Rafael Madrigal

Building apps with LLMs to make life easy | Senior Data Scientist @gcashofficial | MS Data Science @aim_edu | @upsystem BS ChemE

[reia 🍁💫 @reiaposting]

waterloo to san francisco pipeline so strong they're forgetting that it's in canada

[Simon Willison @simonw]

Another prompt injection paper review! This time it's "An Introduction to Google’s Approach to AI Agent Security" by Santiago Díaz, Christoph Kern, and Kara Olive

[giggly ♡ @xgigglypuff]

This guy is smart, will be successful, and I wish him the best of luck, but I wish more people who go viral for getting rejected from top colleges would post their personal statement because it becomes immediately obvious why they were rejected. This is an educational resource.

[Zain @ZainHasan6]

very cool and detailed breakdown of 7 years worth of advancements in post-training LLMs

SFT, RLHF, DPO, visual SFT, MoE, reasoning

[Chase Harris @chaseharris98]

How To Develop Confidence

[Rafael Madrigal @raf_madrigal]

I don't know what to feel about this as an ex-Union banker.

While DS roles are slowly evolving to ML/ AI Engineers or ML/AI Researchers, the role they play in bridging technical and business verticals must not be underestimated.

[Dominic Ligot @docligot]

Opening salvo from the morning panel: Pres of unionbank saying they have too many data scientists. And AI will take them out of a job.

Congrats Ed. You said it before I did.

[Rafael Madrigal @raf_madrigal]

@docligot Finally a long time coming. Hopefully, this would mean that we see more government support/ initiatives related to open data/ colelction, data privacy, and ai compute

[Jeremy Howard @jeremyphoward]

For those that hope (or worry) that LLMs will do breakthrough scientific research, I've got good (or bad) news:

LLMs are particularly, exceedingly, marvellously ill-suited to this task. (if you're a researcher, you'll have noticed this already)

Here's why🧵

[Nick Dobos @NickADobos]

Good diagram for ai engineer

[proper @ProperPrompter]

Anthropic has taken the lead.

The new Clause 3.5 Sonnet model is in a league of it's own.

11 impressive demos of the new model:

[Matthew Carrigan @carrigmat]

Good morning. At some point this summer, perhaps quite soon, @AIatMeta will be releasing a LLaMA-3 model with 400B parameters. It will likely be the strongest open-source LLM ever released by a wide margin.

This is a thread about how to run it locally. 🧵

[Rafael Madrigal @raf_madrigal]

@xvarunxx @cohere Congrats Varun!

[Rafael Madrigal @raf_madrigal]

Is it possible to publish as a non-MS/ PhD student associated with a research lab? Or no choice talaga but to study again?

Would research labs be open in allowing private individuals from industry help in their research in hopes of publishing as a co author?

[ollama @ollama]

.@CohereForAI's Aya 23 is here! Available in 8B and 35B.

ollama run aya:8b

ollama run aya:35b

Learn more:
https://t.co/NC5WeolI86

[Kartikay @Kartikayb77]

Short attention span and too lazy to read this post? Skill Issue

[Patrick Lewis @PSH_Lewis]

New paper from our team, led by @pat_verga
Are you:
* Doing evaluation with LLMs?
* Using a huge model?
* Worried about self-recognition?

Try an ensemble of smaller LLMs.
Use a PoLL: less biased, faster, 7x cheaper. Works great on QA & Arena-hard evals
https://t.co/Lhvx5GN8I8

[Rafael Madrigal @raf_madrigal]

That’s me taking a photo lol. It was fun to meet other builders in the GTA

[Affan Imran @affan_imran]

@cohere Thanks for having us!

[Rafael Madrigal @raf_madrigal]

Prior to coming to the event, I had an inherent bias towards @OpenAI models for RAG applications, since that what I use at work, but after yesterday, I think I have a new bias for RAG apps.

Thanks @cohere for the amazing event! Would love to attend future events like this!

[Rafael Madrigal @raf_madrigal]

A little life update: I was one of the winners of the same-day hackathon hosted by @cohere yesterday! I presented my initial work on equipping multi-agents with multi-tools to speed up the Privacy and Ethics evaluation of AI projects. (Will share initial experiments later)

[Cohere @cohere]

That’s a wrap on Cohere Build Day Toronto! Hosted at our HQ, attendees brainstormed, networked, and built working 27 knowledge agents in three hours. Congratulations to all of the winners! Up next, London and San Francisco!

[Rafael Madrigal @raf_madrigal]

I also had a great time meeting other builders in the GTA and learning how they use LLMs at work!

Command R+ is @cohere 's foundational model tuned specifically for RAG applications. They have made it easy for RAG-builders to incorporate citations in the bot responses