@hazytalks Profile picture

Haziq

@hazytalks

Optimizing one Model at a time. #ML

Similar User
Mathieu Alain photo

@miniapeur

Yandex photo

@yandexcom

Sander Dieleman photo

@sedielem

BartStuck photo

@BartStuck

Native American Pride🇺🇸 photo

@NativeA29927707

Badri Berulava photo

@BerulavaBadri

USS Liberty Veterans Association photo

@usslibertyvets

Marcus Zimmermann photo

@ZimmerMar68

NACTA photo

@nactateachers

Guneet Singh Kohli photo

@guneetsk99

Faiqa Anbreen photo

@faiqanbreen

Shuchita Jha photo

@Shu_cheetah

Lubingzhi Guo photo

@lubingzhiguo_

JoJo Collins photo

@jojo_mama922

Pinned

The paradox with evolutionary systems is that legacy learning is hard to update fast, which is both a good thing and a bad. Bad that it can't improvise to updated data fast , good that a small unintended change won't significantly change the outcomes...


Haziq Reposted

# RLHF is just barely RL Reinforcement Learning from Human Feedback (RLHF) is the third (and last) major stage of training an LLM, after pretraining and supervised finetuning (SFT). My rant on RLHF is that it is just barely RL, in a way that I think is not too widely…

Tweet Image 1

Haziq Reposted

Huge congrats to @AIatMeta on the Llama 3.1 release! Few notes: Today, with the 405B model release, is the first time that a frontier-capability LLM is available to everyone to work with and build on. The model appears to be GPT-4 / Claude 3.5 Sonnet grade and the weights are…


Haziq Reposted

New (2h13m 😅) lecture: "Let's build the GPT Tokenizer" Tokenizers are a completely separate stage of the LLM pipeline: they have their own training set, training algorithm (Byte Pair Encoding), and after training implement two functions: encode() from strings to tokens, and…

Tweet Image 1

Haziq Reposted

I drew this morpho butterfly with mathematical equations.

Tweet Image 1

Haziq Reposted

Introducing AlphaGeometry: an AI system that solves Olympiad geometry problems at a level approaching a human gold-medalist. 📐 It was trained solely on synthetic data and marks a breakthrough for AI in mathematical reasoning. 🧵 dpmd.ai/alphageometry


Haziq Reposted

# On the "hallucination problem" I always struggle a bit with I'm asked about the "hallucination problem" in LLMs. Because, in some sense, hallucination is all LLMs do. They are dream machines. We direct their dreams with prompts. The prompts start the dream, and based on the…


Haziq Reposted

Google (DeepMind) releases AI model Gemini. There is no turning back now, we are in for one mad ride. The multi modality, and fluidity of the model is super clean. My jaw dropped at 4:24 seconds A thread...


Haziq Reposted

In my decade spent on AI, I've never seen an algorithm that so many people fantasize about. Just from a name, no paper, no stats, no product. So let's reverse engineer the Q* fantasy. VERY LONG READ: To understand the powerful marriage between Search and Learning, we need to go…

Tweet Image 1

Haziq Reposted

Runway's new update is producing incredible AI videos. It's a significant leap forward. As someone who's worked with a famous Hollywood producer and dreamed of creating films, I find this so exciting. We're witnessing the birth of a new era in film. Here are the best examples:


Haziq Reposted

OpenAI (vision & voice APIs) for sports narration:

GPT-4V + TTS = AI Sports narrator 🪄⚽️ Passed every frame of a football video to gpt-4-vision-preview, and with some simple prompting asked to generate a narration No edits, this is as it came out from the model (aka can be SO MUCH BETTER)



Haziq Reposted

Turns out, fitting a curve to a dataset produces a model that only generalizes to that specific data distribution -- big if true.

New paper by Google provides evidence that transformers (GPT, etc) cannot generalize beyond their training data

Tweet Image 1


Haziq Reposted

i expect ai to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes


Bias and Variance takes little time to learn but lifetime to master, and at the end it will be at most somewhat an optimized outcome. Applies pretty much to everything in life.


Will be quite useful for case based algorithm development and selection and training. Somewhat like more sophisticated and generalized version of cross validation at a scale for new models, put in rustic terms. Super Amazing

🎉Just released: Eureka!, a new AI agent that uses LLMs to automatically generate algorithms to train robots to accomplish complex tasks. 👀 The #NVIDIAResearch paper includes the AI algorithms and how to experiment with Eureka using NVIDIA Isaac Gym. 👇 nvda.ws/3PWKlk4



Some Beautiful Abstractions!! #DALLE3

Tweet Image 1
Tweet Image 2
Tweet Image 3
Tweet Image 4

Haziq Reposted

Computer vision has been solved.

Tweet Image 1

Haziq Reposted

More and more Plus users are getting the new ChatGPT Vision.. People are posting impressive use cases. 9 really good ones:


Haziq Reposted

Dynamic 3D Gaussians: Tracking by Persistent Dynamic View Synthesis dynamic3dgaussians.github.io We model the world as a set of 3D Gaussians that move & rotate over time. This extends Gaussian Splatting to dynamic scenes, with accurate novel-view synthesis and dense 3D trajectories.


Haziq Reposted

Around 31 hours since Adobe Max dropped the new Firefly 2 models... People are WOW-ing at the real photorealism! 10 examples: (How are these not real photos?)


Haziq Reposted

This is such an interesting work. Video diffusion model is being used as a data-driven physics simulation, in which an agent can plan, explore, and learn optimal actions without touching robot hardware or causing harm. LLM is not only an OS, but also a full reality simulator.

Introducing Universal Simulator (UniSim), an interactive simulator of the real world. Interactive website: universal-simulator.github.io Paper: arxiv.org/abs/2310.06114



Loading...

Something went wrong.


Something went wrong.