thaw13579

9 ms·

61.
by thaw13579 3y ago
Great point, perhaps “confabulation” is a better way of describing it, which means “the replacement of a gap in a person's memory by a falsification that they believe to be true”. For example, the term is sometime used to describe dem
62.
by thaw13579 3y ago
It’s more than next-word prediction though. The supervised fine tuning and RLHF steps are ways to possibly train it to favor truthful answers. Not sure whether this is currently the emphasis of ChatGPT though…
63.
by thaw13579 3y ago
Yep, if there were a macOS app on the iPad that provided a virtual desktop, that would solve it.
64.
by thaw13579 3y ago
Adderall also has addictive properties, so users develop tolerance and habituation, so hitting the correct dosage is quite a challenge, and a real downside in my opinion, given the consequences of over-use.
65.
by thaw13579 3y ago
I don't think it's so cut and dry. The article paints a somewhat simplistic picture of how the best performing LLMs work. The unsupervised pre-trained networks are indeed data hungry, but the secondary supervised learning stages
66.
by thaw13579 3y ago
Did you stop reading at the intro?
67.
by thaw13579 3y ago
I wonder where their revenue comes from, affiliate links, consulting, some eventual hiring at prestigious big Co?
68.
by thaw13579 4y ago
There's quite a bit of drudgery in academic positions, organizing/running meetings/conferences, editing journals, scrounging for reviewers, grant writing, and navigating all of the red tape of universities.
69.
by thaw13579 4y ago
Agreed the world has always had problems, but it feels more like a megaphone than a lens these days (found not only online, but also at work, school, and among friends). These problems also eventually reach people's doorsteps where the
70.
by thaw13579 4y ago
One caveat here is that even if every outcome is taken in some branch of the world, they don’t have equal likelihood. So while there might exist a world where a person makes that cruel decision, it may only be a small fraction. I imagine
71.
by thaw13579 4y ago
Right, I’m not sure whether this refers to finding another earth-like planet with JWST (interesting and plausible though not related to the many-worlds interpretation of quantum), or seeing another version of the Earth with JWST that follo
72.
by thaw13579 4y ago
Curious though, how does this relate to the JWST?
73.
by thaw13579 4y ago
Yes, when mixed together in the world at large, it's rare to run into people who have enough shared context to click with.
74.
by thaw13579 4y ago
It's hard to find opportunities for this outside college. Most people are not interested in intellectual talk, so the question is how to meet like-minded folks? I can't seem to find better alternatives to universities, for both n
75.
by thaw13579 4y ago
This resonances with me as well. Telling people about travel abroad is sort of like telling people about your dreams, rarely ever welcome.
76.
by thaw13579 4y ago
This is the way. Next, you could cache those values for >=6 when you compute them and also use the previously cached values for sub-problem solutions.
77.
by thaw13579 4y ago
It could go the other way as well, fewer people are starting a family because of the cost of housing.
78.
Huell Howser Archives: California’s Gold(blogs.chapman.edu)
78 points by thaw13579 4y ago|22 comments
79.
by thaw13579 4y ago
The sources are there in the training dataset, they are just not linked to the response. I don't think this is an inherent property of LLMs though, and I imagine future iterations will have some sort of attention mechanism that highli
80.
by thaw13579 4y ago
It has idiosyncratic word choice and punctuation, as well as references to other comments out of context.
81.
by thaw13579 4y ago
I asked it to explain some very technical medical testing procedures to me "like I was five years old". It completely nailed it by truly reducing it to its essence and using appropriate metaphors that a child would understand (co
82.
by thaw13579 4y ago
I think the parent is suggesting that the grant-based model of funding research is itself flawed. There are many issues, so it could be related to how the grants are chosen/reviewed, funds are allocated, universities depend on them, c
83.
by thaw13579 4y ago
Right, unfortunately the grant funding model is deeply embedded in university research. Some folks get some startup funds on hiring (usually for lab construction/setup), but the rest is dependent on outside resources via grants. Th
84.
by thaw13579 4y ago
At a research university, it would help get grants, a large chunk of which goes to the university via indirect costs. In practice, they expect the grants to flow on top of of these other minutiae often as unpaid overtime.
85.
by thaw13579 4y ago
Perhaps it’s best to avoid attribution at all when presenting scientific and mathematical ideas themselves? There are many contexts when priority claims are messy (colonialism and not) and are fundamentally an oversimplification of an incr
86.
by thaw13579 4y ago
It seems like an impractical but useful thought experiment, which I think makes a good point about why these lists should be taken with a grain of salt. To take things further, we could improve the experiment by randomly assigning books to
87.
by thaw13579 4y ago
Yes, I agree it’s good to sprinkle them in, but not for every slide, like the OP advises. I think the challenge is that evergreen content, by its nature, is general and can come across as irrelevant to the specific topic at hand (and easil
88.
by thaw13579 4y ago
I recall phone lines being overloaded shortly after, so that’s one bottleneck For dial-in.
89.
by thaw13579 4y ago
Agree, although I’d wager that prenatal environment still has a big role, especially in cases of poor nutrition, parental stress, and substance abuse.
90.
by thaw13579 4y ago
I could see that working for presentations for a general audience like a TED talk or maybe intro/closing slides, but otherwise many presentations are context-specific, and trying to frame the content as a "deep truth" might n

More ›