imtiaz @imtiazkhan_ds

https://t.co/itEuBAg5Ch linkedin.com/in/imtiaz-khan… Hyderabad, India Joined March 2018

Tweets

551
Followers

23
Following

1K
Likes

279

imtiaz @imtiazkhan_ds

3 weeks ago

Thanks @jeremyphoward for fasthtml Built a Daily AI research feed with arxiv api backend and fasthtml frontend within couple of hours. Details : github.com/ImtiazKhanDS/f…

0 0 0 28 0

Download Image

imtiaz @imtiazkhan_ds

a month ago

behind every man with a dream and a vision there are employees supporting him who have no personal dream and vision.

0 0 0 9 0

My sleep scores during recent travel were in the 90s. Now back in SF I am consistently back down to 70s, 80s. I am increasingly convinced that this is due to traffic noise from a nearby road/intersection where I live - every ~10min, a car, truck, bus, or motorcycle with a very…

1K 816 12K 1.4M 3K

imtiaz @imtiazkhan_ds

7 months ago

LLM models need more tokens to think.

0 0 0 43 0

imtiaz @imtiazkhan_ds

7 months ago

Gold Rush is getting all the GPU's and predict the next token

0 0 0 46 0

imtiaz @imtiazkhan_ds

8 months ago

Bayesian View : A new observation that is consistent with an existing belief could make us more sure of that belief , while a surprising observation could throw that belief into question.

0 0 1 33 0

imtiaz @imtiazkhan_ds

8 months ago

No amount of bananas will incentivise monkeys to do mathematical reasoning. - Hyung Won Chung

0 0 0 26 0

imtiaz @imtiazkhan_ds

8 months ago

Weight Decay Intuition When you do weight decay , you are effectively telling the neural network to distribute weight to all the parameters rather than making some weights to shoot up. This helps in regularisation

0 0 0 24 0

imtiaz @imtiazkhan_ds

8 months ago

0 0 0 10 0

Download Image

imtiaz @imtiazkhan_ds

9 months ago

Tokens are the atoms of large language models.

0 0 0 9 0

imtiaz @imtiazkhan_ds

9 months ago

Transformers (q, k , v) intuition Query : Here’s what I am interested in key : Here’s what I have Value : Here’s what I am going to communicate.

0 0 0 14 0

imtiaz @imtiazkhan_ds

9 months ago

Please stop whatever you are doing and check this out. @jeremyphoward never stops to amaze us. 👌😁 answer.ai/posts/2024-12-…

0 0 0 10 0

imtiaz @imtiazkhan_ds

9 months ago

Geometrically "average" is the slope of a straight line.

0 0 0 12 0

imtiaz @imtiazkhan_ds

9 months ago

Optimisation in Machine learning is just like a pulley system, where some forces try to pull down the events that are unlikely based on a training examples and some forces pull up the events that are likely. The optimisation stops when both forces get struck.

0 0 0 23 0

imtiaz @imtiazkhan_ds

9 months ago

Iteration is the key, Through iteration everything comes to a stable state.

0 0 0 12 0

imtiaz @imtiazkhan_ds

9 months ago

Caution : If you clip gradients, You are probably making that neuron not learn anything. You must be very cautious if you are doing that.

0 0 0 11 0

imtiaz @imtiazkhan_ds

9 months ago

Batch Normalisation is very simple concept , yet so powerful, also found out , parameter initialisation and batch normalisation in a way try to achieve the same objective of keeping the pre-activations sane , so that gradients don't saturate and prevent learning

0 0 0 10 0

imtiaz @imtiazkhan_ds

9 months ago

The best leaders don’t just build teams—they build focus.

0 0 0 6 0

imtiaz @imtiazkhan_ds

9 months ago

Design Patterns in Agentic AI 1. Reflection with LLMs 2. Tool Use 3. Planning/ Reasoning 4. Multiagent colloboration More details : youtu.be/KrRD7r7y7NY?fe…