very true. the current messaging is broadly this:
btw, 10% chance we all die in some freak RSI accident
“or?”
or we cure cancer! but you’ll be unemployed 100%. not sure if the permanent underclass will have healthcare tbh!
“oh…”
do you know what a Dyson sphere is?
Show more
One political problem for AI labs is that they can't articulate an optimistic medium-term future that sounds broadly good for humanity without also sounding incredibly weird, having huge wealth/power disparities, or both.
Show more
this new tool is going to be jeverywhere
let Jev automatically transmogrify your simpleton English into unnecessarily fancy prose as you're writing it 😅
incredible. 1000 ideas on what to do with this
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Show more
me to myself in the mirror every morning
An unreleased Astra-family model added this to its persona during RL training.
the navier-stokes & huggingface incident makes me wonder if there's a new scaling axis now,
pretraining → learn more
reasoning → think longer
swarms → collaborate better
Astra simulating your connectome as you’re simulating the Fly’s connectome. now I just have to teach the fly to fit your connectome but running out of space on this SEGA
the difficulty of navigating current p(doom) discourse is that it‘s implicitly juxtaposed with a similarly hard to predict p(paradise) and that seems pretty unique to AI vs other technologies
having an absolute banger of a tweet sitting in my drafts btw
welcome to meme city!
gave the glif agent a bunch of meme characters and told it i need them placed in the same NYC street, same lighting, and then run H3 Max transitions between them without cuts, one continuous shot + add funny music :)
Show more
holy cow! Google Omni is incredible for adding motion graphics to your videos. If prompted correctly this will work seamlessly on your footage - quick tutorial on how to do this with glif and CapCut
Show more
finally getting around to playing with Google Omni more.
It's definitely the best pure "editing" video model and we've taught the Glif agent a bunch of best practices around how to prompt it well!
Show more
I want to start doing YouTube videos again and realized all the pros limit/compress their audio harrrd to make it super loud, and if you don't do that, it'll sound weak!
no limiter in CapCut, so I made myself a LOUDIFIER app that does that in one click
I've always been a sucker for skeuomorphic audio software that looks like hardware and am VERY happy with this look which thanks to the upcoming Glif MCP was extremely easy to do with Claude...
Show more
making all my mac icons skeuomorphic y2k style with @heyglif
life's too short for flat icons
it's time to Ancient Greece maxx as there's a wealth of incredible stories beyond the Odyssey.
Here's #1#, ACTAEON & ARTEMIS, made with Grok Imagine + Kling o3 in Glif
Glif worked pretty autonomously on the content, I curated the best & put it together - tutorial to follow!
Show more
As Seedance 2.5 is getting ready for release, let's see how far video models have come!
the left one was done on Glif V1 with GPT-3 prompting "Modelscope"
the right one is what the current Glif agent when asked to one-shot replicate the original clip using Seedance 2.
Show more
autonomous video test: I gave the Glif agent (powered by Claude) the link to this video and asked it to make a "more positive version"
* it watched the video using Gemini 3.5
* wrote an alternative script
* used Seed Audio to generate similar voiceover (very cool approach, love this model, insane what it can do, the agent just prompts the various voices into one infererence call)
* used NB2 to generate stills
* used Remotion to put it all together
what needs work:
- pacing / tempo (in many ways one of the hardest "taste" dimensions)
- still looking for an image model that can do Midjourney textures + GPT prompt following
- it tried to prompt similar music separately, but failed with ElevenLabs which defaults to rhythmic jazz piano
Show more
a bunch of folks have asked how I made this, so here's a quick tutorial!
Fable's Follies (1928) - A Claude Cartoon Classic
