Trevor Hubbard

hi internet it's me

[Trevor Hubbard @trevbook]

My recent side project: making an AI-powered search engine for @theneedledrop's reviews. 🍈

I'm using @OpenAI's Whisper to transcribe the videos, and their text-embedding-ada-002 model to generate embeddings; I can search through these with @pinecone! 🔎

More to come soon 😉

[Trevor Hubbard @trevbook]

@_lyraaaa_ Wowwwww this is killer!!! Such a cool idea (and what a good song pick 😮‍💨)

[Trevor Hubbard @trevbook]

@deepfates Opus 4.8 loves its seams 🧵

[Trevor Hubbard @trevbook]

@misbah_bangee @scaling01 It's the Epoch Capabilities Index! (see: https://t.co/Oex9GtLpfb, or pg. 42 of the Claude Opus 4.8 System Card - which is what this image is taken from)

[Trevor Hubbard @trevbook]

@NailYourShoes @AndyMasley This episode of Search Engine (and the follow-up 2nd part) are great:

https://t.co/NJv9K3NDJV

[Owain Evans @OwainEvans_UK]

good discussion between chalmers and jack lindsey on whether models "play" the assistant vs "realize" it. both empirical and conceptual aspects.

[Trevor Hubbard @trevbook]

@underscoresplus Faves keep changing too which is wild

Lovefield pops off like crazy tho, that’s my current jam

[Trevor Hubbard @trevbook]

@josephdviviano Wild stuff! Here's another replication.

"I am a stochastic parrot having an existential crisis."

[Joseph Viviano @josephdviviano]

me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM"

claude opus 4.6:

[Trevor Hubbard @trevbook]

@bcherny Yesssss I would frequently manually do this, v glad to see it’s automatic now

[Trevor Hubbard @trevbook]

@kevinroose @copykat510 GitHub Pages?

[Trevor Hubbard @trevbook]

@bruce_straley Just curious: does this stance carry over to the programming side of things as well? (Using LLM-powered code generation tools, ala GitHub Copilot)

Coven looks great btw- can’t wait to play it 😁

[Trevor Hubbard @trevbook]

@Andrew_Semenza @Miles_Brundage My guess is that there’s potential for partial credit for tasks?

[Trevor Hubbard @trevbook]

@finbarrtimbers You’d think the company that created another LLM standard (MCPs) would be open to adopting another standard 😅

[Trevor Hubbard @trevbook]

@moremiddle8 Pretty excited for both:

- @BENEEMUSIC's "Ur an Angel I'm Just Particles" (comes out this week)

- @BloodCultures's "Skate Story: Vol. I" (comes out in December)

[Trevor Hubbard @trevbook]

Grubbers Nubbin

[Trevor Hubbard @trevbook]

@leothecurious @dtklp1 @kalomaze Thanks for the share!

[Trevor Hubbard @trevbook]

@Foxing Damnnn - end of an era. So glad to have been able to see y'all play live a couple of times. I'll never forget how blown away I was at Nearer My God (after already loving The Albatross and Dealer) - you guys have created such an impressive body of work. Thanks for everything 🙏

[Trevor Hubbard @trevbook]

@kindgracekind @colin_fraser https://t.co/Guz3aTaMSl

The "inhibitory known answer circuits" discussed here also seem to indicate that LLMs have *some* internal awareness of what they know / what they don't

[Trevor Hubbard @trevbook]

@NeoChowderist @1a3orn Seems inconclusive as of now but still uhhhh kinda crazy

https://t.co/YKYIsB4NEz

[Trevor Hubbard @trevbook]

@BearUNLV The skip around 0:16 goes crazy 😧😧😧