Remember the outcry from our entire political class back in October when it was reported that Israel had hit 1 hospital? They immediately screamed fake news. Now it’s 8 hospitals, and they’ve all gone silent.
3 replies
153 reposts
416 likes
Reposted by Tyler
It’s pretty telling that all the folks funded by billionaires to speak out against cancel culture and allowing all points of view on campus have gradually but consistently picked a time to go “surprise it was Nazi stuff all along!”
10 replies
76 reposts
384 likes
Sorry, wrong Greene book - but still, this guy's list should've been rejected on principle
0 replies
0 reposts
0 likes
Lucy Sante's and John Irving's are interesting, but everyone else's lists are mostly homogeneous, inoffensive, or baffling (the president of Wesleyan picks Thinking Fast And Slow? what on earth is The 48 Laws of Power doing anywhere here?) www.nytimes.com/interactive/...
1 replies
0 reposts
0 likes
Sitting down and reading these things really underline this bsky.app/profile/afiv...
0 replies
0 reposts
1 likes
From a similar paper from OpenAI - this is just an outright falsehood, right? Are the questions from different years of AP STEM exams 1) significantly different from each other and 2) unique from any other published problems? arxiv.org/pdf/2305.20050
2 replies
1 reposts
2 likes
From a similar paper from OpenAI - this is just an outright falsehood, right? Are the questions from different years of AP STEM exams 1) significantly different from each other and 2) unique from any other published problems? arxiv.org/pdf/2305.20050
2 replies
1 reposts
2 likes
And of course, it would only be really helpful in terms of existing math - the AlphaGo model they're trying to take inspiration from is neat, but it's not going to lead to what this guy claims
1 replies
0 reposts
1 likes
Which would indeed be helpful, in that the model would have its own little automated search engine to check against, but it's not really an improvement in 'reasoning' so much as an improvement in tilting the table to help it
1 replies
0 reposts
1 likes
All this was triggered by wondering what happened to the Q* thing that OpenAI people were supposedly freaking out about nine months ago, and according to this piece, it's literally just trying to combine the existing GPT with a verifier model, GAN style www.understandingai.org/p/how-to-thi...
1 replies
0 reposts
1 likes
2. In case you were wondering - no, they didn't include the actual generated paragraph, just the 'plans'
1 replies
0 reposts
2 likes
1. The two 'plans' are exactly the same, but the second one makes an assertion at the beginning of the first example, which leads it to pick that one as more coherent even though the other three still have nothing to do with it
(because presumably, it resembles the data more with the assertion)
1 replies
0 reposts
2 likes
Getting round to reading the Train Of Thought paper, where LLMs are basically made to pick and choose from sets of their own generations in order to improve accuracy, and this graph of the 'creative writing' test is an illustration of why I don't trust these people with this arxiv.org/pdf/2305.10601
1 replies
0 reposts
3 likes
Reposted by Tyler
Was going to reply "I'm a data scientist, and am not a fan." reading this, then clicked through and scrolled up to get the context. Guess what they're selling.
3 replies
11 reposts
37 likes
Woodsy Owl swapping minds with George Will has produced exactly the results you would expect
0 replies
0 reposts
0 likes
In their newly converted zeal and previous secular ignorance, Archie and Jughead began praying to a picture of Topol
0 replies
0 reposts
0 likes
Reposted by Tyler
2 replies
1 reposts
10 likes
Reposted by Tyler
3 replies
1 reposts
8 likes
'partisan (corporations are evil)'
0 replies
0 reposts
0 likes
Reposted by Tyler
Good reporting on an entirely predictable problem. Fun fact: Airbnb also covers up thousands of sexual assaults every year! Its internal safety team has "cool-down rooms" to take a break from the trauma of what they hear.
It's a bad company, a bad industry, and people should just stay in hotels.
20 replies
444 reposts
878 likes
Reposted by Tyler
do you think they bond over amassing power and wealth off others' hard work they didn't pay for?
7 replies
30 reposts
168 likes
According to the recent Atlantic article, 'using more AI' is Bezos's number one demand for the Post, so
0 replies
0 reposts
1 likes