Joe @twitskeptic@qoto.org
Unprofessional data wrangler and Mastodon’s official fact checker. Older and crankier than you are.
Joined Nov 2022
Looks like blocking no longer works on Twitter, at least if you click on a trending hashtag.
Joe
boosted
Some good news:
WHO says only 1 of 12 samples taken from contacts of the 11-yo girl who died earlier this week from #H5N1 has tested positive for the virus. It was from the girl's father.
WHO is providing flu antivirals & helping with the field investigation.
@paul@social.paulkedrosky.com Meta just announced a new model (LLaMA) that they claim achieves GPT3 performance with less than 10% of the parameters. Not sure how this will ultimately play out, but I think it’s way too early to make definitive predictions about the GHG impact of LLMs.
Meta just released a publicly available pretrained LLM (research access only, apparently) that claims GPT-3 level performance with < 10% of the parameters (13B vs 175B)
https://research.facebook.com/publications/llama-open-and-efficient-foundation-language-models/
https://github.com/facebookresearch/llama
@pbump More optimistic view - it's remarkable that ~70% of older Republicans still see Ukraine as an ally, given the barrage of anti-Ukraine propaganda on Fox.
Just the kind of unironic self examination I'd expect from a billionaire Silicon Valley VC bro.
@katestarbird Curious about the answer!
@ct_bergstrom Ugh, looks like 1 of the 4 contacts - the girl's father - has tested positive. https://bnonews.com/index.php/2023/02/cambodia-reports-2nd-human-case-of-h5n1-bird-flu/
Joe
boosted
Anybody know of any good articles considering the possibilities of the resonant effects of AI training (for both predictive and generative tasks) on inputs increasingly produced by generative AI? And/or where strategic manipulation of AI might occur (through intentionally crafted content that ends up as training data)?
@katestarbird What a great research idea! It would also be interesting to study the effects of *adversarial* "intentionally crafted content" if it were widely distributed.
Depressingly dystopian but likely relevant, unfortunately.
Strange visualization here about the efficacy of sanctions against Russia by the NYTimes. Each country is represented as a circle sized by population which tells you, well, nothing when you're really interested in the economic might of a country exporting goods to Russia.
Not to mention that it's very difficult to quantitatively compare the area of 2D objects (circles).
https://www.nytimes.com/interactive/2023/02/23/world/russia-ukraine-geopolitics.html
@kristinmbranson @cstanhope I tried copilot but didn't like it due to the annoying UI in the IDE I use (PyCharm). I felt like I was fighting with it all of the time. But it seemed like it might have potential with a better UI.
As for ChatGPT, check out this example I tried. No, it won't write your program for you, but it's certainly good at generating snippets that are useful.
@kristinmbranson @cstanhope Disagree - there are areas where LLMs are useful. For instance, ChatGPT is pretty good at code generation. Yes, it's often wrong but even the incorrect code can be helpful. I'm guessing that these kind of specialized applications are where LLMs will prove to be useful.
Mystified, though, at the rush to deploy them in search engines where the reputational risk is much higher.
Hugh Hewitt proving once again that even GPT-2 can write less nonsensical columns than he can.
@chockenberry Apple's reluctance to develop decent System Settings search is an ongoing mystery.
@ct_bergstrom $5.37B, Carl. The Times style guide has very high standards on precision.
@ct_bergstrom Anyone else notice the problem with the $1.4B metaverse real estate revenue statement in the article?
If fake real estate grows by $5.37billion I will pay each of you a buck.
@amyfou Yes, another embarrassing Times story. https://qoto.org/@twitskeptic/109899110134926201
@ct_bergstrom Different subject, different organization, similar bullshit problem. https://www.nytimes.com/2023/02/19/realestate/metaverse-vr-housing-market.html
Unprofessional data wrangler and Mastodon’s official fact checker. Older and crankier than you are.
Joined Nov 2022
