woof

Summary using Claude 3.7 Sonnet:

"Your Brain On Chat GPT" Paper Analysis

In this transcript, neuroscientist Ashley and psychologist Cat critically analyze a controversial paper titled "Your Brain On Chat GPT" that claims to show negative brain effects from using large language models (LLMs).

Key Issues With the Paper:

Misleading EEG Analysis:

The paper uses EEG (electroencephalography) to claim it measures "brain connectivity" but misuses technical methods EEG is a blunt instrument that measures thousands of neurons simultaneously, not direct neural connections The paper confuses correlation of brain activity with actual physical connectivity Poor Research Design:

Small sample size (54 participants with many dropouts) Unclear time intervals between sessions Vague instructions to participants Controlled conditions don't represent real-world LLM use Overstated Claims:

Invented terms like "cognitive debt" without defining them Makes alarmist conclusions not supported by data Jumps from limited lab findings to broad claims about learning and cognition Methodological Problems:

Methods section includes unnecessary equations but lacks crucial details Contains basic errors like incorrect filter settings Fails to cite relevant established research on memory and learning No clear research questions or framework The Experts' Conclusion:

"These are questions worth asking... I do really want to know whether LLMs change the way my students think about problems. I do want to know if the offloading of cognitive tasks changes my own brain and my own cognition... We need to know these things as a society, but to pretend like this paper answers those questions is just completely wrong."

The experts emphasize that the paper appears designed to generate headlines rather than provide sound scientific insights, with potential conflicts of interest among authors who are associated with competing products.

"1) Python is unreadable."

Would you prefer C or C++?

"2) AI companies are content with slop and do not even bother with clear problem statements."

It's a filter. If you don't get the problem, you'll waste their time.

"3) LOC and appearance matter, not goals or correctness."

The task was goal+correctness.

"4) Anthropic must be a horrible place to work at."

Depends on what you do. For this position it's probably one of the best companies to work at.

It was never about Perl, it was the plethora of alternatives.

Python evolved, PHP had 1000 times more "how to get started" articles, Node happened. And LAMP became the default for noobs.

Nothing beats Maccy.app for clipboard management, but DoubleMemory does look great.

For me, DoubleMemory falls between the cracks: - bookmarks in multiple browsers on multiple devices - Maccy.app for clipboard management on MacOS - Bear.app for notes on MacOS, iOS and iPad

"Made" or "designed"?

Our Tesla Model 3 was made in China…

Seems like the bias is on your part… Which is ironic, given the articles content….

Or do you have _evidence_?

After using the 8GB M1 MacBook Pro, my new job gave me a 16" 32GB MacBook Pro 2,6GHz i7 - and I actually miss my M1!

Battery life and fan is rather annoying now :(

I loved mine at least a year and was quite happy for another year. Some sites worked awesome, others sucked hard due to crazy pay loads.

I blame shitty sites more than Apples architecture :(

Amazon, Google, Microsoft is not a duopoly :)

And while there are probably some services on AWS you can’t find at a competitor - that’s mostly a choice (aka profit margin) made by the competitor.

Apache Solr?

Maybe not "better", but useful and with a Apache 2.0 license :)

She wasn't "forced out"?

"hCaptcha now ruins fifteen percent of the internet"

I loathe captchas, especially Googles who seems to punish my use of privacy extensions (ie ublock origin, Ghostery Lite).

Captcha is a ok tool when you have valid reasons to assume the user is a bot (multiple failed logins, unusual traffic, password resets etc.). Used as a default it only antagonizes users.