If rumours OpenAI are doing 70% margins on inference and Anthropic doing 30% margins, then open weights models hosted on clouds happy with 10% margins increase competition and decrease cost. I’m game, like most. Much easier compliance with data sovereign concerns too.
> OpenAI are doing 70% margins on inference and Anthropic doing 30% margins
That difference is actually pretty surprising. Is Claude that much more expensive to host? The end-user pricing seems to be pretty similar, or better for OpenAI.
In this context, it means being able to deterministically predict properties of the output based on properties of the input. That is, you don’t treat each distinct input as a unicorn, but instead consider properties of the input, and you want to know useful properties of the output. With LLMs, you can only do that statistically at best, but not deterministically, in the sense of being able to know that whenever the input has property A then the output will always have property B.
I mean can’t you have a grammar on both ends and just set out-of-language tokens to zero. I thought one of the APIs had a way to staple a JSON schema to the output, for ex.
We’re making pretty strong statements here. It’s not like it’s impossible to make sure DROP TABLE doesn’t get output.
You still can’t predict whether the in-language responses will be correct or not.
As an analogy: If, for a compiler, you verify that its output is valid machine code, that doesn’t tell you whether the output machine code is faithful to the input source code. For example, you might want to have the assurance that if the input specifies a terminating program, then the output machine code represents a terminating program as well. For a compiler, you can guarantee that such properties are true by construction.
More generally, you can write your programs such that you can prove from their code that they satisfy properties you are interested in for all inputs.
With LLMs, however, you have no practical way to reason about relations between the properties of inputs and outputs.
I think they mean having some useful predicates P, Q such that for any input i and for any output o that the LLM can generate from that input, P(i) => Q(o).
Having that property is still a looooong way away from being able to get a meaningful answer. Consider P being something like "asks for SQL output" and Q being "is syntactically valid SQL output". This would represent a useful guarantee, but it would not in any way mean that you could do away with the LLM.
Same here - I'll only buy books I can read on any device of my choosing. Kindle+dedrm was an option. https://www.kobo.com/gb/en/p/drm-free is another option I have used a lot. But if it's not available, I will go to the modern day library of alexandria. I will not pay for crap that will just stop working in a few years - a book can sit on my shelf for 15 years before I get around to reading it.
Most definitely not true. Maybe enough area to cover current electric usage, but to truly decarbonize society a lot more renewable energy is needed - for transport, heating, iron industry, chemical industries, fetilizers etc. Massive amounts of electricity is needed unless you export your industries to china.
Heating with heat pumps is highly efficient and already the cheapest way of heating your home. The grids are ready for it - especially considering the amount of residential solar.
Yes but the major energy consumption of a household is heating than transport than utility.
Using ground heat (deep ones) reduces the electricity need sign.
Also if a heat pumpt creates 3-6 the energy from 1kwh, its even more efficient to burn oil and gas to make energy out of it and remote heat than just burning it locally in your burner.
Why not? Germany's total energy consumption is estimated to be around 1-2 TWh/y. This could be generated by photovoltaics covering less than 5% of its land surface.
There are significant problems around rolling out that much capacity quickly enough, and I also don't think nuclear should have been shut down that hastily, but I don't think "only nuclear can cover long-term energy needs" is true in any way.
Wtf are your numbers from, but they are wrong. It's over 2200TWh per year. And it you truly want to be renewable, the numbers go up. Upcycling waste to plastics or using hydrogen to make steel is more energy intensive than using fossil fuels.
https://www.umweltbundesamt.de/en/indicator-final-energy-con...
Those are total energy numbers, which includes fossil fuels, but those are famously misleading because replacing those with electricity reduces the number of Wh needed. An electric car needs roughly 15kWh for 100 kilometers, a gas powered car typically at least 60kWh for the same distance.
Electrifying reduces energy consumption only in selected use cases. Such as EVs yes. However other usescases such as making steel with hydrogen, plastic fromwaste or fuel for planes require vastly more energy when electrified.
Unsurprisingly the use cases where energy consumption is going down lead on electrification (because it's a cost advantage), so it may seem like electrification reduces energy consumption.
But if you really want to leave fossil fuels behind, the electric consumption will go up, up and beyond.
Electricity consumption will go up* but energy consumption will go down. You will not need 2200TWh of energy in Germany when all is said and done. Heating is one of the top reasons we spend energy and heat pumps are just tremendously more efficient than something like gas heating. You can get the same amount of heating for 3-4 times less energy with a heat pump than gas. So obviously you will not need 2200TWh of electricity like you do now with fossil fuels for energy.
* It's also debatable how much electricity use will actually go up. Logic says this must happen, but logic is not science. We have millions of EVs now in the EU and electricity production is less than it was 20 years ago. Efficiency is a source of energy. If you look at the US for example, it uses almost twice as much electricity per capita than Germany, and I would say they both get the same high level if living. If you look at it that way, Americans can cut their use almost in half and live the same standard of living. This can power a lot of EVs and heat pumps without adding a single GW of new capacity.
Energy consumption in total in Germany will only go down if you decide to export your steel and chemical industries to china. The high temperatures needed by industrial processes can't be achieved with heat pumps.
No. If you electrify residential heating and transportation it will obviously go down. If there are sectors where you can't do that, it will still overall go down because those other sectors will not go up to make up for the reduction. Not sure what your argument is.
The number Lxgr gave, 1-2 TWh/year, is simply completely wrong. Germany's annual electricity use alone is around 500 TWh/year. 1-2 THw/year would be the electricity use of 300-600k average German houses.
I have my doubts about short and medium term feasibility, and much more importantly storage and adapting carbon-based industrial processes.
But yes, if all it took was 5% of landmass (which also doesn’t get permanently unusable nor polluted), I’d say that would be a pretty good deal, yeah. This is significantly less than what’s used for livestock farming, to put it into perspective.
Realistically, I don’t think we’ll solve storage fast enough to be able to afford zero nuclear power in Europe.
And of course, you can combine those things sometimes - I've seen cattle munching on grass under solar panels in Baden-Württemberg (state just west of Bavaria).
You can install solar panels over areas that are already developed — rooftops (lol), parking garages, highways, and so on. Some agricultural land even benefits from being covered by solar panels. This has great potential and was first researched in the United States. China is covering water reservoirs with solar panels, which has the additional positive effect of reducing evaporation. And then there is the incredibly large amount of energy that the North Sea, far from any beaches or islands, could provide in consistent wind energy.
Rooftop solar is prohibitively expensive in Germany. My installation would only cover its costs if electricity becomes so expensive that it would lead to complete economic collapse.
No. In Germany, rooftop solar is usually economically attractive, not prohibitively expensive. Especially on a decent roof and if you use a fair share of the power yourself. Verbraucherzentrale(1) says PV systems for private homes are “usually worthwhile” economically, and that self-consumption is the key driver of profitability.
Over 40% of the German landmass is currently used to produce food for farm animals. The space requirement for solar is far off from that. And you can use rooftops etc.
reply