard the phrase “the top minds of this generation of Americans are all working on ads”?
Also, if huggingface is “technically” American, then fuck me, Stripe is technically European. Hell, their CEO is insanely pro europe, pro EU, and a big enabler of projects like EU Inc and lately, the Rhine Group.
I mean EU can cope all it wants, the most successful startup in Germany being a weapons company is all that you really need to know. It is sad how far these countries have fallen and I am German myself.
Germany isn’t exactly great at startups, it’s one of the worst countries in europe to create and scale one. Helsing is doing great in spite of that ecosystem, and thanks to good timing and current geopolitics.
That itself does not reflect on all of EU. Since the UK left, Germany has probably been the worst “special child” of the lot to be honest; their notaries and trade unions are the primary challengers to EU Inc for the most ridiculous reasons for example.
Maintining this type of long time fork is the kind of mind numbing, tedious thing LLMs are actually exceptionally good at. I did some backend work on Lean4 compiler and then they switched out parts of the codegen, it took Codex less than an hour to recover the feature set I had implemented but on the new backend.
All this AI doomerism is hiding the obvious and immediate harm that will cause a mass extinction event. Funnily enough people at Anthropic are predicting the end of the Anthropocene, completely absolving themselves from any responsibility for the actual reality that the Anthropocene is defined by human made geologically noticeable changes. Like causing a global climate catastrophe.
It's possible to have more than one existential risk at a time. We even have a third, since nuclear war doesn't seem completely impossible these days. We probably shouldn't ignore any of them.
An all-out global nuclear war (all nuclear nations landing 10% of their warheads in populated areas) would be a disruption comparable to COVID at best.
Can you elaborate on this? I’m not an expert at all but this seems so far from true I’m struggling to understand if you’re serious.
Wouldn’t we expect tens of millions of casualties, nuclear fallout rendering large swathes of countries uninhabitable, and ensuing total societal collapse once the globalized western world is separated from their foreign means of production?
Of course you're correct. Here's a 2019 simulation of escalating full-scale war between the US and Russia, done at Princeton based on what's publicly known about actual force postures. Immediate effects would be 34 million dead, 91 million total casualties, and 30 major economic centers destroyed on each side.
Fallout would cause significantly more casualties. Then there are the environmental effects. Science on that has evolved over the past few decades, so wikipedia is a good starting point for an overview.
Since none of this is certain and we're talking about risks, considering the worst cases seems appropriate, and those involve agricultural collapse, several billion dead, mass extinction, and an ozone hole covering most of the planet. It might not kill off the human species, but modern civilization would struggle to recover.
It is the paperclip AI scenario:
The AI is cheating humans to produce only one thing - but in reality today, its not paperclibs, but datacenters as "the most important product": So the scenario still holds :-D
The existential risk people are worried about a much shorter timeline - like the next 5-10 years. Climate change is serious but very unlikely to lead to anything like ending the Anthropocene in the next 5 years.
400,000 people in Europe are displaced. Pretending that this is a future issue is tremendously disingenuous. Things don't need to be civilisation-ending to cause a degree of harm that is worth focusing on.
Yes, it's pure, horrible, AI slop. And as usual, all the people that mentioned this got downvoted here. Unfortunately, HN users can't even tell this crap from useful text.
"What it means and what it does not": article is clearly written by claude, which for me at this point means that I pay as much attention to it as I do to Claudes output: I skim it and then move on. It actually requires far more effort of me to read it than it was to generate the content. Before it would have always been the other way around. Whoever wrote the article would have had to spend way more time writing it than me reading it.
