gbasin

11 ms·

31.
by gbasin 3y ago
The latencies and costs are changing so quickly...
32.
by gbasin 3y ago
you don't need access to the weights though, unless I missed something? this is akin to the various groups using GPT-3.5/4 to generate data for fine-tuning smaller models
33.
by gbasin 3y ago
Yes this feels analogous to ensembling which can help measure variance, also
34.
by gbasin 3y ago
Very exciting, thanks for sharing all this
35.
by gbasin 3y ago
Does funding matter? <insert one of a dozen studies here showing little relationship between student funding and outcomes, across and within countries>
36.
by gbasin 4y ago
The RLHF process performed on that chat models actively confuses their ability to estimate confidence. The raw models are quite well calibrated though
37.
by gbasin 4y ago
At least in the US, this will only be true before a company goes public. After that point, the shares will all be freely salable and have the same value (except for a small premium for supervoting shares which some founders may retain).
38.
by gbasin 4y ago
When was that not the case? Only time I can think of is A/B testing facebook.com and maybe the transition to mobile feed?
39.
by gbasin 4y ago
There is "long range" dependence, it's just only on the prompt: the conversation with the user and the hidden header (e.g. "Answer as ChatGPT, an intelligent AI, state your reasons, be succinct, etc."). That ends up
40.
by gbasin 4y ago
ngl would've killed for more alpha back then
41.
by gbasin 4y ago
Wow I would have killed for this in my algo trading days :) Great looking tool
42.
by gbasin 4y ago
Ya, luckily we already have one a large bank ready to use it and that's enough to start: https://twitter.com/garybasin/status/1624134955566788608?s=2...
43.
by gbasin 4y ago
Cofounder here, happy to answer questions
44.
Pineapple Litepaper: mortgage banking on a blockchain(pineapple.mortgage)
20 points by gbasin 4y ago|5 comments
45.
by gbasin 4y ago
The 3rd one is interesting and something I hadn't considered as much before — may be applicable still in less regulated markets
46.
by gbasin 4y ago
Not to create unnecessary paranoia, and the odds are low, but there is a possibility of early onset alzheimer's...
47.
by gbasin 4y ago
some charts here https://twitter.com/Econimica/media
48.
by gbasin 4y ago
To play devil's advocate, these big institutional buyers moving into the single-family home market after a huge run up just feels like suckers rushing in at the top. We're possibly at a multi-generational peak in inflation-adjuste
49.
by gbasin 4y ago
ya, also keep everything in cache to be below a few mics
50.
by gbasin 4y ago
sub 10 in Java was very doable 10 years ago, fwiw. even better than collecting once a day is doing everything off-heap — C-style java :D
51.
by gbasin 4y ago
Sure ping me on twitter
52.
by gbasin 4y ago
I founded and ran a quant trading firm for about a decade :)
53.
by gbasin 4y ago
If your predictions are good, I'd be happy to get you $100 million in assets to manage. It's very unlikely that your predictions are good...
54.
by gbasin 4y ago
ok and then what? :D you still can't recover original
55.
by gbasin 4y ago
Nothing a little Stable Diffusion can't fix. Soon, digital memories will be just like real ones — slightly different every time we look back, but with no way of knowing what's changed
56.
by gbasin 4y ago
the innovator's dilemma
57.
by gbasin 4y ago
anti-trust law would prevent the most egregious forms of this
58.
by gbasin 4y ago
Take a look at Manhattan's Financial District — seems like half of the commercial buildings have already been converted to residential. Some layouts are a bit awkward but it works fine
59.
by gbasin 4y ago
The CoreLogic data shows that what it calls “mega” investors, with a thousand or more homes, bought 3% of houses last year and in 2022, compared with about 1% in previous years, with the bulk of investor purchases made by smaller groups.
60.
by gbasin 4y ago
Nothing an LCD and some cameras can't solve

More ›