Just as an example: I learn languages using Anki, and I always do it the same way: I use decks that
* exclusively quiz entire sentences
* introduce around 500 new words (a nice mix of nouns, verbs and adjectives)
* use a wide variety of grammatical constructs (including all conjugations of the new verbs),
* and that have audio of a native speaker reading the entire sentence after I "flip" the card
Such a deck needs to be thoroughly designed, and while I could choose the new words and then write software to make sure they are all used equally in sentences and no conjugations are missing, I actually can't easily make sure they are correct and I can't record the audio of the text.
> Many of my colleagues cracked 100k€ this year without being AT and having crazy high position ratings.
And for each of those guys there's 2 people working for 48k and happy about it. They've been at the same shop for 15 years, in a team of the only 3 people doing software in the entire company. Probably somewhere a bit rural, and/or north of Frankfurt.
Yes, medicine pays better, median is around 100k but with significant back loading towards the second half of the career.
Finance can be (much) better, but feels like far fewer jobs, especially outside Frankfurt. I'm not sure finding a high paying finance jobs is easier than finding a software job at the German office of an American firm (which pay similarly well).
> I suppose factory workers cannot be let go as easily.
It's important to look at comparable companies. If you're a SE at a company with many factory workers, firing the SE is usually equally as difficult as firing the factory worker. They usually have the same protections and are in the same union. Software shops just tend to be smaller and those have lower job security.
Even more interesting is if a thinking LLM would come up with tricks mitigating its own known limits - like listing animals in alphabetical order, or launching a shell/interpreter with a list that contains previous answers (which it then checks each new answer against).
> The problem is that european politicians don't want to kill the tech $$$. They just want to bring the revenue home. They don't understand that they will never make EU big tech and that their only feasible path forward to get rid of US tech is also the path that kills the goose.
Not necessarily. Red Hat is a billion dollar company just on FOSS support services and consulting. And if you put hundreds of thousands of clients on a completely novel FOSS stack, you're going to need several of those.
Sounds like you have a music discovery process in nicotine? Can you elaborate on how you find new things to listen to? Just my looking at what individual other users listen to?
Music discovery is the one thing I cannot drop Spotify for. I want to make a playlist with 10 songs and then have an algorithm suggest 20 more - ideally songs I have never listened to before, or songs I haven't listened to in a long time.
Spotify is mediocre at that task, but I just can't find a replacement at all...
Then, when downloading in nicotine, you can click a user to see all their shares, so I just scroll through what other kind of stuff they have, and download anything that strikes my fancy.
Dota 2 is a real time strategy game with an arguably more complex micro game (but a far simpler macro game than AoE2, but that's far easier for an AI to master), and OpenAI Five completely destroyed the reigning champions. In 2019. Perfect coordination between units, superhuman mechanical skill, perfect consistency.
I see no reason why AoE2 would be any different.
Worth noting that openAI Five was mostly deep reinforcement learning and massive distributed training, it didn't use image to text and an LLM for reasoning about what it sees to make its "decisions". But that wouldn't be a good way to do an AI like that anyway.
Oh, and humans still play Dota. It's still a highly competitive community. So that wasn't destroyed at all, most teams now use AI to study tactics and strategy.
> especially silly to lug that around if 90% of your trips are in battery range
The same argument works for large batteries, right? On 90% of your trips, you're lugging around several hundred pounds of battery you're not using.
If you want to tackle the weight argument, you could always drop 40 kWh battery capacity from the truck. That frees up around 600 lbs you can now use for the genset.
The maintenance thing is a real problem, of course. A 50 kW genset that almost never runs will be much better on mainenance than a classic ICE car, but still add significant maintenance cost to a BEV.
Which is part of why I think Range is often a distraction in EV discussions. The people asking for 600+ mile Range EVs may see those built because it always sounds like there are plenty of people "demanding" that, but the weight trade-off in batteries isn't going to make those great cars, most of that range will be "waste" given average trips. But it is easier to get to 600+ miles of range by adding more (and better) batteries than by making larger gas tanks and engines.
> my bet is it's almost identical to the BEV version but with an engine where the big beautiful frunk is now
Would be interesting how small and how cheap you can design a ~50kW genset to be (any smaller and you don't gain that highly coveted towing range). I don't think it's an easy task, you still need to integrate the crash compliant fuel tank, the emissions compliant exhaust system, water cooling for the engine, ect.
It's a pretty long BOM you're adding to an already expensive BEV, so you don't really have thousands of dollars of budget to add to the production cost.
> It's a pretty long BOM you're adding to an already expensive BEV, so you don't really have thousands of dollars of budget to add to the production cost.
Agreed. And I don't think it will be cheaper. The Lightning was already selling for less than the comparable ICE equivalent, there is no way they will sell the Lightning EREV at the same price point after adding a generator along with the associated supporting parts. I bet it will be at least 10K more, and I won't be shocked if it's closer to 20K.
If I had to bet; they'll put their 3.7L V6 in and run it on the miller cycle with a fixed drive to hit @130+kW or so.
The changes for cooling, etc. will be substantial, but the problem space is already well-known by the team, so the time to market probably won't be as long as we think.
That's probably a quick way to do it, but considering that using a miller cycle means we're going to want the turbo version of the engine, that alone is going to cost like $4k on Fords end. Add a 100kW+ generator with the power electronics to charge while driving, fuel system, exhaust system and cooling system, and we're probably approaching $10k upcharge for the customer.
Gotta remove a whole lot of batteries from that car to make it cost competitive again. Realistically, with an engine this powerful, we can probably cut down to like 30kWh of total battery capacity, which gets us back to where we started financially. And 30 kWh is enough to drive 70 miles all electric, which should pretty much should cover most daily use for people who charge at home.
Now, the questions if we can do that cheaper with a much smaller engine. Ford has a 1 liter inline 3 in the Fiesta and Focus that makes half as much power. Should be enough...
The dreamers amongst us have noted that Ford has a patent (at least an application for one, I don't recall if it was granted) for putting an EREV generator under the bed near the back of the truck. Since it can be a smaller engine and does not require an attachment to the drivetrain, maybe this is feasible.
If they did that, it would remove one of my reasons for not being too interested in the Lightning EREV -- the anticipated loss of the frunk. It still introduces a bunch of mechanical bits and associated maintenance requirements, that is unavoidable, along with a substantial increase in cost.
I bought my Lightning with the intent of keeping it 7-8 years, and it meets my needs very well, so this is mostly just navel gazing for me. The EREV version would have more range that I would rarely benefit from, and be substantially less powerful, which is also a negative from my perspective, in addition to costing a bunch more. My current truck is by far my favorite so far. I hope when I'm finally ready to try something new, there are better options. It's a high bar.
* exclusively quiz entire sentences
* introduce around 500 new words (a nice mix of nouns, verbs and adjectives)
* use a wide variety of grammatical constructs (including all conjugations of the new verbs),
* and that have audio of a native speaker reading the entire sentence after I "flip" the card
Such a deck needs to be thoroughly designed, and while I could choose the new words and then write software to make sure they are all used equally in sentences and no conjugations are missing, I actually can't easily make sure they are correct and I can't record the audio of the text.
reply