jumpCastle

8 ms·

31.
by jumpCastle 3y ago
Not by far, there's bloom and opt if you count it.
32.
by jumpCastle 3y ago
It is fine tuned to maximize reward though, not likelihood. And it provides an answer in both cases, just not as well.
33.
by jumpCastle 3y ago
In what sense it was more complete or replicable than this paper?
34.
by jumpCastle 3y ago
It’s easy to quit smoking. I’ve done it hundreds of times.
35.
by jumpCastle 3y ago
Attention is all you need had no code or model and still led somewhere.
36.
by jumpCastle 3y ago
If you apply for the api you can use the time-stamped original version.
37.
by jumpCastle 3y ago
Is Linux battery life good?
38.
by jumpCastle 3y ago
Closed source models are also improving. Not clear to me why would open source models will catch up.
39.
by jumpCastle 3y ago
It does not seem conversational though
40.
by jumpCastle 3y ago
The name 3.5-turbo sounds to me like it implies distillation. The release notes at the time also hinted at it IIRC.
41.
by jumpCastle 3y ago
In my experiments bard is weaker than 3.5, but if it wasn't, than I would prefer the fresh data of bard.
42.
by jumpCastle 3y ago
Optimazing for training could help distillation also.
43.
by jumpCastle 3y ago
These aren't their core product.
44.
by jumpCastle 3y ago
Did relevant tax rules change recently? If not, why is this behavior new?
45.
by jumpCastle 4y ago
One reason might be that some people value open source.
46.
by jumpCastle 4y ago
Looks like llama 7b was trained on 4 times more tokens.
47.
by jumpCastle 9y ago
I didn't need to call. I canceled last year via chat. It took me 15 minutes, answering about 10 quite annoying questions.
48.
by jumpCastle 10y ago
I'm not an expert on routers but if I was in the market I would look at the Wirecutter. I find their recommendations pretty good. thewirecutter.com/reviews/best-wi-fi-router/
49.
by jumpCastle 10y ago
I subscribe to a newsletter of a product recommendation service like wirecutter.com, which provides discovery. Ideally I would be able to pay for this service so that their incentive to provide quality recommendations would align better wit
50.
by jumpCastle 10y ago
I read people claim that they like it since it reminds them that they pay for the government's operations.
51.
by jumpCastle 10y ago
priceline 86B$, ctrip 24B$
52.
by jumpCastle 10y ago
That's true, although significant corruption might make GDP growth too slow such that the start ups will run out of money before they can benefit from it.
53.
by jumpCastle 10y ago
Yet Nigeria's GDP is smaller than Florida's according to google.
54.
by jumpCastle 10y ago
I liked all three of them. I also feel Murphy's MLAPP provides additional useful material and is well written.
55.
by jumpCastle 10y ago
The difference in the quality of the microphones make it quiet impractical in phone, but very easy in a google home device.
56.
by jumpCastle 10y ago
Game theory is domain specific. Generic methods in AI tend to dominate domain knowledge over time. Although I agree that other game-theoretic techniques might help here.
57.
by jumpCastle 10y ago
I might pay subscription. I agree that the market is not likely to be big. Also data network effects and difficulty with price discrimination can be big obstacles. EDIT: I think I misunderstood your question. I will probably pay that much f
58.
by jumpCastle 10y ago
Or a paid search engine incentivized to provide quality results rather than maximize ad network profits
59.
by jumpCastle 10y ago
Thanks again. So I guess for multiplayer games a better approach would be reinforcement learning with no game theoretic heuristics?
60.
by jumpCastle 10y ago
Thanks for the explanation. I will also appreciate a reference to a good survey paper if you are in the mood :)

More ›