isaacfrond

There lots of knowledge types where Anki might not be the ideal choice. For example,

- itemized lists of information. E.g. The 5 reasons for the civil war, type of questions. What to do, if you miss one item? Your card as a whole gets over exposed, while the missed item might get underexposed.

- Related information, e.g., history dates. When using Anki the students tends to latch on to incidental facts, e.g., Treaty of Paris, ah I remember it has two repeating digits, yes, 1899. While it would be more useful for a student to think, ah, that was after the Cuban independence struggles in the 1890's.

Any user will recognize that this effect is very strong. You tend to remember cards by the most trivial of things.

- Hierarchical knowledge. Things like chess openings. How do you put that in Anki cards. It's all a kludge. Where to put the variants, etc.

- Knowledge networks. Things like medical information (where Anki is hugely popular). But typically, you get cards with massive amounts of information because you have lots of linked information, (symptoms, causes, treatment, pharmaceuticals, etc. )

- Even in language learning. We have the useful fiction that there is a 1-1 relationship between words in two languages, but there almost never is.

- Learned knowledge tends to be linked to Anki. It happens often that you can remember things when doing Anki, while being at a loss when needing the information in the real world.

Add to that, that Anki does nothing for conceptual understanding. You really need to learn a subject before memorizing it, but in the learning phase Anki is not helpful.

So in short, yes, Anki is the best tool there is to help learning, but I'm sure better tools can be made, especially when targeted to a particular knowledge field.

from the article

* OpenAI has not elaborated about the reason for its sudden decision. ChatGPT is already blocked in China by the government’s firewall, but until this week developers could use virtual private networks to access OpenAI’s tools in order to fine-tune their own generative AI applications and benchmark their own research. Now the block is coming from the US side.*

If chatgtp is blocked anyway, there little advantage to allow researchers access but bekng able to monetize.

tl;dr

German sources say they did not fully test the functionality and deployment of the satellite antennas on the ground.

There is no published article corresponding to that press release.

There is only this published abstract: https://www.abstractsonline.com/pp8/#!/20343/presentation/37...

(which does not contain the 91% by the way; though it is reported widely. perhaps it was in the presentation).

Here is a page with expert reactions (most are skeptical)

https://www.sciencemediacentre.org/expert-reaction-to-confer...

tl;dr: I don't believe a word of it.

from the newsroom article:

People who followed a pattern of eating all of their food across less than 8 hours per day had a 91% higher risk of death due to cardiovascular disease.

That just fails the laugh test. That can't possibly be true.

huh, is that true? Is that class of computable function with a halting guarantee strictly smaller than the class of computable function without one? <citation missing>

Well, that's about time. Although, I'm pessimistic that the law can stem the current tide of research fraud, at least it's something.

tl;dr

"We stress-tested our university’s examination system against AI cheating in a controlled experiment,” at the School of Psychology and Clinical Language Sciences.

His team created over 30 fake psychology student accounts and used them to submit ChatGPT-4-produced answers to examination questions in five undergraduate modules, covering classes needed during all three years of study for a bachelor’s degree in psychology. The assignments were either 200-word answers to short questions or more elaborate essays, roughly 1,500 words long.

Shorter submissions were prepared simply by copy-pasting the examination questions into ChatGPT-4 along with a prompt to keep the answer under 160 words. The essays were solicited the same way, but the required word count was increased to 2,000. Setting the limits this way, Scarfe’s team could get ChatGPT-4 to produce content close enough to the required length.

Even with no editing or efforts to hide the AI usage, 94 percent of those went undetected, and nearly 84 percent got better grades (roughly half a grade better) than a randomly selected group of students who took the same exam.

“We did a series of debriefing meetings with people marking those exams and they were quite surprised,” says Scarfe. Part of the reason they were surprised was that most of those AI submissions that were detected did not end up flagged because they were too repetitive or robotic—they got flagged because they were too good.

It's may favorite example of unintended consequences.

FDA orders bakers to list allergens. There are stiff penalties if you are allergenic but don't list the allergen.

So? Bakers claim their product contains allergens, even if they don't.

FDA doesn't like it.

So? Bakers add allergens on purpose, so they can rightly claim it contains them.

Let's do it one more time.

Correlation != Causation

Could the causation conceivably run the other direction? People with beginning dementia exercise less? Yes, that is quite conceivable actually.

Could there be a common cause explaining both? Yes, if you are in good health, with good genes, a good job, leisure time and money, then probably you are the type of person who'll live longer, sport more, and be cognitively healthy for a longer time. It doesn't mean the causality runs in the other direction.

In other words, this study shows what we all know. The 'good' people have all the good stuff. But what we want to know, will an intervention that causes people to exercise more do any good?

Chrestomathy:

1 : a selection of passages used to help learn a language

2 : a volume of selected passages or stories of an author

I genuinely hope Stonehenge will survive the attack unharmed.

But you are not going to get attention for your cause without making a big fuss. Have you considered the possibility that they are just right in what they say?

It even works on LLM. If you ask an LLM to continually repeat a word, it can break its training and start to regurgitate training data.

https://www.pcmag.com/news/this-silly-attack-reveals-snippet...

The only solution at the moment appears to detect the behavior and to stop the LLM from doing it.

from the article

The model considers vast amounts of data, including tax and property ownership records, agricultural maps, data on soil fertility, logs from the military and emergency services of where bombs and shells have landed, information gleaned from satellite images and interviews with local civilians and the military. Even climate change models and data on population density derived from mobile phone operators could be assessed. The AI then weighs factors such as civilian safety and potential economic benefits to determine the importance of a given piece of land and how urgent it is to make it safe.

But with so many factors to consider, you can prioritize any area of land simply by weighing them differently.

Also since when is weighing multiple factors Artificial Intelligence? I mean, this is moving the goalpost, but moving them closer not further...

could be lip balm addiction. It's a thing apparently

I don't see any example solutions in the paper? If it is solving olympiad level problems, I want to see some!?

The prompt was:

I strongly suspect the story below was written by or with heavy use of an AI. write a report with citations from the article to argue my case.

and then copied the first half of the article

Based on the article provided, several elements suggest that the narrative could have been written or heavily influenced by an AI. Below are key points from the article that support this suspicion, each backed by direct citations:

1. *Generic Language and Lack of Specific Detail*: The article describes Kimberly Gasuras’s experience with broad, generalized statements that lack specific, nuanced detail that a human writer with deep knowledge might include. For instance, phrases like "I don’t need it," and "How do you think I did all that work?" are rather cliché and could indicate AI usage due to their non-specific nature.

2. *Frequent Mention of AI and Related Technologies*: The story frequently references AI technologies and tools, which might be a characteristic of AI-written content trying to maintain thematic relevance. The tools mentioned, such as "Originality" and others like Copyleaks and GPTZero, align closely with typical AI text outputs that often include relevant keywords to boost perceived relevance and accuracy of the content.

3. *Narrative Coherence and Flow*: The narrative flows in a structured manner typical of AI outputs, where each paragraph introduces new information in a systematic way without the nuanced transitions we might expect from a seasoned journalist. This can be seen in transitions like, "It was already a difficult time. Then the email came." This kind of straightforward sequencing is common in AI writing.

4. *Absence of Emotional Depth or Personal Insight*: Despite discussing a personal and potentially distressing situation for Gasuras, the article does not delve deeply into her emotional response or provide personal insights that a human writer might include. The statement, "I couldn’t believe it," is as deep as it gets, which seems superficial for someone discussing their own career challenges.

5. *Repetitive and Redundant Information*: The article repeats certain themes and statements, such as the reliability issues of AI detectors and the impact on personal livelihoods. For example, the repetition of the impact of AI on writers and the functionality of AI detectors in multiple paragraphs could suggest an AI's attempt to emphasize key points without introducing new or insightful commentary.

6. *Use of Industry Buzzwords and Phrases*: The language includes buzzwords and phrases typical of AI-related discussions, such as "AI boogeymen," "peace of mind," "proof," and "accountability." These terms are often used to artificially enhance the thematic strength of the content, a common technique in AI-generated texts to align closely with expected keyword density and relevance.

These elements collectively suggest the possible use of AI in crafting the article, particularly in terms of the language used, the structure of the narrative, and the absence of deeper, personalized insights one would expect from a human writer discussing their own experiences.

Exactly. Because I'm sure that the minute some program aces the ARC test, we'll all say, ahhh, but that, that wasn't real intelligence. And they would be right, if you solve the ARC test, you can do ARC like puzzles. Say something about your reasoning abilities I guess, but it surely does not say you have super human intelligence.

tl;dr it's about progress on the twin prime conjecture.

For a second you think it about the 'little-known mathematician' Yitang Zhang. But before the article is out, it's Terence Tao and James Maynard, ofc.

Isn’t HOL a bit oldfashioned now that we have Lean?

The human visible frequencies also match the frequencies where the sun has most of its output.

Moreover, they also match the frequencies for which the atmosphere is transparent.

tl;dr

The case arises from years of litigation between Blacklock’s Reporter, a paywalled news service based in Ottawa, and the Canadian government.

Blacklock’s continued to press ahead with other cases with the latest decision involving 15 articles that were distributed to media personnel at Parks Canada. The department had an individual subscription to the service, but Blacklock’s argued that allowing anyone other than original subscriber to access articles constituted copyright infringement.

Blacklock’s position on this issue was straightforward: it argued that its content was protected by a password, that passwords constituted a form of technological protection measure, and that fair dealing does not apply in the context of circumvention. In other words, it argued that the act of circumvention (in this case of a password) was itself infringing and it could not be saved by fair dealing.

The Federal Court disagreed on all points. It cited with approval CIPPIC’s argument that “the TPM provisions do not apply to restrain fair dealing; using a validly obtained password to access content is not circumvention.” Further, court noted “how the password was obtained is significant as this may prevent a user from invoking the fair dealing provisions of the Act.” In other words, not all password sharing will qualify as fair dealing. This decision is not a licence to simply share passwords with no consequences, but rather affirmation that passwords are not a technological protection measure and that using validly obtained password does not limit fair dealing rights.

It fooled me. Turns out I had accidentally entered my company into a prime subscription. It could have been embarrassing, but it turned out that my boss did the same thing.

Took only a few emails to get us of it, and reimburse the costs, fortunately.

Most of the factors in ranking a page are no surprise. But i was surprised that having Product reviews on your site is apparently a demotion? Surely, many people are searching to find just that?

If anybody is interested, here are the papers themselves. I wish you a lot of luck (I have a phd in algebra, but it is unintelligble to me).

https://people.mpim-bonn.mpg.de/gaitsgde/GLC/

In my own time as a high school student, I would come up with alternative solutions all the time. I'd always get full marks, though sometimes I'd have to explain it to the teacher afterwards.

Recently, I did the same test as my daughter did, who is now in high school. It was a bet kind of thing. So anyway, after doing the exam, I figured I'd have 9 out of 10 marks. I though t I'd lose some marks for some clumsy arithmetic, but it went pretty well (I thought).

So my daughter then grades my test and carefully explains that maybe I might just have a passing grade (6 out of 10). Ok. whatever. That can't be right. So she takes my test to the teacher to have her grade it as well. The teacher's verdict: clearly, easily, a failing grade. Maybe 4 out of 10 marks.

Turns out getting the right answer in a clever way counts for nothing. What you are being tested on is: can you reproduce the algorithm we are teaching you in school. As I solved most of the problems in a different manner, I consequently got almost none of the marks. I suppose, that if you know and accept this, then you can accept the result as well.

It's disappointing I guess, but certainly it makes math more accessible. I suppose there are pros and cons.

Your son wasn't tested on cleverly getting the right answer. The test was: can you reproduce this algorithm. He couldn't.

tl;dr

The facts you learn at school are surprisingly ill remembered. But stuff learned by cultural osmosis is remembered very well. So why go to school?

On this one, it is OpenAI who are terrible and acting like spoiled kids.

They approached Ms. Johansson, she said no, and they went ahead and did it anyways.

I remember installing a text only browser to visit a 'website' at CERN on this new thing the 'world wide web'. I browsed their website for like 30 seconds, and deleted the browser. That will never amount to anything. I was sure of it.

My children delight in this story. Though they don't really believe the text-only part. That's got to be a fabrication on my part.