Amrit Krishnan

I like to build things. Optimist. Improving healthcare using machine learning @VectorInst. #AI #Healthcare

[Amrit Krishnan @amritk110]

My team at Vector is excited to share the latest release of CyclOps! @VectorInst #healthcare #ai

Link to blog - https://t.co/KJ0nD5pR61

Link to github - https://t.co/cKpFjhXorH

Some key feature updates! (1/n)

[Amrit Krishnan @amritk110]

We are spending billions to marginally improve LLMs, maybe even making them worse in terms of over fitting to benchmarks. Meanwhile we are spending peanuts on coral reef restoration, something that could actually save the planet. #LLMs #coralreefs #OpenAl

[Amrit Krishnan @amritk110]

I spent the weekend implementing a simple and hackable vector database in rust. The purpose is mostly educational, but it can be used in lightweight ML pipelines. https://t.co/wM0c9hUCQK. #rustlang

[Amrit Krishnan @amritk110]

@AravSrinivas, generating markdown on perplexity doesn't format nicely if it also involves code snippets. The markdown content appears inside the "text" card with clipboard copy but the code snippet appears outside.

[Orla Joelsen @OJoelsen]

Hi Canadian neighbors,🇨🇦
We also have Canadian products here in Nuuk - Greenland.🇬🇱
From today, we will begin to buy some your
products.

#standwithcanada 🇨🇦 stay strong!

[Made In Canada @MadelnCanada]

Buy Canadian
Support Canadian

Let’s also follow Canadian 🇨🇦

Like ❤️ & RT 🔄 this tweet

Reply with your favourite Canadian X account, let’s help grow Canadians on social media.

[Amrit Krishnan @amritk110]

@AravSrinivas Yes. I'd like to see more evaluations, but the reasoning traces are just a lot better from a user perspective.

[Narendra Modi @narendramodi]

Game-changer for Indian academia and for youth empowerment!

The Cabinet has approved ‘One Nation One Subscription’, which will strengthen our efforts to become a hub for research, learning and knowledge. It will also encourage interdisciplinary studies.

https://t.co/qbBoT71vON

[Eric Topol @EricTopol]

Does an AI/ML model provide more accurate warning of clinical deterioration than older models (logistic regression or point-based) among >360,000 hospitalized patients in 7 hospitals?
Yes. eCART (@AgileMD) performed best; EPIC (EDI) was one of the worst.
https://t.co/mfGZ0Ub5Tv by @dpedelson and colleagues
Editorial https://t.co/rBrWde86UO by @AmolAVerma

[Amrit Krishnan @amritk110]

Ever notice how when a startup “benchmark” themselves to other LLMs, they always seem to do better than GPT-4 while GPT-4 does better than other models. I see this even if the domain is very specialized or if metrics are very specific. Almost like if they were just faking it…😞

[Delip Rao e/σ @deliprao]

Folks are surprised that they are giving away gpt-4o for free. They are collecting your speech prompt data silly. Remember if you are not on the team plan, unless you go through a complicated opt-out process, all your data is by default used by openai. Not surprising anymore.

[Bindu Reddy @bindureddy]

Early Results From Our First Eval Of GPT-4o - Hard Coding and Reasoning

GPT-4o
Successful tasks - 79 / 96
Coding tasks - 52 / 65

GPT-4
Successful tasks - 90/ 96
Coding tasks - 60 / 65

The model is way faster, but it's unclear why it's doing much worse on hard tasks 😢

Trying to debug to see if we can find a common theme/issue

[Amrit Krishnan @amritk110]

@OpenAI The hype behind AI is over. This cringe demo is proof of that.

[Qiao Jin, MD @DrQiaoJin]

PMC-Patients, an open dataset of 167k patient summaries and their relations, finally published at https://t.co/6xRvuZ3PXH.

Many exciting things to try!

[Amrit Krishnan @amritk110]

@fabian_theis @LukasHeumos This toolkit looks awesome! We have built something similar yet complementary to focus more on evaluation and monitoring! https://t.co/cKpFjhXorH.
Would love to create some joint tutorials to show end-to-end use cases on EHR data that use both APIs highlighting their strengths.

[Fabian Theis @fabian_theis]

Excited to share ehrapy, our data & software framework for exploratory electronic health record (EHR) analysis. Led by @LukasHeumos and inspired by scanpy, we enable vendor-independent exploratory end-to-end analysis of EHR data. https://t.co/nKKEpe2sly
https://t.co/CJRIsM8jzi

[Bo Wang @BoWang87]

congrats on the amazing work @fabian_theis! great to see such tools that will make a huge impact on #AI & #healthcare research!

[Amrit Krishnan @amritk110]

Mistral overshadowing billion dollar Gemini with a torrent link 😂. The power of open source ❤️.

[clem 🤗 @ClementDelangue]

Super happy to support the AI alliance. We need more openness, transparency & collaboration in AI! https://t.co/e2DnnGB4uo

[clem 🤗 @ClementDelangue]

Open models now starting to surpass GPT4 for specialized tasks. Let's go!

Model by @NexusflowX: https://t.co/TBUBrevTpJ

Leaderboard: https://t.co/jbvk3U8Ibt

[Amrit Krishnan @amritk110]

@t3dotgg Use chatGPT and make it seem obvious