Avery Lamp
40 posts
Avery Lamp
@AveryLamp
:), doing stuff
San Francisco, CA Katılım Nisan 2011
1.2K Takip Edilen761 Takipçiler
Avery Lamp retweetledi
Gemini 3!
An AGI model is supposed to be superhuman at most things humanity cares about, all in one! This has been one of the early philosophies we've had with Gemini, studying modality transfer and scaling very carefully.
English
Avery Lamp retweetledi
Avery Lamp retweetledi
Introducing Gemini 3 Pro, the world's most intelligent model that can help you being anything to life.
It is state of the art across most benchmarks, but really comes to life across our products (AI Studio, the Gemini API, Gemini App, etc) 🤯
English
Avery Lamp retweetledi
Congrats @aquariusacquah and @michael_silver on the launch! Exciting stuff
ken@aquariusacquah
Humans invented software to free ourselves from manual work. Somehow, we still waste billions of hours clicking across websites. Today, @joinkaizen launches to fulfill that original promise. Announcing our $4M raise from @8VC and the public launch of the Kaizen platform.
English
DSL’s have a very natural fit with LLMs where LLMs are better at coding than language so they learn a DSL much better than abstract English concepts. Super powerful to teach an LLM a DSL or partial DSL to control behavior. Function calling is essentially a one-time use DSL with PMF
English
OK, on whether DSLs are a good idea. It comes down to the following:
Either you're implementing a language, or you aren't. If you are, do it right (lex, parse, bind, opt, plan, interpret).
And realize it's probably overkill, and has likely already done that better.
English
Avery Lamp retweetledi
two days left to submit your apps to our startup program Embed!
we're really excited about the quality of applications we've already read through and a bunch of the teams we've met these past couple weeks
get your apps in by tomorrow at midnight PT! DMs open for q's
English
Sora Unreal Engine hypothesis:
A lot video content online is games, so simply training on it leads some generations to look very game-like. Specifically for drone car following shots, those clips are more likely to exist in games because the difficulty of those real world shots.
English
Why has no one re-written Whisper to be `torch.compile` compatible? I'm surprised that there's not an easily accessible version that has fused kernels & high MFU.
English
Avery Lamp retweetledi
1/ Applications are now open for the S24 cohort of Embed, @W_Conviction’s program for early-stage AI-native startups! $150K uncapped SAFE investment, $600K in credits with our partners, and a hand-selected group of peers to accelerate amazing teams.
English
Avery Lamp retweetledi
1/ Amazing job to all the @W_Conviction Embed teams at Demo Day! With Embed, we hoped to create a Schelling Point, a way for people who want to build products on the frontier of AI to find one another. CC: @pranavreddy
🧵on all the startups in the batch:
English
If you want an EA to help with scheduling meetings but can't justify the cost, reply here or tag someone and I'll share our new AI assistant. Only have a few accounts to give out.
I don't currently have an EA and I've been using it for the past month, it's 🤯🔥‼️
English
Just raised series B!! Come join us build the future of computer interaction :)
English
@wmertens @ShaunEls @verge I believe it ~1c per query, but verve said 0.03 cents per query and other nonsense numbers (og post based on this). My guess is like 5 seconds a generation, $8/hr for compute node subsidized heavily by azure. Leads to about a cent per query, without batching. Same 100k/day number
English
@AveryLamp @ShaunEls @verge Wait I thought it was cents per query? Here's an estimate of 3M$/month twitter.com/tomgoldsteincs…
Tom Goldstein@tomgoldsteincs
I estimate the cost of running ChatGPT is $100K per day, or $3M per month. This is a back-of-the-envelope calculation. I assume nodes are always in use with a batch size of 1. In reality they probably batch during high volume, but have GPUs sitting fallow during low volume.
English
