Offshore
Photo
Dave Craige
Be very aware. This guy doesn’t give a shit about our community.
/ #HustleGPT https://t.co/WJJupHcTYb
tweet
Be very aware. This guy doesn’t give a shit about our community.
/ #HustleGPT https://t.co/WJJupHcTYb
tweet
Offshore
Photo
Ethan Mollick
One way AI is going to change research is by making hard and inaccurate data analysis much easier.
Here GPT-3.5 (not even GPT-4!) outperforms human annotators and costs 20x less per annotation. (Data annotation is one of the most annoying & expensive parts of research projects) https://t.co/oliUujap2J
tweet
One way AI is going to change research is by making hard and inaccurate data analysis much easier.
Here GPT-3.5 (not even GPT-4!) outperforms human annotators and costs 20x less per annotation. (Data annotation is one of the most annoying & expensive parts of research projects) https://t.co/oliUujap2J
LLMs Can Outperform Humans on Data Annotation
-Compare 0-shot accuracy of ChatGPT vs crowd-workers on:
relevance
topic detection
stance detection
general frame detection
policy frame detection
-ChatGPT better on 4/5 tasks
-20x cheaper than MTurk
Paper: https://t.co/lcVMqbFBFu https://t.co/VS3wRCkuUG - John Naytweet
Offshore
Photo
Daniel Vassallo
🚨 You can now get free access to GPT-4 if you join our https://t.co/NdIyIcbyCB community. Just talk to our smallbot and it will enlighten you with its vast wisdom.
Our smallbot has a peculiar personality though 😂 https://t.co/0hfF0SJ6Pr
tweet
🚨 You can now get free access to GPT-4 if you join our https://t.co/NdIyIcbyCB community. Just talk to our smallbot and it will enlighten you with its vast wisdom.
Our smallbot has a peculiar personality though 😂 https://t.co/0hfF0SJ6Pr
tweet
Adam.GPT
RT @OfficialLoganK: @gopikl @DavidSacks Klarna, Milo, & Speak are all startups and initial featured plugins.
As @TheRealAdamG said, we are always looking for great use cases from high trust partners.
No AI company is more focused on enabling startups and developers right now than us.
tweet
RT @OfficialLoganK: @gopikl @DavidSacks Klarna, Milo, & Speak are all startups and initial featured plugins.
As @TheRealAdamG said, we are always looking for great use cases from high trust partners.
No AI company is more focused on enabling startups and developers right now than us.
tweet
Offshore
Photo
Stephanie 🏹
hokay i just listened to fifty fifty's entire discography and I AM IN LOVE???? i know everyone's going crazy over cupid and i am too but can we PLEASE talk about higher?? that song is ANGELIC. that vocal line is something else, man. gives me chills & makes me wanna eat a cloud. https://t.co/0Ow9HB4wNZ
tweet
hokay i just listened to fifty fifty's entire discography and I AM IN LOVE???? i know everyone's going crazy over cupid and i am too but can we PLEASE talk about higher?? that song is ANGELIC. that vocal line is something else, man. gives me chills & makes me wanna eat a cloud. https://t.co/0Ow9HB4wNZ
tweet
Offshore
Photo
Victor Sanh
RT @margs_li: 📰 📸: the comeback of 🌟BTM🌟, your fave Kpop boyb-- er, 🌟LM algorithm🌟
cluster-BTM (c-BTM) discovers domains in data ➡️ trains independent domain experts from a pretrained LM ➡️ sparsely activates experts for inference
⏭️ costs less💲, outperforms baselines!
⚪️ / n; 👇 https://t.co/4ciumHeQDG
tweet
RT @margs_li: 📰 📸: the comeback of 🌟BTM🌟, your fave Kpop boyb-- er, 🌟LM algorithm🌟
cluster-BTM (c-BTM) discovers domains in data ➡️ trains independent domain experts from a pretrained LM ➡️ sparsely activates experts for inference
⏭️ costs less💲, outperforms baselines!
⚪️ / n; 👇 https://t.co/4ciumHeQDG
tweet
Offshore
Photo
Arvind Narayanan
This is one example pointing to an emerging paradigm where LLMs are used for labeling data, which is then then used to train a small, specialized model. The use of small models in production avoids the security, reliability, and cost drawbacks of LLMs. https://t.co/qdOVk3XgED
tweet
This is one example pointing to an emerging paradigm where LLMs are used for labeling data, which is then then used to train a small, specialized model. The use of small models in production avoids the security, reliability, and cost drawbacks of LLMs. https://t.co/qdOVk3XgED
LLMs Can Outperform Humans on Data Annotation
-Compare 0-shot accuracy of ChatGPT vs crowd-workers on:
relevance
topic detection
stance detection
general frame detection
policy frame detection
-ChatGPT better on 4/5 tasks
-20x cheaper than MTurk
Paper: https://t.co/lcVMqbFBFu https://t.co/VS3wRCkuUG - John Naytweet
Carlos E. Perez
Cognitive psychologists and traditional ML researchers are confused about LLMs. These fields have existed for decades, and generations of PhDs have been trained using their methodologies. But here we are, none of the old methods predicted what we see https://t.co/UaHR3ubrBJ… https://t.co/CH9rYL0HFC
tweet
Cognitive psychologists and traditional ML researchers are confused about LLMs. These fields have existed for decades, and generations of PhDs have been trained using their methodologies. But here we are, none of the old methods predicted what we see https://t.co/UaHR3ubrBJ… https://t.co/CH9rYL0HFC
tweet
inFERENCe
We May be Surprised Again: Why I take LLMs seriously.
"Deep Learning is Easy, Learn something Harder" - I proclaimed in one of my early and provocative blog posts from 2016. While some observations were fair, that post is now evidence that I clearly underestimated the impact simple techniques will have, and…
near
it would be reasonable to let users toggle this off, as that seems in line with the general ethos elon has in mind for twitter https://t.co/u9CqGgxV55
tweet
it would be reasonable to let users toggle this off, as that seems in line with the general ethos elon has in mind for twitter https://t.co/u9CqGgxV55
Starting April 15th, only verified accounts will be eligible to be in For You recommendations.
The is the only realistic way to address advanced AI bot swarms taking over. It is otherwise a hopeless losing battle.
Voting in polls will require verification for same reason. - Elon Musktweet
Robert Scoble
All AIs start worse than a child and keep improving. Eventually will be better than any human. https://t.co/IVEclgreyK
tweet
All AIs start worse than a child and keep improving. Eventually will be better than any human. https://t.co/IVEclgreyK
Around my parts, the highway stack for FSD v11.3.3 is the most impressive technical feat I’ve ever seen and I’m not joking when I say this. The car legitimately drove better than I did on the highway. It was just as smooth and decisive, yet way safer.
The world is not ready. - Farzad Mesbahitweet
You.com
RT @DrFantinuoli: https://t.co/2NrUe7kZzs is a new search engine with a ChatGPT like search feature (called Chat) which gives short explanations for your queries, with precise sources of the information it presents.
This is an examples of the plethora of #AI tools we will be using every day.
tweet
RT @DrFantinuoli: https://t.co/2NrUe7kZzs is a new search engine with a ChatGPT like search feature (called Chat) which gives short explanations for your queries, with precise sources of the information it presents.
This is an examples of the plethora of #AI tools we will be using every day.
tweet
You
The AI Search Engine You Control | AI Chat & Apps
You.com is a search engine built on artificial intelligence that provides users with a customized search experience while keeping their data 100% private. Try it today.
Stephen Young
RT @rachelmetz: So far dinner is going great: the instant pot refuses to seal and I accidentally made scones instead of biscuits like what the actual fuck am I doing
tweet
RT @rachelmetz: So far dinner is going great: the instant pot refuses to seal and I accidentally made scones instead of biscuits like what the actual fuck am I doing
tweet
Adam.GPT
RT @vboykis: in two years, every single platform will offer a SQL interface, vector database, K8s integration, DAG-based job scheduler, and API builder, and we will interact with them using only a series of prompts that takes 17 hours to get right and we immediately commit to YAML manifests
tweet
RT @vboykis: in two years, every single platform will offer a SQL interface, vector database, K8s integration, DAG-based job scheduler, and API builder, and we will interact with them using only a series of prompts that takes 17 hours to get right and we immediately commit to YAML manifests
tweet