jvm

If consumers paid close to wholesale rates for their home energy they would be highly incentivized to do these sorts of things: they'd pay almost nothing (or maybe even less than nothing) in the day and big bucks from 5 PM to 8 PM. There would be whole industries helping people shift consumption to daylight hours. Unfortunately legislatures have consistently been acting to shield consumers from variable time of day costs, preventing behavior adjustment.

I'm curious, what actually happens when you're fined some comically large amount of money? Do you just immediately declare bankruptcy and let the courts carve up your assets? Is that usually the end of that or do you still end up having your wages garnished after bankruptcy?

Doesn't make much sense to compare a model that's not fine tuned to flan models that are fine tuned. Makes more sense to compare to something like T5 base where it's probably a lot more comparable.

My (John McDonnell) conversation with Kanjun Qiu and Josh Albrecht are the Co-founders of Generally Intelligent.

We talked about the state of reinforcement learning and what the field needs to succeed in building autonomous agents, the challenge of founding and building a company with a research-first agenda, which AI technical advances are over- and under-hyped, and much more.

Kanjun thinks language models are _still_ underrated!

My (John McDonnell) conversation with Kanjun Qiu and Josh Albrecht are the Co-founders of Generally Intelligent. We got into the state of reinforcement learning and what the field needs to succeed in building autonomous agents, the challenge of founding and building a company with a research-first agenda, which AI technical advances are over- and under-hyped, and much more.

Kanjun thinks language models are _still_ underrated!

Author here, yeah I agree the real world is very complicated. For any given task I'm imagining something like

1. Use some sort of instruction tuning to get the thing "good enough" that it gives decent results 75% of the time and the other 25% a human has to take over. 2. Use the actual usage data as training input. Punish bad behaviors and show the model what the human did to solve the problem. 3. Use this training loop to progressively have the model take over a larger % of the time.

…and I think if you can't get (1) good enough to be worth using it's going to be really hard to get the loop going.

Selling online without negotiated price is a big change. That should make the dealerships really uncomfortable in the long run… he's kind of saying "oh the dealerships could rent out Mustangs" but is that actually realistic?

I keep hearing this business about the US being a reserve currency and it makes no sense.

First off, lots of other currencies are valuable without being reserve currencies. Swiss franc, Japanese yen, etc. Being a reserve currency probably boosts USD a bit but it's not the only pillar holding the thing up that people imply. If you disagree show your model: You have to believe that value for the euro, GBP, JPY, Swiss Franc is determined by thing X, USA does not have thing X and instead depends on being a reserve currency.

Second, what's the alternative reserve currency? Bitcoin? Don't make me laugh. RMB? Even more ridiculous. Euro? Probably the most realistic contender but seems to have major governance problems. As with democracy, the USD is the worst reserve currency, except for all the alternatives.

Evergreen

7% is a low interest rate for an unsecured small business loan.

I'm not 100% sure what the FRED chart you cited means by "Effective Loan Rate for Small Business Administration" but those are most likely collateralized.

To actually answer your questions:

- Noise means you can't do US domestic

They don't seem to be targeting this.

- Concorde didn't have the range for Pacific

They do seem to be attempting Pacific range.

- Costs didn't work for Atlantic routes

They are trying to bring down costs considerably.

- And airlines want lots of identical planes, not one special one for one route

This will certainly be a drawback, although if they could take e.g. 50% of premium transoceanic it won't be so specialized.

Not sure where indeed is getting their numbers, but I can assure you levels.fyi are well calibrated. Glassdoor also has weirdly low estimated salaries.