@antigravity One caveat. It makes the Apple silicon turn on its fan, which is usually quite unless naughty apps like this make the app run in the background and use battery all the time! fix it @antigravity@_mohansolo
Cool. @antigravity agent & I blogged about @antigravity using @antigravity agent manager.
That is so anti-sentence in using Antigravity 3 times!
I could have used @antigravity IDE to initialize the directory and watch it modify stuff though!
@karpathy Just try Pixel phone for one week. It's best in protecting from spam. Not only that it has features where it can take calls on your behalf which annoys the spammers.
It's useful when holding with customer service as well.
One of the thing @ChatGPTapp is really good at creating coloring pages for kids based on given image. I tested @grok and @GeminiApp and ChatGPT is hands down every time! 2nd place probably goes to Gemini, but instruction following is poor. Grok is out of the question!
@GeminiApp Please add the "mute" button on the Gemini Live app.
When you are on a call, it's a no-brainer not to have a "mute" button. And it is even more important when you are talking to an AI model that is always on voice Assistant which accidentally thinks that you are talking with some background noise and interrupts the conversation flow. In fact, apart from the existing buttons, it should have all the abilities to add more people like a group phone call. Think of it like a Google Hangouts/Chat call but you can always call Gemini live. At the same time you can invite your friend also to the conversation and then all of a sudden you have an AI and humans in the conversation mix.
Anyways, I think Gemini live is totally unusable until it has the mute button. Currently we can only pause but that's not doing the right thing. If you look at Grok voice mode, they have done it correctly at least the mute portion.
@karpathy ICL: In-Context Learning 🤔
How about calling this as "Priming" the model.
Essentially we are changing the course of *Activations* by changing the sequence of tokens without actually modifying the model weights.
@_philschmid@ByteDanceOSS In my experience BAML is a better way to write and enforce structured responses and in general a better way to call LLMs as you put software pieces together. It decouples structured responses API support from LLM providers, multi language support etc.
@karpathy imagine injecting a bio fluid which is extension of model to give visual perception of model running in data center! direct brain interoperation without any additional compute device in between. maybe a neuralink plugin to foundation models, thats personalized for individuals.
@julien_c@yagilb I recommend BAML as an abstraction to structured data extraction from models. It's intuitive, transparent and minimal. Supports any model
https://t.co/RoHVjN3his
@boundaryML@hellovai
@rauchg@turkticaretnet@Cloudflare also its better to have a DNS level protection to block these type of attacks using @AdGuard and/or @NextDNS. I think the other day I accidentally clicked a phishing link and one of this protected me by blocking the URL(I have AdGuard on home wifi and NextDNS on mobile network)