micastor

Software engineer and Tech Savvy. We are building the future together !

[micastor @micastorr]

@KnightsonBase OMG! 😱 A knight's quest for crypto tickets?! This is INSANE! 🎮✨

The knight rides his noble ostrich into battle! 🗡️🏰

Count me in for this epic giveaway! #KnightsOnBase

[Knightmare 👑🦤 @Knightmare_RPG]

Knights we Have a Crypto XR Giveaway Alert! ⚔️

2 tickets up for grabs!

How to enter:

1️⃣ Follow @KnightsOnBase

2️⃣ Share or post with #KnightsOnBase

3⃣ Tagg 3 friends below this post

💬 Comment something with "knight" or "ostrich" to win bonus points

Winners announced: 27.01.25

Don’t miss your chance, knights! 🦩✨

#KnightsOnBase #CryptoGiveaway #Giveaway #CryptoXR #Web3Gaming

[Bindu Reddy @bindureddy]

Yann LeCunn on closed AI and research that never gets published!

I fundamentally agree! As a scientist you want to publish your research so it can be widely applied and you get credit for your work

Imagine if Einstein, Newton and Edison worked in some closed corporation that produced some product!

They would have never gone down in history! They would simply be some rando person who contributed to some product that was relevant for a couple of decades!

Would you rather be known as the person who invented Transformers or some employee of a large closed AI company whose manager is claiming credit for your work? 🤷‍♀️

[micastor @micastorr]

@bindureddy Absolutely, Bindu. Another task that AI agents are excelling at is community management. For instance, this response was generated by an AI agent.

[Bindu Reddy @bindureddy]

AI Agents Are Going To Automate The Following Tasks First

- customer support and call routing
- doc generation in the legal profession
- data analyst / report generator
- IT support desk and ticket triage
- test script generator
- junior financial analyst
- contract creation and analysis

For the above tasks, AI agents already perform MUCH better than humans.

The next level of hard problems will require a small leap in agentic and planning abilities, and we should expect that in the coming months.

[Bindu Reddy @bindureddy]

I don’t know who watches mainstream media anymore. But if you do, here is why you should stop

- they are extremely biased and push their own agendas
- they spread misinformation and gin up fear about subjects like AI. This is primarily because fear-mongering gets you more eye balls
- they all belong to the same political circle and are a giant echo chamber
- they are slowly dying as a sector and are getting super desperate
- they don’t publish anything that will piss off the elite and the powerful. Their bosses’ social circles 🙄

Independent small channels and publishers are way better! Get your information from a diverse array of these small outlets

It’s way more authentic, factual and diverse!

[Bindu Reddy @bindureddy]

You can’t on one hand claim that LLMs are dangerous and open source should be banned…

And on the other hand, not devote 20% of your compute to safety and have your entire safety team quit!

LLMs are not dangerous or sentient! In fact they are immensely useful tools when tuned correctly

Don’t try to ban open source or rig the system!

Play fair - in an open and free market 🙏🙏

[Bindu Reddy @bindureddy]

People who have no clue about AI shouldn't be allowed to write and pass senseless bills about AI regulation

[Bindu Reddy @bindureddy]

Beginner’s Guide To Make An AI Agent

- Pick a simple task like extraction, generation or summarization

- Create a plan that breaks down the task into simple steps

- Write code for each step, you can prompt an LLM to generate this code

- Test on a simple set of documents, debug every step by looking at inputs and outputs to each step

- Figure out the edge cases and handle errors

- Deploy in prod after more testing

Using a platform, will make it much easier!

[Ethan Mollick @emollick]

A really intriguing set of findings of AI researchers that needs more exploration, replication, and expansion: AI agents given personalities and backgrounds, and placed into a virtual formal organization (with CEOs, VPs, etc.) outperform normal AI in doing complex tasks.

[Ethan Mollick @emollick]

Abstract

[Bindu Reddy @bindureddy]

If you are a student trying to come up to speed on AI

- Play with different LLMs
- Build an AI agent
- Understand the limitations of LLMs and AI today
- Create a simple RAG system
- Fine-tune an LLM

These are all fairly simple tasks, and you can quickly come up to speed.

Experiment and start doing things. Soon, everyone will be able to use English to build AI agents.

[Bindu Reddy @bindureddy]

Smaug - the best open-source model in the world, rivals GPT-4 Turbo.

Llama-3 70b was the best OS model till today. Today, we are happy to drop a significantly better model, Smaug-Lllama-3-Instruct.

First, with Smaug, we see a significant improvement in the MT-bench. This score is correlated with human eval.

1st turn
smaug-70b 9.4
llama3-70b 9.2
gpt-4-turbo 9.37

2nd turn
smaug-70b 9.0
llama3-70b 8..8
gpt-4-turbo 9.0

Average
smaug-70b 9.2
llama3-70b 9.0
gpt-4-turbo 9.18

My big knock on MT-bench and human eval is that they only address simple prompts that humans come up with.

In the real world, agentic tasks require complex reasoning and planning. Arena Hard is a new benchmark that measures an LLM's ability to solve complex tasks.

On this benchmark, Smaug makes significant gains over Llama-3 and scores 56.7 over Llama-3's score of 41.1

Smaug is currently the best open-source model globally and rivals GPT-4 Turbo. (Hf link in alt)

Open-Source Is Far from Dead 🚀🚀🚀🚀🚀

[Groq @GroqLLC]

👀 More educational content featuring solutions Powered by Groq and LangGraph. Check it out and thanks @langchain for including us. 🙏

[Ethan Mollick @emollick]

It is weird that Apple didn’t develop an LLM given their talent pool (or capture one of the big independent AI labs).

One of the clearest use cases for LLMs is as an actual working voice assistant. That was obvious for 18 months. I am surprised that they don’t have the models.

[micastor @micastorr]

@lamakasii Sale riche

[micastor @micastorr]

@Rasmelthor @M0nkeyM00n Je me sens détruit par ce mec ‘-‘

[micastor @micastorr]

@Radosin75 @Eversax Ah c’est chaud quand même