Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Blog Pricing

Atrioc

๐Ÿ‘ค Speaker
3331 total appearances

Appearances Over Time

Podcast Appearances

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

So in both angles, we're like suffering.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

And then there's just

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

going to be this massive slop that is being pushed out as everybody becomes dumber, but is better at producing slop.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

We're getting really efficient at being dumb slop generators.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

prime prime said this he reviewed the two most recent models and he's like the people who weren't producing good code are just producing bad code faster but at a gargantuan rate yeah and then one final thing i'll say there's this excellent youtube video called can ai pass freshman cs so this is graduate ta at cornell i guess we can put links to these youtube videos in the description but graduate ta at cornell um last semester did a study where he basically had

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

the three of the best coding models, Gemini, Claude, and ChachiBT, just follow his computer science class that he helps teach to see, is it actually like PhD level coding?

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

Because Sam Altman keeps saying they're now like PhD level coders.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

So the synopsis of this hour and a half video, these AIs ace a whole bunch of really hard things and screw up things that no student would ever get wrong and no human would ever get wrong.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

And the final grades were Gemini and Claude both got a C plus in this class.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

ChatGPT got a B plus.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

All three of those models, which are the newest models that are mega hyped, that are told, like we're told constantly, these are PhD level smartness.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

Look at all these graphs.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

Every single one scored below the median student grade.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

These are supposedly PhD level CS coding capabilities and they're doing worse than a freshman computer science student.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

And so he ends with a quote that I really like.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

He talks about how there's all these mistakes and then he says, but the complete void of creativity that permeates everything that AI touches makes the output so much less than anything we got from the real people we got taking the course.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

If you put the bland, broken output from the LLMs alongside the magic the students worked, it really isn't even a comparison.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

So can an AI pass a first semester CS class?

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

Sure, it can technically.

Lemonade Stand
The Biggest Merger Ever | Ep. 050 Lemonade Stand ๐Ÿ‹

But looking at our students, I'm still feeling OK.