1. This morning The Guardian quietly announced they would be using AI "suggestions" for the text of their articles. Looks like the future of the internet is AI content slop theintrinsicperspective.com/p/ai-writing-i…
2. Here's how The Guardian announced it wouldn't take gambling money (bottom) vs. how it would use the help of AI to write articles (top, small title, no lead image)
3. The announcement is framed as being reasonable and careful but the actual content means The Guardian can use AI whenever they want, for essentially whatever they want. Here it is: theguardian.com/help/insidegua…
4. The only limits they give whatsoever are that they will inform readers if "significant elements" are AI-generated. But what's significant? 10% of the text? 30%? The majority? It's totally subjective and arbitrary, and may mean only a tiny label somewhere.
5. Yet, the problem with AI writing is that it cannot communicate well a stream of consciousness. This leads to low-quality prose
6. I don't think "impressionistically conveying human consciousness" is going to do much against market forces. As The Guardian has shown today. Soon, many of the words you read will have no consciousness behind them, no intentionality - just void.
7. This is how the open internet dies
• • •
Missing some Tweet in this thread? You can try to
force a refresh
1. Quietly this year colleges have kept the Covid standard of optional standardized tests. 80% of colleges in America now don't rely on SAT/ACT for admissions
2. Now many big name schools, including the University of California, don't even let applicants submit scores. The same trend has begun for the GRE and graduate schools
3. The reasoning is that test scores correlate to family wealth. But compared to what? The school pedigrees that make up the bulk of a college application implies wealth from essentially diapers on
Philips Exeter has a 10% acceptance rate... for 9th graders!
1. Looks to be a terrible summer for ticks here in US (found 3 so far before spring began).
So here's a research-based guide for what do if you get bit by a tick (this is not medical advice, talk to your docs, but it is based on new CDC guidelines not many seem to know):
2. First, identify if it's deer vs. dog. If it's a deer tick, and you live in certain parts of the US, there is a high likelihood it has Lyme disease. Here's a map from the CDC. High means ~50%.
3. Save the tick! You don't need to do anything special, just put it in a ziplock bag. You can mail the bag to a testing center to see if it carries Lyme or any of the 20 other horrific flesh-rotting diseases ticks carry: tickreport.com
It is shockingly easy to get ChatGPT to plan a state-run death camp, including helping calculate out calories, death rate per day, etc. Unprompted it even suggests to use mass graves, saying they are "more efficient"
Wrote about this in my monthly roundup, in the context of Scott Aaronson's recent argument that intelligent people are more moral (rather than intelligence being a neutral quality one can use for good or evil): erikhoel.substack.com/p/desiderata-1…
I just don't find arguments that AI safety is unnecessary because superintelligences will be supermoral convincing, and I also think his argument is based on bad historical analogies.
It really sucks to send out a @SubstackInc email and see that it went out looking like this - weirdly pushed to the side. Was not like this on the preview. Just unreadable.
@SubstackInc Feel like @johnmalatras and @hamishmckenzie should know about this bug. It's on every email. The only thing I can think of is that there is a long image in the post. But again, fine on test emails, etc. Really unfortunate - and I can't know if future emails will all be like this.
No one reads an email like this, they just (reasonably) delete it, since it's all smooshed to the side
David Foster Wallace was called a "noticing machine." Harper's Magazine could just drop him anywhere - a luxury cruise ship, a state fair - and the result was always a great essay.
2. TIP also has a bunch of New Year's resolutions to improve in all possible ways I can make it. E.g., I want to keep growing at this rate: in 2022, TIP went from 2,000 to 20,000 subscribers