The new Grok seems like it could be useful as a search replacement for combing through tweets to find useful news. Alternatives for me might be the feed or lists.
At the beginning of March, will I judge that I've been using Grok a lot to keep up with developments in Computer Science, AI, Mathematics, etc.? (with an expectation that I follow the AI news the most)
My baseline for the resolving probability will be: "Percentage of weeks that I used Grok at least once to summarize or understand technical news". If Grok releases a conversation history, it will be a well-defined number. Otherwise, I'm estimating based on my memory.
If I cancel my Twitter Blue/Premium+, this market resolves NO because Grok wasn't worth the cost. I won't cancel it just to manipulate this market.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ70 | |
| 2 | Ṁ31 | |
| 3 | Ṁ22 | |
| 4 | Ṁ10 | |
| 5 | Ṁ9 |
I have not used Grok really at all. I asked it about "top papers from NeurIPS" 2-3 different ways and got generic responses instead of specific tweet threads about specific papers. If it had at least pointed me somewhere useful, I would've counted it; but as it is this currently is on track to resolve 0%.
Some casual use today, just to see what kinds of responses it returns. Today is "reinforcement learning", next time I will probably ask about "any recent new quantum algorithms?"(I know of Shor's Algorithm, Grover's Algorithm, and the random quantum circuit that Google claimed gave them "quantum supremacy"; but these are years old by now)
If I didn't know anything about reinforcement learning, the 2013 "Playing Atari with Deep Reinforcement Learning" is the first paper mentioned and would be a good place to start(checking for heavily-cited papers that cite it).
The AlphaStar Unplugged I would've wanted to know about had I not already seen it. So points for that.
I think I remember seeing the Schmidhuber "World Models" paper a few years ago and it sounded very interesting at the time.
I would've expected the search "introduction to reinforcement learning" to link to https://spinningup.openai.com/en/latest/
Google has it as the 3rd result, and has Sutton and Barto on Amazon as the 5th result.
So it seems okay at finding relevant papers that are talked about on Twitter, but Google so far is competitive.
@Mira so far AI history, I can't wait to see your updates on AI "news" performance, because that seems to be my {arbitrary platform} wrapped 2023 subject. You know there's gonna be really personal favourites missing from it's summaries but will it actually keep you updated
@DanPowell I don't currently use any LLMs to summarize news on a regular basis. That includes locating articles/tweets/etc. and given an article summarizing it into fewer words.
Partly because, while LLMs are slightly faster at reading than me, they're much slower at using search tools. I can search 10 articles and skim them for important information probably 10x faster than GPT-4 or Bing can. Grok is the first one that seemed fast enough at searching when I first used it to be useful.
GPT-4 might have an advantage summarizing arxiv papers, because it takes me longer than 30s to get information I want from most papers. And the feed is regular enough as to be easy to search.
@Mira I asked gpt4 to summarise an arxiv link and it had to be fed the link->pdf I gave it, was I meant to give it the title to search for