Twitter Beğeni Hilesi

In case anyone is wondering if o1 is still based on GPT-4o - yes, it is, even though they like to call it a new model. If you work at OAI, it’s not called o1, but actually "GPT-4o with reasoning"

7/ PS. You want to see a cute kid failing Theory of Mind like GPT-4o?

Google's new DataGemma uses Retrieval *Interleaved* Generation to mitigate against hallucination:

What would you do if you were David?

Prompting OpenAI's o1 ("Strawberry") feels like learning a brand new game. Rate limits just increased: now it's 50 messages per day for o1-mini (instead of 50 per week). What have you learned about using o1 and o1-mini so far?

Want a simple explanation of a research paper? Google's NotebookLM generates two podcasters explaining to a lay audience. It's free. I was skeptical—but listen to how well executed it is. Just try 60 secs. It's not perfect: it's like listening to podcasters discuss a paper. 👇

Do you have a technical question that AI can't answer, but you can? The top 50 questions will earn $5,000 each. The next 500 will earn $500 each. Co-authorship on the paper also offered if your question gets in.

Ethan BdM I like your stance. But just in case someone else uses the term, I like this mnemonic:

People complain we don't build modern wonders. How about:

Do you have examples of AI weirdness? Can be text based, images, whatever. Share examples you've seen below.

Anyone building an AI agent that cancels your subscriptions —to see if you get a massive discount offer? I just cancelled a $1k a year sub, and they were immediately like: would you stay for 50% off?

Such a good use of ChatGPT for writing: - having a clear and detailed sense of what good looks like - systematising repetitive edits/comments you make again and again - recognising that it's maybe ~80% accurate And human in the loop for judgement, writing, and discernment.