4 ms·
> If training costs 17x revenue you'd better bring in some customers, but you might be alright in the long term if you've trained something useful. Yes. The ke
by rvz 3y ago
> If training costs 17x revenue you'd better bring in some customers, but you might be alright in the long term if you've trained something useful.
Yes. The key is 'if' and that is the hidden risk when training a model which can overfit or under-perform.
That cost is in the tens of millions and they (Cognition Labs) better make some revenue to cancel all of this out.
- fancyfredbot 3y agoThere's nothing especially remarkable about taking some risk - everything from drug discovery to drilling for oil involves large up-front costs with uncertain returns. It sounds a bit like Cognition labs can afford to take a few losing bets costing tens of millions (based on $2bn valuation) but investors seem to be speculating that they'll find something which works in the end.