Ben Serridge

Curious skeptic by nature, product manager by trade. Did AI stuff back when it wasn't called that. Lived in Mexico & Brazil, saudades. Running out of sp...

[Ben Serridge @bserridge]

@simonw I (well, Claude, actually) added a hook in my settings.json that appends to a PROMPTS.md file after every prompt. Seeing this thread here, I think I'm going to add the response as well so that I always have the full transcript available.

[Ben Serridge @bserridge]

@brookejlacey @clairevo Love the list! For me, too, the definition of tools was counterintuitive. I think of tools as a way for the AI to communicate with outside systems. By analogy, we don't define a hard drive as "a way to get information in and out of a computer" (even though it is).

[Ben Serridge @bserridge]

Interesting choice my phone just presented me with...

[Ben Serridge @bserridge]

"Never trust anything that can think for itself if you can't see where it keeps its brain." - @jk_rowling - More than 25 years before the launch of ChatGPT!

[Ben Serridge @bserridge]

@zhuokaiz @ylecun Where does Waymo's world model sit in this taxonomy?

[Ben Serridge @bserridge]

@SomTheBuilder @akshay_pachaar I had the same question!

[Ben Serridge @bserridge]

Sorry, but if you call a white paper published by a tech company a "banger", that's an automatic unfollow.

[Ben Serridge @bserridge]

@jetpen @karpathy He's not trying to make the model more useful, the whole point of this model is that it's a tool for learning about how LLMs work under the hood. Demonstrating on this classic failure mode (understanding why simple tasks fail at the implementation level) is super helpful.

[Ben Serridge @bserridge]

@davidretegan @antonhowes @paulg Google's AI research was published.

[Ben Serridge @bserridge]

@shreyas One reason feedback can feel like a gift is that one almost never receives it. Or at least, not the "constructive" kind. After a talk, for example, the people who talk to you are the ones who liked it. Those who didn't rarely say anything, so there's no actionable feedback cycle.

[Ben Serridge @bserridge]

@clairevo @elawless "no designs to get pixel perfect" <-- this alone is a huge productivity unlock!

[Ben Serridge @bserridge]

@jasondebolt @paulg The "don't say anything" part of this advice has never played out well for me...

[Ben Serridge @bserridge]

Well, if it makes you feel any better, at least the Warriors are now above 500.

[Ben Serridge @bserridge]

United just wrote me saying "You deserve some more PQP". 🤣🤣🤣

I hope their localization team didn't just straight up translate to Portuguese. 🤦‍♂️

(For the gringos / uninitiated: https://t.co/RpOtYDo7xo)

[Ben Serridge @bserridge]

Helpful discussion in the replies...

[Yann LeCun @ylecun]

Genuine question about image generation:
If someone uses a generative AI tool to produce an image that is substantially similar to a copyrighted piece (a drawing, painting, movie screenshot, etc), who should be liable for copyright infringement?
Should it be:
A. the company producing the tool?
B. the (human) creator of the piece?
C. the person or entity posting/publishing the piece?
D. the communication platform through which the piece is distributed?

1. Copyright law protects against unauthorized exact or near-exact copies of a painting, photo, movie, or other visual piece.
2. When a person distributes a sufficiently similar copy of an art piece, it's a violation of copyright regardless of the tools and process used to produce it.
3. The liable person is the person *distributing* the piece, not the artist, and not makers of the tools.
4. Image generation systems are trained to generate images that are on the "manifold" of nice-looking images. Obviously, the training images are on this manifold.
5. Hence, sufficiently detailed prompt will produce images that are substantially similar to images from the training set. It is not at all surprising that a prompt like "The Batman movie, rooftop scene, screenshot, 4k..." will produce an image very similar to an actual screenshot from the movie.
6. Whether using publicly available-yet-copyrighted screenshots and other materials as part of the training set constitutes a violation of copyright is a separate question. As far as I can tell, this question is not legally settled in the US.

[Ben Serridge @bserridge]

@JoannaStern First question: What does ADS stand for?

[Ben Serridge @bserridge]

This webshite is for a city at the heart of Silicon Valley.

[Ben Serridge @bserridge]

@clairevo +1 postmortems are useful when there's clarity as to exactly who did what, when. "Blameless" refers to the psychological safety needed for real transparency, blameless by obfuscation / anonymization is not actionable.

[Ben Serridge @bserridge]

"Dad, don't be silly, that's not Dale, that's Chip, Dale is the clumsy one with the red nose!"

[Ben Serridge @bserridge]

Nice touch at SFO, water to fill up your bottle right after you come through security (which requires you to first empty the bottle).